System and method for predicting brain tau or amyloid levels

Abstract

A machine learning model can be trained to predict a patient's cerebral tau or amyloid-beta state using, at least partially, patient biomarker data. The cerebral tau or amyloid-beta state may involve one or more continuous values ​​of cerebral tau or amyloid-beta levels and / or may relate to cerebral tau or amyloid-beta levels in multiple regions of the patient's brain. Predictions of a patient's cerebral tau or amyloid-beta state can be generated by applying patient-specific data to a machine learning model. The operation may further include providing predictions of a patient's cerebral tau or amyloid-beta state. The machine learning model may be used in patient selection for treatment with anti-tau therapy and / or anti-amyloid therapy, in the treatment of patients with or suspected of having Alzheimer's disease (AD), or in monitoring the effectiveness of AD treatment.

Description

[Technical Field]

[0001] Cross-reference of related applications This application claims the interests of U.S. Provisional Patent Application No. 63 / 513,839, filed on 14 July 2023. This application also claims the interests of U.S. Provisional Patent Application No. 63 / 618,754, filed on 8 January 2024. This application also claims the interests of U.S. Provisional Patent Application No. 63 / 625,245, filed on 25 January 2024. This application also claims the interests of U.S. Provisional Patent Application No. 63 / 625,247, filed on 25 January 2024. This application also claims the interests of U.S. Provisional Patent Application No. 63 / 644,301, filed on 8 May 2024. This application also claims the interests of U.S. Provisional Patent Application No. 63 / 644,322, filed on 8 May 2024. The patent applications identified above are incorporated by reference as a whole to provide continuity of this disclosure.

[0002] This disclosure relates to the training and use of machine learning or statistical models for predicting continuous or discrete brain tau or amyloid-beta states in individual patients. [Background technology]

[0003] Conventional methods for determining brain tau or amyloid positivity in patients may require the administration of radioactive tracers to the patient and subsequent collection of imaging data (e.g., positron emission tomography (PET) scans). The additional time and cost can be burdensome for patients, while the necessary clinical resources (e.g., PET scanners) can be burdensome for healthcare providers. These burdens may hinder widespread screening of patients for tau or amyloid positivity, impacting patient treatment and the design and evaluation of clinical trials. [Overview of the project] [Means for solving the problem]

[0004] A system and method for predicting cerebral tau or amyloid-beta status, as well as a computer-readable medium, are disclosed. In accordance with the disclosed embodiments, a machine learning model can be trained to predict a patient's cerebral tau or amyloid-beta status. The machine learning model can predict a patient's cerebral tau or amyloid-beta status based on the patient's biomarker data.

[0005] The disclosed embodiments include a system. The system may include at least one processor and at least one non-temporary computer-readable medium containing instructions. When executed by at least one processor, instructions can cause the system to perform an action to predict a brain tau or amyloid-beta state. The action may include obtaining a machine learning model trained to predict a patient's brain tau or amyloid-beta state from the patient's subjective data. The subjective data may include the patient's biomarker data. The brain tau or amyloid-beta state may include one or more continuous brain tau or amyloid-beta levels and / or relate to brain tau or amyloid-beta levels in multiple regions of the patient's brain. The action may further include generating a prediction of the patient's brain tau or amyloid-beta state by applying the patient's subjective data to the machine learning model. The action may further include providing a prediction of the patient's brain tau or amyloid-beta state.

[0006] Embodiments disclosed include another system, which may include at least one processor and at least one non-temporary computer-readable medium containing instructions. When executed by at least one processor, instructions can cause the system to perform an action to train a machine learning model to predict brain tau or amyloid-beta states. The action may include obtaining brain tau or amyloid-beta states of a training patient. The action may further include obtaining reference data of the training patient, which is reference data including biomarker data. The action may further include generating training samples that associate brain tau or amyloid-beta states with reference data for the training patient. The action may further include training a machine learning model using the training samples to predict brain tau or amyloid-beta states from the reference data. The action may further include providing the trained machine learning model to enable prediction of brain tau or amyloid-beta states.

[0007] The disclosed embodiments include a method for selecting a patient for treatment with anti-tau therapy and / or anti-amyloid therapy. The method may include, in accordance with the disclosed embodiments, identifying a patient who has AD or is at risk of AD, and administering anti-tau therapy or anti-amyloid therapy.

[0008] The disclosed embodiments include methods for treating patients who have or are suspected of having Alzheimer's disease. The methods may include, in accordance with the disclosed embodiments, identifying the patient as having or being at risk of having AD, and administering anti-tau therapy and / or anti-amyloid therapy.

[0009] The disclosed embodiments include a method for monitoring the therapeutic efficacy of AD. The method may include, in accordance with the disclosed embodiments, measuring elevated levels and / or numbers of brain regions containing neurofibrillary fibrous masses by obtaining and applying levels of p-tau 217 to a system. The method may further include administering a therapeutic agent and repeating the measurement. A reduction or delay in the progression of brain regions containing neurofibrillary fibrous masses may indicate therapeutic efficacy.

[0010] The general descriptions above and the detailed descriptions below are illustrative and explanatory only and do not limit the scope of the claims.

[0011] The accompanying drawings, incorporated into and constituting part of this disclosure, along with the description, are useful in illustrating and illustrating the principles of various exemplary embodiments. [Brief explanation of the drawing]

[0012] [Figure 1] This document presents an exemplary platform for developing, validating, and deploying predictive models for predicting cerebral tau or amyloid-beta states, consistent with the disclosed embodiments. [Figure 2] An exemplary process for predicting the tau or amyloid-beta state of a subject brain, consistent with the disclosed embodiments, is shown. [Figure 3] This document describes an exemplary process for treating a subject, consistent with the disclosed embodiments. [Figure 4] An illustrative diagram of the progression of Alzheimer's disease along the Braak stage, consistent with the disclosed embodiments, is provided. [Figure 5A] Exemplary structural brain network (SBN) hubs and modules consistent with the disclosed embodiments are illustrated. [Figure 5B] Exemplary structural brain network (SBN) hubs and modules consistent with the disclosed embodiments are illustrated. [Figure 5C] Exemplary structural brain network (SBN) hubs and modules consistent with the disclosed embodiments are illustrated. [Figure 6] 1 illustrates an example table summarizing patient data consistent with the disclosed embodiments. [Figure 7] illustrates an example distribution of standard uptake value ratio (SUVR) values in subjects having different tau-positive statuses consistent with the disclosed embodiments. [Figure 8] illustrates an example table showing performance for tau-positive prediction consistent with the disclosed embodiments. [Figure 9A] illustrates example top predictive factors for detecting tau-positive subjects in Braak stages 3 to 6 using a stochastic gradient boosting machine model consistent with the disclosed embodiments. [Figure 9B] illustrates example top predictive factors for detecting tau-positive subjects in Braak stages 3 to 6 using a stochastic gradient boosting machine model consistent with the disclosed embodiments. [Figure 9C] illustrates example top predictive factors for detecting tau-positive subjects in Braak stages 3 to 6 using a stochastic gradient boosting machine model consistent with the disclosed embodiments. [Figure 10A] illustrates example individual conditional expectation profiles of several features for predicting tau positivity consistent with the disclosed embodiments. [Figure 10B] illustrates example individual conditional expectation profiles of several features for predicting tau positivity consistent with the disclosed embodiments. [Figure 10C] illustrates example individual conditional expectation profiles of several features for predicting tau positivity consistent with the disclosed embodiments. [Figure 11A] illustrates an example heatmap showing interactions between several features for predicting tau positivity consistent with the disclosed embodiments. [Figure 11B] illustrates an example heatmap showing interactions between several features for predicting tau positivity consistent with the disclosed embodiments. [Figure 11C] illustrates an example heatmap showing interactions between several features for predicting tau positivity consistent with the disclosed embodiments. [Figure 12A] Examples of plasma phosphorylated tau 181 (p-tau 181) corresponding to tau-positive status, consistent with the disclosed embodiments, are provided. [Figure 12B] Examples of plasma phosphorylated tau 181 (p-tau 181) corresponding to tau-positive status, consistent with the disclosed embodiments, are provided. [Figure 13A] Using a Bayesian ordered logistic model consistent with the disclosed embodiments, we illustrate predictors for identifying tau-positive subjects with Braak disease stages 3–4 and Braak disease stages 5–6. [Figure 13B] Using a Bayesian ordered logistic model consistent with the disclosed embodiments, we illustrate predictors for identifying tau-positive subjects with Braak disease stages 3–4 and Braak disease stages 5–6. [Figure 13C] Using a Bayesian ordered logistic model consistent with the disclosed embodiments, we illustrate predictors for identifying tau-positive subjects with Braak disease stages 3–4 and Braak disease stages 5–6. [Figure 14AB] Figures 14A and 14B illustrate top MRI predictors consistent with the disclosed embodiments. [Figure 14C] Examples of top MRI predictors consistent with the disclosed embodiments are provided. [Figure 14D] Examples of top MRI predictors consistent with the disclosed embodiments are provided. [Figure 14E] Examples of top MRI predictors consistent with the disclosed embodiments are provided. [Figure 15A] The cortical thicknesses in two regions corresponding to tau-positive status in Braak stages 0-2, Braak 3-4, and Braak 5-6 are illustrated, grouped by inferior parietal cortical thickness-right (VCIPCR) thickness, consistent with the disclosed embodiments. [Figure 15B] The cortical thicknesses in two regions corresponding to tau-positive status in Braak stages 0-2, Braak 3-4, and Braak 5-6 are illustrated, grouped by inferior parietal cortical thickness-right (VCIPCR) thickness, consistent with the disclosed embodiments. [Figure 16]A table illustrating how adding amyloid PET centiloid data, consistent with the disclosed embodiments, can improve tau-positive prediction is provided. [Figure 17A] This illustrates how amyloid PET levels can play a complementary role in predicting tau positivity, consistent with the disclosed embodiments. [Figure 17B] This illustrates how amyloid PET levels can play a complementary role in predicting tau positivity, consistent with the disclosed embodiments. [Figure 18A] This provides an overview of key demographic, clinical, and genomic characteristics of subjects included in the data used to construct and validate a particular brain tau state prediction model consistent with the disclosed embodiments. [Figure 18B] This exhibits a nonlinear pattern in the relationship between the plasma phosphorylated tau-217 (p-tau-217) / unphosphorylated tau-217 (np-tau-217) ratio (p-tau-217R) and tau-PET SUVR values ​​across selected cortical regions and in six Braak stage regions, consistent with the disclosed embodiments. [Figure 18C] This exhibits a nonlinear pattern in the relationship between the plasma phosphorylated tau-217 (p-tau-217) / unphosphorylated tau-217 (np-tau-217) ratio (p-tau-217R) and tau-PET SUVR values ​​across selected cortical regions and in six Braak stage regions, consistent with the disclosed embodiments. [Figure 19] This provides an overview of the predictive performance of a particular brain tau state prediction model across various brain regions via cross-validation, consistent with the disclosed embodiments. [Figure 20A] The relative effects of predictors in a multivariate model for predicting brain tau state are shown, consistent with the disclosed embodiments. [Figure 20B] The relative effects of predictors in a multivariate model for predicting brain tau state are shown, consistent with the disclosed embodiments. [Figure 20C]The relative effects of predictors in a multivariate model for predicting brain tau state are shown, consistent with the disclosed embodiments. [Figure 20D] The relative effects of predictors in a multivariate model for predicting brain tau state are shown, consistent with the disclosed embodiments. [Figure 20E] The relative effects of predictors in a multivariate model for predicting brain tau state are shown, consistent with the disclosed embodiments. [Figure 21] This document demonstrates the predictive performance of a brain tau state prediction model using p-tau217R observed during cross-validation extended to a validation set, consistent with the disclosed embodiments. [Figure 22] This document demonstrates the predictive performance of a brain tau state prediction model using p-tau 217R and apolipoprotein Eε4 (ApoE4) allele counts, consistent with the disclosed embodiments. [Figure 23] This document demonstrates the predictive performance of a brain tau state prediction model that uses p-tau-217 concentration instead of p-tau-217R, consistent with the disclosed embodiments. [Figure 24A] This document demonstrates the differences in SUVR predictive profiles and reliable tau-PET SUVR boundaries using a brain tau state prediction model based on plasma p-tau 217R from the entire cortical gray matter (WCGM) and medial temporal lobe (MTL), consistent with the disclosed embodiments. [Figure 24B] This document demonstrates the differences in SUVR predictive profiles and reliable tau-PET SUVR boundaries using a brain tau state prediction model based on plasma p-tau 217R from the entire cortical gray matter (WCGM) and medial temporal lobe (MTL), consistent with the disclosed embodiments. [Figure 25A] We present reliable SUVR predictive profiles and boundaries for tau-PET SUVR using a plasma p-tau 217R-based brain tau state predictive model for additional cortical regions beyond those shown in Figures 24A and 24B, consistent with the disclosed embodiments. [Figure 25B]We present reliable SUVR predictive profiles and boundaries for tau-PET SUVR using a plasma p-tau 217R-based brain tau state predictive model for additional cortical regions beyond those shown in Figures 24A and 24B, consistent with the disclosed embodiments. [Figure 25C] We present reliable SUVR predictive profiles and boundaries for tau-PET SUVR using a plasma p-tau 217R-based brain tau state predictive model for additional cortical regions beyond those shown in Figures 24A and 24B, consistent with the disclosed embodiments. [Figure 25D] We present reliable SUVR predictive profiles and boundaries for tau-PET SUVR using a plasma p-tau 217R-based brain tau state predictive model for additional cortical regions beyond those shown in Figures 24A and 24B, consistent with the disclosed embodiments. [Figure 26A] This study demonstrates reliable tau-PET SUVR boundaries for six Braak disease stages using a plasma p-tau 217R-based brain tau state prediction model. [Figure 26B] This study demonstrates reliable tau-PET SUVR boundaries for six Braak disease stages using a plasma p-tau 217R-based brain tau state prediction model. [Figure 26C] This study demonstrates reliable tau-PET SUVR boundaries for six Braak disease stages using a plasma p-tau 217R-based brain tau state prediction model. [Figure 26D] This study demonstrates reliable tau-PET SUVR boundaries for six Braak disease stages using a plasma p-tau 217R-based brain tau state prediction model. [Figure 26E] This study demonstrates reliable tau-PET SUVR boundaries for six Braak disease stages using a plasma p-tau 217R-based brain tau state prediction model. [Figure 26F] This study demonstrates reliable tau-PET SUVR boundaries for six Braak disease stages using a plasma p-tau 217R-based brain tau state prediction model. [Figure 27]This document shows the upper limit of reliable predictions for tau-PET SUVR using a brain tau state prediction model for Braak staging regions and selected cortical regions, consistent with the disclosed embodiments. [Figure 28] The ability of a brain tau state predictive model to identify subjects in a test set below or above two SUVR thresholds, using plasma p-tau 217R, consistent with the disclosed embodiments, is demonstrated. [Figure 29A] Using a brain tau state model employing plasma p-tau 217R, consistent with the disclosed embodiments, ROC curves and corresponding AUROC values ​​are shown to predict tau positivity and a tau-PET SUVR level of 1.5 in selected cortical regions and six Braak stage regions. [Figure 29B] Using a brain tau state model employing plasma p-tau 217R, consistent with the disclosed embodiments, ROC curves and corresponding AUROC values ​​are shown to predict tau positivity and a tau-PET SUVR level of 1.5 in selected cortical regions and six Braak stage regions. [Figure 29C] Using a brain tau state model employing plasma p-tau 217R, consistent with the disclosed embodiments, ROC curves and corresponding AUROC values ​​are shown to predict tau positivity and a tau-PET SUVR level of 1.5 in selected cortical regions and six Braak stage regions. [Figure 29D] Using a brain tau state model employing plasma p-tau 217R, consistent with the disclosed embodiments, ROC curves and corresponding AUROC values ​​are shown to predict tau positivity and a tau-PET SUVR level of 1.5 in selected cortical regions and six Braak stage regions. [Figure 30] The results of repeating the performance evaluations described in Figures 28 and 29A-29D for each demographic and genomic subgroup in the validation set, consistent with the disclosed embodiments, are shown. [Figure 31]This demonstrates the significant potential reduction in PET scans achievable by utilizing a brain tau state prediction model using p-tau217R, consistent with the disclosed embodiments. [Figure 32A] The predicted amyloid-PET CL values ​​are consistent with the disclosed embodiments. [Figure 32B] The predicted tau-PET values ​​are consistent with the disclosed embodiments. [Figure 33] A table summarizing patient characteristics in the study dataset, consistent with the disclosed embodiments, is shown. [Figure 34] The performance of the SGB and BLLR models on the VC-1 and VC-2 datasets, consistent with the disclosed embodiments, is shown. [Figure 35A] The percentage of the relative influence of the predictors in the SGB model is shown, consistent with the disclosed embodiments. [Figure 35B] The ICE profiles, showing individual subject-level and mean outcomes, are presented in accordance with the disclosed embodiments. [Figure 35C] The ICE profiles, showing individual subject-level and mean outcomes, are presented in accordance with the disclosed embodiments. [Figure 36A] The CL prediction ranges and observed CL values ​​for three prediction models using different combinations of biomarker inputs, consistent with the disclosed embodiments, are shown. [Figure 36B] The CL prediction ranges and observed CL values ​​for three prediction models using different combinations of biomarker inputs, consistent with the disclosed embodiments, are shown. [Figure 36C] The CL prediction ranges and observed CL values ​​for three prediction models using different combinations of biomarker inputs, consistent with the disclosed embodiments, are shown. [Figure 37] The performance of a predictive model for predicting amyloid states across a set of CL thresholds, consistent with the disclosed embodiments, is demonstrated. [Figure 38A]ROC curves and AUROC values ​​for different combinations of CL levels and models for the subject in VC-1 and VC-2, consistent with the disclosed embodiments, are shown. [Figure 38B] ROC curves and AUROC values ​​for different combinations of CL levels and models for the subject in VC-1 and VC-2, consistent with the disclosed embodiments, are shown. [Figure 38C] ROC curves and AUROC values ​​for different combinations of CL levels and models for the subject in VC-1 and VC-2, consistent with the disclosed embodiments, are shown. [Figure 38D] ROC curves and AUROC values ​​for different combinations of CL levels and models for the subject in VC-1 and VC-2, consistent with the disclosed embodiments, are shown. [Figure 38E] ROC curves and AUROC values ​​for different combinations of CL levels and models for the subject in VC-1 and VC-2, consistent with the disclosed embodiments, are shown. [Figure 38F] ROC curves and AUROC values ​​for different combinations of CL levels and models for the subject in VC-1 and VC-2, consistent with the disclosed embodiments, are shown. [Figure 38G] ROC curves and AUROC values ​​for different combinations of CL levels and models for the subject in VC-1 and VC-2, consistent with the disclosed embodiments, are shown. [Figure 38H] ROC curves and AUROC values ​​for different combinations of CL levels and models for the subject in VC-1 and VC-2, consistent with the disclosed embodiments, are shown. [Figure 39] This document demonstrates the predicted effects of using a CL predictive model to screen subjects for confirmatory PET scans, consistent with the disclosed embodiments. [Figure 40] The patient characteristics of a cohort for constructing a brain Aβ predictive model consistent with the disclosed embodiments are shown. [Figure 41A]Further details are provided regarding the distribution of Aβ42 / Aβ40 ratios and p-tau 217R values ​​in the training cohort, consistent with the disclosed embodiments. [Figure 41B] Further details are provided regarding the distribution of Aβ42 / Aβ40 ratios and p-tau 217R values ​​in the training cohort, consistent with the disclosed embodiments. [Figure 42] The odds ratios and significance levels of each predictor in the combined biomarker prediction model, consistent with the disclosed embodiments, are shown. [Figure 43] The ROC curves and AUROC values ​​for three different CL values ​​are shown, consistent with the disclosed embodiments. [Figure 44] The predicted amyloid-beta PET status is shown, classified by the demographic characteristics of a patient cohort obtained through community-based screening, consistent with the disclosed embodiments. [Modes for carrying out the invention]

[0013] The following detailed description refers to the accompanying drawings. Wherever possible, the same reference numerals are used in the drawings and the following description to refer to the same or similar parts. While several exemplary embodiments are described herein, modifications, adaptations, and other implementations are possible. For example, substitutions, additions, or alterations may be made to components illustrated in the drawings, and the illustrative methods described herein may be modified by substituting, rearranging, removing, or adding steps to the disclosed methods. Therefore, the following detailed description is not limited to the disclosed embodiments and examples. Instead, the appropriate scope is defined by the appended claims.

[0014] Alzheimer's disease (AD) is a progressive primary neurodegenerative disease characterized by cognitive and functional decline, posing significant global challenges to patients, healthcare providers, and healthcare systems. Key pathological features of Alzheimer's disease include the accumulation of amyloid-beta (Aβ) plaques and tau protein aggregation in the brain, which can lead to neuronal dysfunction, synapse and neuronal loss, and ultimately, clinical deterioration. Such tau proteins (or tau) belong to the microtubule-associated protein (MAP) family, are primarily expressed in neurons, and are found in axons and dendrites. Tau proteins play a crucial role in the assembly of tubulin monomers into microtubules, constitute the cytoskeleton, and function as tracks for axonal transport. Tau proteins are translated from a single gene located on chromosome 17, and alternative mRNA splicing leads to the formation of six distinct central nervous system tau isoforms, five of which are found in the adult human brain. These isoforms are distinct, possessing three (R1, R3, and R4) or four (R1-R4) repeating regions in their carboxyl (C)-terminus, and exhibiting diverse microtubule-binding region (MTBR) appearances. The amino (N)-terminal domains, which establish junctions between microtubules and other parts of the cytoskeleton or the cell membrane, have a variable presence of 0, 1, or 2 inserts of 29 amino acids. The full-length tau isoform sequences are provided in PCT / US Patent Application Publication 2023 / 081441, which are incorporated herein by reference in their entirety. Similarly, the sequences of therapeutic anti-amyloid and anti-tau antibodies available in the disclosed embodiments are provided in PCT / US Patent Application Publication 2023 / 081441, which are incorporated herein by reference in their entirety.

[0015] PET imaging serves as the gold standard for visualizing and quantifying Alzheimer's disease-related pathologies in vivo. The use of PET tracers targeting amyloid-beta plaques and tau aggregates allows clinicians to non-invasively evaluate these pathological features, supporting diagnosis, prognosis, and treatment monitoring. In some embodiments, tau-PET may refer to tau positron emission tomography. “Tau-PET level” may be identified by the standardized uptake ratio (SUVr) compared to a reference region, as measured by tau-PET imaging. As used herein, “tau-PET level,” “tau level in the brain,” “cerebral tau level,” and “tau load” are used interchangeably. As used herein, tau-PET level refers to a measurement of tau levels in a brain region, e.g., the temporal lobe, as measured by PET. Methods for calculating tau-PET SUVr are known in the art and may include those described herein. In some embodiments, quantitative analysis of the standard uptake ratio of tau-PET levels is completed using PMOD PNEURO Biomedical Image Quantification Software (PMOD Technologies, Zurich, Switzerland).

[0016] Over the past decade, significant advances in tau-specific PET radioligands such as MK6240 have facilitated targeted visualization of tau pathogenesis in vivo. These radioligands offer the ability to quantify tau accumulation across various brain regions, providing valuable insights into the spatial and temporal progression of tau pathogenesis in Alzheimer's disease. In some embodiments, tau pathogenesis can refer to pathological forms of tau, such as intracellular fibrous condensates and their components, present in Alzheimer's disease (AD) and other neurodegenerative disorders referred to as tauopathy. The aggregation of highly phosphorylated tau into insoluble paired helical fibrils (PHFs) that accumulate within nerve cells to form neurofibrillary condensates (NFTs) is characteristic of tau pathogenesis. In AD, NFTs typically occur in a neuroanatomically characteristic pattern of increasing severity, generally defined according to Braak stages 1–6, which correlates well with progressive neuronal loss and worsening clinical status. Extracellular tau dispersants are also pathological forms of tau.

[0017] In some embodiments, brain amyloid-beta levels may be determined by longitudinal positron emission tomography (PET) assessment of the uptake of a contrast agent (e.g., an amyloid contrast agent) into the brain. In some embodiments, brain amyloid-beta levels are continuous levels. In some embodiments, a subject is determined to be amyloid-positive or amyloid-negative by evaluation of amyloid PET imaging assessment. In some embodiments, a subject is "amyloid-negative" if PET SUVR negativity is below a threshold determined for the amyloid PET tracer. In some embodiments, the amyloid PET tracer may be florbetabene (e.g., 18F-florbetabene (Nueraceq®)), florbetapil (e.g., 18F-florbetapil (Amyvid®)), and / or flutamethamol (e.g., 18F-flutamethamol (Vizamyl®)). In some embodiments, the PET SUVR threshold for the amyloid PET tracer is approximately 1.17, and a measurement below this threshold may indicate that the subject is "amyloid-negative".

[0018] As those skilled in the art will recognize, amyloid-beta levels from amyloid PET can be reported using the centiloid method in "centiloid" units (CL). (Klunk WE et al. The Centiloid Project: standardizing quantitative amyloid plaque estimation by PET. Alzheimer's Dement. 2015;11:1-15 e1-4). The centiloid method measures the tracer on a scale of 0 CL to 100 CL, where 0 is considered the anchor point and represents the mean in young, healthy controls, and 100 CL represents the mean amyloid load present in subjects with mild to moderate severity dementia due to AD. (Id.) An elevation in amyloid levels can be set against a baseline threshold in healthy controls determined according to methods known to those skilled in the art (POSA). For example, a centiloid value of 32.5 can be used as a threshold for "elevated amyloid," and a "moderate amyloid" level may refer to a centiloid value in the range of 20 to 32.5 CL (e.g., 30 CL). In another example, a centiloid value of 40 can be used as a threshold for "amyloid elevation," and a "moderate amyloid" level may refer to centiloid values ​​in the range of 20–40 CL.

[0019] Further details regarding amyloid and tau-PET imaging are provided in International Publication No. 2024 / 118665, which is incorporated herein by reference in its entirety.

[0020] While PET imaging holds significant promise for in vivo assessment of tau and / or amyloid-beta pathogenesis, its widespread clinical adoption may be hindered by cost, limited cyclotron availability, limited patient accessibility, the scarcity of approved tau-PET or amyloid-PET tracers, and the need for specialized expertise for scan interpretation. In some cases, measurements of biomarkers in cerebrospinal fluid (CSF) and blood (e.g., plasma, serum, whole blood) may offer more readily available and cost-effective options for assessing tau and / or amyloid-beta pathogenesis. In particular, plasma biomarkers (e.g., plasma markers of tau phosphorylated at specific sites, such as 217 (p-tau 217) or 181 (p-tau 181)) have emerged as promising options for characterizing central nervous system pathogenesis in Alzheimer's disease. Plasma concentrations of p-tau 217 may correlate with the diagnosis, severity, and progression of Alzheimer's disease. The ratio of p-tau-217 to np-tau-217 (p-tau-217R) in plasma may reflect observed tau or amyloid-beta pathology in the brain.

[0021] As disclosed herein, predictive models can predict brain tau or amyloid-beta status in the entire brain of interest (e.g., the entire cortical gray matter), brain regions designated in neuroanatomical atlases (e.g., Hammers Atlas, Desikan-Killiany Atlas, Harvard-Oxford Atlas, Automated Anatomical Labeling Atlas, Brainnetome Atlas, etc.) (e.g., hubs as described herein), sets of such brain regions (e.g., modules as described herein), or other desired parts of the brain of interest.

[0022] In some embodiments, the cerebral tau state may include indicators of tau concentrates or tau deposition, while the cerebral amyloid-beta state may include indicators of amyloid-beta deposition. Such indicators may relate to the whole brain, a brain region (e.g., a hub), a collection of brain regions (e.g., a module), etc. Such indicators for tau or amyloid-beta may be continuous values ​​(e.g., tau-PET or amyloid-beta SUVR levels, amyloid-beta centroid levels) or discrete values ​​(e.g., binary classifications representing the fulfillment of diagnostic criteria such as tau or amyloid-beta positivity; multi-class classifications indicating the stage or class of cerebral tau or amyloid-beta deposition, etc.). In some embodiments, the cerebral tau state may include predictions of tau-PET SUVR levels in multiple hubs or modules within the brain. In some embodiments, the cerebral amyloid-beta state may include predictions of amyloid-beta SUVR and / or centroid levels in multiple hubs or modules within the brain. Such predictions may be made along a continuum of early Alzheimer's disease. In some embodiments, predictive models can predict regional brain tau or amyloid-beta levels in Aβ+ early-stage Alzheimer's disease patients. Therefore, such predictive models can reduce the need for PET scans to identify patients with varying degrees of brain tau or amyloid-beta accumulation. Simultaneous prediction of tau and / or amyloid-beta levels across multiple brain regions can provide increased flexibility and accessibility in patient screening and monitoring procedures for both clinical trials and real-world clinical settings.

[0023] As can be understood, predictive models consistent with the disclosed embodiments can be used as a patient screening tool to exclude individuals predicted to have no or low levels of tau or amyloid-beta accumulation from clinical trials or therapies, thus limiting the need for expensive and burdensome tau or amyloid-beta PET scans for patients before or during treatment. Such an approach may reduce the burden on patients and result in cost and time savings.

[0024] Predictive models consistent with the disclosed embodiments can predict continuous tau levels over a wide range of tau levels and / or continuous amyloid-beta levels over a wide range of amyloid-beta levels. Such continuous predictions improve binary positive-negative assessments by enabling direct patient monitoring, evaluation of treatment effectiveness (e.g., whether treatment reduces predicted tau or amyloid-beta levels in the patient's brain), and evaluation of disease progression (e.g., whether predicted tau or amyloid-beta levels are increasing over time in the patient's brain).

[0025] In some embodiments, predictive models consistent with the disclosed embodiments exhibit good performance (e.g., mean squared error (MSE) or root mean squared error (RMSE), mean absolute error (MAE), coefficient of determination (R). 2 ), receiver operating characteristic (ROC) curve or area under the ROC curve (AUROC), confusion matrix, sensitivity or selectivity, specificity, or as measured by such area under the curve, precision and reproducibility, F scale, or any other suitable performance measure can be achieved. In some embodiments, the output range of the predictive model consistent with the disclosed embodiments may extend to the range of observed outputs (e.g., tau-PET SUVR values ​​obtained from patients, amyloid-β SUVR values ​​obtained from patients, amyloid-β centroid values ​​obtained from patients, etc.). Furthermore, tau predictive models may demonstrate consistent performance across demographic factors (e.g., age, sex) and genomic factors (e.g., ApoE4 status) for whole-brain and regional tau level predictions. Thus, such models improve upon predictive models that provide a single binary prediction (e.g., presence or absence of tau accumulation in the patient's brain).

[0026] In some embodiments, predictive models consistent with the disclosed embodiments can assist in identifying patients for further testing (e.g., imaging, cognitive assessment tests), monitoring (e.g., follow-up by clinicians), and / or treatment (e.g., administration of therapeutic agents such as anti-tau antibodies or anti-amyloid-beta antibodies). Suitable therapeutic anti-amyloid-beta antibodies are known in the art and include lecanemab, while suitable therapeutic anti-tau antibodies are also known and include E2814. The predictive capabilities of the disclosed predictive models may enable physicians and caregivers to adjust optimal treatment and care strategies, which may lead to improved clinical outcomes.

[0027] Predictive models consistent with the disclosed embodiments may include statistical and machine learning models suitable for identifying relationships between input data and output results. Such models may include regression models (e.g., logistic regression models; ridge, lasso, or elastic network regression models; time series regression models, etc.), support vector machines, gradient boosting machines, Bayesian classifiers, neural networks, decision trees, random forests, other ensemble-based models, or other suitable statistical and machine learning models.

[0028] In some embodiments, preferred predictive models may include regularized logistic regression models and ensemble tree-based models. Regularized logistic regression models may include Bayesian elastic net models or Bayesian linear lasso regression (BLLR). Such models can use preferred prior distributions selected to reduce the complexity of the model (e.g., mixed double-exponential prior distributions such as spike-and-slab mixed double-exponential prior distributions), thus preventing overfitting and increasing the robustness of the model.

[0029] In accordance with the disclosed embodiments, the ensemble tree-based model may include a stochastic gradient boosting (SGB) model that can combine predictions from multiple decision trees to generate a final prediction. Each node in the multiple decision trees can be trained using a different random subset of input features. Thus, the individual decision trees are different and can potentially capture different signals from the data. In some embodiments, training an SGB model may involve iteratively strengthening a series of simple decision trees over a large number of iterations to yield a more refined prediction. This iterative process can progressively correct errors from previous models and eventually converge to a more accurate prediction. In this way, the SGB model can dynamically adapt to the complexity of the data during training to acquire nonlinearity and predictor interactions without pre-specifying or imposing restrictive assumptions about distributions or mathematical relevances.

[0030] Suitable input data for a predictive model consistent with the disclosed embodiments may include biomarker data, demographic data, genomic data, cognitive measurement data, and / or image data. In some embodiments, the input data for a predictive model consistent with the disclosed embodiments may include biomarker data and demographic data. In some embodiments, the input data for a predictive model consistent with the disclosed embodiments may include biomarker data, demographic data, and genomic data. In some embodiments, the input data for a predictive model consistent with the disclosed embodiments may include biomarker data, demographic data, genomic data, and one or more of cognitive measurement data and image data. In some embodiments, the input data for a predictive model consistent with the disclosed embodiments may not include cognitive measurement data or image data.

[0031] In some embodiments, biomarker data may represent biomarker levels or functions of biomarker levels (e.g., normalized levels of a biomarker, levels of two biomarkers, or ratios of normalized levels). Such biomarker data may indicate whether the biomarker level or function of biomarker levels satisfies a condition (e.g., detectable or undetectable, exceeding a threshold) or falls within a category (e.g., indicating a level that maps to "high" in a low-transition-high scheme). Biomarker data may be continuous, Boolean, or categorical values. In some embodiments, biomarker data may be plasma, serum, or cerebrospinal fluid biomarker data. In some embodiments, biomarker data may relate to phosphorylated tau levels, unphosphorylated tau levels, or functions or combinations thereof. For example, biomarker data may relate to p-tau 181 or p-tau 217 levels. As an additional example, biomarker data may relate to the p-tau 217 / np-tau 217 ratio (referred to herein as p-tau 217R). In some embodiments, biomarker data may relate to amyloid-beta protein levels, monomer levels, or functions or combinations thereof. For example, biomarker data may relate to amyloid-beta 1-42 (Aβ42) levels or amyloid-beta 1-40 (Aβ40) levels. As an additional example, biomarker data may relate to the Aβ42 / Aβ40 ratio. Information on biomarkers including p-tau 181, p-tau 217, Aβ42, and Aβ40, and methods for measuring such biomarkers, can be found in PCT / U.S. Patent Application Publication No. 2022 / 073576, which is incorporated in whole by reference. As exemplary examples, several biomarker levels (e.g., plasma p-tau 217, plasma np-tau 217, plasma Aβ42, and / or plasma Aβ40 levels) can be quantified by immunoprecipitation-mass spectrometry.In some embodiments, certain biomarker levels (e.g., plasma p-tau 181, neurofilamentous light chain (NfL), and / or glial fibrillary acidic protein (GFAP) levels) may be quantified using single-molecule array assays. For example, amyloid levels may be measured using immunoassays (e.g., Quanterix® Simoa® p-tau assay), C2N Diagnostics' mass spectrometry platform (PrecivityAD®) for quantifying levels and / or ratios, and / or methods based on mass spectrometry (IP / LC-MS / MS).

[0032] In some embodiments, demographic data may include demographic characteristics of the subject. For example, demographic data may represent one or more of the following: age, sex, geographical region, occupation, education level, body mass index (BMI), race, country of origin, or ethnicity. In some embodiments, genomic data may include ApoE4 status (e.g., ApoE4 allele count).

[0033] In some embodiments, cognitive measurement data may be or include at least one assessment value such as the Total Clinical Dementia Scale (CDR-SB) measurement, the Alzheimer's Disease Composite Score (ADCOMS) measurement, the Alzheimer's Disease Assessment Scale (ADAS) measurement (e.g., ADAS-Cog-14), the Alzheimer's Disease Collaborative Study - Activities of Daily Living (ADCS-ADL) score, the Mini-Mental State Examination (MMSE) score, the Integrated Alzheimer's Disease Assessment Scale (iADRS) score, the Wechsler Memory Scale, Fourth Edition - Logical Memory (Subscale) I (WMS-IV LMI) score, or the Wechsler Memory Scale, Fourth Edition - Logical Memory (Subscale) II (WMS-IV LMII) score. As can be understood, cognitive measurement data may be or include any preferred component or subscore of the aforementioned assessments. Non-exclusive examples of preferred components or subscores include delayed word recall (e.g., ADCDRL), word recall (e.g., ADCRL), ADCCMD (e.g., direction), ADCOF (e.g., object naming), ADCCP (e.g., acts such as constructing or ideating), CDR0101 (e.g., memory), orientation (e.g., time and place), encoding (e.g., names of objects), attention and calculation (e.g., spelling), language (e.g., conversation), word recognition, comprehension, and word retrieval.

[0034] In some embodiments, the image data may include raw or processed MRI, PET, or CT images, etc. In some embodiments, the image data may include measurements derived from such images (e.g., volume, surface area, cortical thickness, etc.). In some embodiments, the image data may include images and / or measurements of brain regions identified (e.g., by an algorithm) as having specific predictive values.

[0035] In some embodiments, the tau PET level is calculated from the overall tau load from the whole-brain signal. In some embodiments, the “tau PET level” may be identified in tau PET imaging by a standardized uptake ratio (SUVr or SUVR) to determine, for example, the level of neurofibrillary ensembles in a brain region or the whole brain. SUVr may be a measure of tau PET tracer uptake in a region of the patient’s brain compared (e.g., normalized) to a reference region in the same patient. In some embodiments, elevated levels of tau, e.g., tau ensembles, may be determined by comparing the PET level in the subject to that in a control subject without AD. Methods for calculating tau PET SUVr are known in the art and may include quantitative analysis of SUVr (e.g., computer processing) by PMOD PNEURO Biomedical Image Quantification Software (PMOD Technologies, Zurich, Switzerland). In some embodiments, the tau PET level is evaluated by a PET tracer.

[0036] In some embodiments, the cerebral tau state may include any indicator of the degree of neurofibrillary tau condensation or tau deposition in the brain, including presence or absence, quantity, location, etc. In some embodiments, the cerebral tau state may relate to the entire brain (e.g., the entire cortical gray matter) or to one or more regions of the brain. In some embodiments, the cerebral tau state may relate to multiple regions of the brain. In some embodiments, multiple regions may correspond to one or more Braak stages.

[0037] In various embodiments, the Braak staging system may be used to assess tau levels in the brain based on the anatomical localization of tau neurofibrillary condensates or phosphotau (e.g., Braak H, Braak E. Neuropathological staging of Alzheimer-related changes. Acta Neuropathol. 1991;82:239-59.), or based on a PET-based Braak staging system (e.g., as described by Therriault et al., Nature Aging volume 2, pp. 526-535 (2022)). Thus, Braak regions may refer to anatomical regions typically affected by tau aggregation in Braak staging (e.g., the temporal cortex, which is generally affected in early AD). For example, early Braak regions may refer to the temporal lobe, such as the medial temporal, metatemporal, or temporal lobe, or the entorhinal cortex and / or hippocampus.

[0038] Generally, the early Braak stages can be characterized as follows: Braak stage I may be characterized by tau aggregation in the entorhinal cortex; Braak stage II may be characterized by tau aggregation in the hippocampus; and Braak stage III may be characterized by tau aggregation in the amygdala, parahippocampal gyrus, fusiform gyrus, and lingual gyrus. The later Braak stages, for example, Braak stages IV and V, may be characterized by tau aggregation in the association cortex; and Braak stage VI may be characterized by tau aggregation in the primary sensory cortex.

[0039] In some preferred embodiments, the tau-PET staging classification may use the following Braak classification: Braak I (entorhinal cortex); Braak II (hippocampus); Braak III (amygdala, parahippocampal gyrus, fusiform gyrus, and lingual gyrus); Braak IV (middle temporal, caudal anterior cingulate cortex, rostral anterior cingulate cortex, posterior cingulate cortex, isthmus of the cingulate cortex, insula, inferior temporal, temporal pole); Braak V (superior frontal, lateral orbitofrontal cortex, medial orbitofrontal cortex, frontal pole, caudal middle frontal, rostral middle frontal, operculum, orbital, triangular, lateral occipital, suprapetrimal, inferior parietal, superior temporal, superior parietal, precuneus, bank of the superior temporal sulcus, transverse temporal cortex); and Braak VI (pericalvinic cortex, postcentral gyrus, cuneiform cortex, gyrus precentral, paracentral lobule, pericalvinic cortex, postcentral gyrus, cuneiform cortex, gyrus precentral, paracentral lobule).

[0040] In some embodiments, an alternative tau-PET staging system following Therriault et al., Nature Aging volume 2, pp. 526-535 (2022) may be used to classify the disease stage and brain regions as follows: Braak stage I (transitional entorhinal), Braak stage II (entorhinal and hippocampal), Braak stage III (amygdala, parahippocampal gyrus, fusiform gyrus and lingual gyrus), Braak stage IV (insula, inferior temporal, lateral temporal, posterior cingulate gyrus and inferior parietal), Braak stage V (orbitofrontal, superior temporal, inferior frontal, cuneiform, anterior cingulate gyrus, supramarginal gyrus, external occipital, precuneus, superior parietal, superior frontal and rostromedial) and Braak stage VI (paracentral, postcentral gyrus, precentral and pericarpal sulcus).

[0041] In the Braak staging system (e.g., tau-PET Braak staging), the classification of disease stage and structure may differ slightly between methods due to natural variability among patients and / or variations in staging methods (e.g., different tau-PET tracers and analytical methods). The results of tau-PET Braak staging may differ from the staging determined during autopsy.

[0042] In some embodiments, multiple regions include two or more regions designated in a neuroanatomical atlas. In some embodiments, a region of the brain may be a meta-region consisting of all or part of two or more anatomical regions. The temporal region may, for example, include at least a portion of the temporal lobe. For example, the temporal region may include the superior posterior, superior anterior, posterior, middle inferior, and fusiform gyrus of the temporal lobe. The temporal region may include these structures from both the left and right hemispheres of the brain. In some embodiments, a meta-region of interest (ROI) is defined as a region of the brain having certain characteristics, such as a region with the most tau deposition in AD patients, or a region where tau PET imaging differs among groups of patients (e.g., cognitively unimpaired individuals, e.g., those with normal amyloid PET compared to cognitively unimpaired individuals with abnormal amyloid PET). Examples of meta-ROIs in the temporal region can be found in International Publication No. 2024 / 118665, both of which are incorporated herein by reference, and in Jack et al., Alzheimer's Dementia 13, 205-216 (2017).

[0043] In some embodiments, the multiple regions include regions as specified in the neuroanatomical atlas. In some embodiments, the multiple regions include at least 2, 5, 10, 20, 50, 100, 200, and 500 regions; up to 500, 200, 100, 50, 20, 10, 5, and 2 regions; or a number of regions between any two preceding lower and upper limits. In some embodiments, the multiple regions include combinations of such regions. For example, regions corresponding to the major cortical lobes of the brain (e.g., cortex, parietal, occipital, lateral temporal, and medial temporal) and the entire cortical gray matter (e.g., the entire cortical gray matter or WCGM) may be combined into a composite region, resulting in six such composite regions. Predictions regarding composite regions corresponding to the major cortical lobes and WCGM of the brain may be evaluated in addition to predictions regarding composite regions corresponding to Braak stages.

[0044] In some embodiments, the predicted brain tau or amyloid-beta state may be a predicted brain tau or amyloid-beta level, such as a predicted quantity or amount. The brain tau or amyloid-beta level may be a continuous, Boolean, or categorical value. In some embodiments, the brain tau or amyloid-beta level is a continuous level. In some embodiments, the predicted brain tau or amyloid-beta level may refer to a level that will be measured by a PET scan (e.g., tau-PET SUVR, amyloid-PET SUVR, amyloid-PET centroid value), other imaging modalities (e.g., MRI or other modalities as described herein), histopathological methods, etc. Methods for calculating tau-PET or amyloid-PET SUVR levels are known in the art and may include those described herein.

[0045] In some embodiments, medical record information may include medical records, case notes, clinical trial records, requirements information (e.g., relating to biomarker testing), image data, or clinical test results. In some embodiments, medical record information may include (or be available for generating) patient medical data.

[0046] Figure 1 shows an exemplary platform 100 for developing, validating, and deploying predictive models for predicting brain tau or amyloid-beta levels in a subject, consistent with the disclosed embodiments. The platform 100 may be configured to acquire patient clinical data from other systems (e.g., electronic health record (EHR) systems, insurance systems, imaging systems, or medical testing systems, not shown in Figure 1) or records 101.

[0047] In accordance with the disclosed embodiments, such patient clinical data may include image data, biomarker data, genomic data, demographic data, cognitive measurement data, etc. Platform 100 may be configured to generate datasets suitable for training predictive models using components such as an extract-transformation-load (ETL) engine 110 and a dataset creation engine 115. Platform 100 may be configured to train models using a training engine 120. The trained models may be used in the prediction phase by a prediction engine 130. Users can control and configure Platform 100 by interacting with a user device 199. Users can also provide target data to Platform 100 and receive predictions from Platform 100 by interacting with the user device 199.

[0048] As can be understood, the specific arrangement of components shown in Figure 1 is not intended to be limiting. Platform 100 may include additional components (e.g., additional databases, data sources, processing systems, etc.) or fewer components (e.g., by combining databases or processing systems). The functionality of existing components may be combined or distributed among additional systems without departing from the envisioned embodiments.

[0049] The components in Figure 1 may be implemented using one or more computing systems (e.g., laptops, desktops, workstations, computing clusters, on-premises or off-premises cloud computing platforms). For example, a computing cluster or workstation may implement the ETL engine 110, the dataset creation engine 115, or the training engine 120. As an additional example, a desktop or laptop (e.g., user device 199 or another device) may implement the prediction engine 130. As an additional example, components of platform 100 (e.g., apart from user device 199) may be implemented using containerized services on a cloud computing platform. As should be understood, these examples are not intended to be limiting.

[0050] In accordance with the disclosed embodiments, record 101 may include one or more storage locations for data available to platform 100 for predicting cerebral tau or amyloid-beta status. In various embodiments, such data may include medical record information relating to a subject. In some embodiments, the medical record information contained in record 101 may include (or may be available for generation of) patient medical data as described herein.

[0051] In accordance with the disclosed embodiments, the ETL engine 110 may be configured to acquire data in various forms from one or more sources (e.g., record 101). The disclosed embodiments are not limited to any particular form of acquired data or method for acquiring such data. For example, acquired data may be or include structured data or unstructured data. The ETL engine 110 may interact with various data sources to receive or retrieve data.

[0052] The ETL engine 110 can convert the data into a suitable format and load the converted data into a target component or database of the platform 100. In some embodiments, converting the data may include performing quality control processing on the acquired data. Such quality control processing may include verifying that the data is usable (e.g., that the subject meets inclusion criteria for the model to be trained, that the input data required for the subject is complete, etc.). In some embodiments, converting the data may include processing the data into a standard format or structure. As can be understood, input data acquired from record 101 may not be in a suitable format for training a predictive model. Similarly, input data acquired from different records 101 may have different formats. Therefore, the ETL engine 110 may clean the acquired input data so that the input data has a consistent format despite originating from various different sources.

[0053] In some embodiments, the ETL engine 110 may augment image data or medical record information by generating additional data using the image data or medical record information. For example, the ETL engine 110 may convert biomarker levels into scores (e.g., using population distribution information, clinical range, etc.) or normalize image data (e.g., normalizing area and volume by intracranial volume). In some embodiments, the ETL engine 110 may remove unnecessary or undesirable variables or data from the input dataset. For example, if the medical record contains information unrelated to the prediction of tau or amyloid-beta status, the ETL engine 110 may create a version of the medical record that contains only the information relevant to the prediction of tau or amyloid-beta status.

[0054] In accordance with the disclosed embodiments, the ETL engine 110 may load the transformed data into another component of the platform 100, such as a dataset creation engine 115 (or into a suitable data storage device from which the dataset creation engine 115 can retrieve the data).

[0055] In some embodiments, the dataset creation engine 115 may be configured to generate training samples or inference samples from data received from the ETL engine 110. The dataset creation engine 115 may be configured to extract any required input data features from the transformed data received from the ETL engine 110. In some embodiments, the dataset creation engine 115 may generate features based on a combination of biomarkers (e.g., p-tau 217R, Aβ142 / Aβ40 ratio, etc.), determine correlations between input data (e.g., between thickness, area, or volume measurements for different brain regions), convert amyloid-PET SUVR scores to centiloid levels, identify brain regions as modules or hubs as described herein, or perform other feature extraction.

[0056] In some embodiments, the dataset creation engine 115 may be configured to accept label information provided by the user through the user device 199. For example, the dataset creation engine 115 may be configured to provide data (or metadata about the data) received from the ETL engine 110 to the user device 199 for display. In response, the dataset creation engine 115 may receive label information (e.g., identification of subjects having specific brain tau or amyloid-beta levels, tau or amyloid PET image labels, Braak stage information, etc.).

[0057] In some embodiments, the dataset creation engine 115 may be configured to associate labels with training samples. For example, when predicting brain tau or amyloid-beta status, the data may include labels indicating a measurement or determination of brain tau or amyloid-beta status. In some embodiments, the prediction may relate to brain tau or amyloid-beta levels measured using brain imaging data, such as tau-PET or amyloid-PET image data. In such embodiments, the labels may be or indicate the brain tau or amyloid-beta levels measured for the training sample. In some embodiments, the units for the prediction and the units for the labels may be the same (e.g., SUVR, centiloid, etc.). The dataset creation engine 115 can associate such labels with input data. The training sample may then include input data and associated labels for the subject. As an additional example, when predicting brain tau or amyloid-beta status, findings of the presence of brain tau or amyloid may be described in the subject's medical record (e.g., based on a biomarker assay or visual reading of radiotracers in PET image data). Next, training examples may include indicators of findings regarding the presence of brain tau or amyloid in the input data for the subject. The dataset creation engine 115 may be configured to store the training samples in the data storage device 105.

[0058] In accordance with the disclosed embodiments, the model storage device 103 may be a storage location for models available to components of the platform 100 (e.g., the training engine 120 or the prediction engine 130). The disclosed embodiments are not limited to any particular implementation of the model storage device 103. In accordance with the disclosed embodiments, the model storage device 103 may be implemented using one or more relational databases, object-oriented or document-oriented databases, tabular data stores, graph databases, distributed file systems, or other preferred data storage devices.

[0059] In accordance with the disclosed embodiments, the data storage device 105 may be a storage location for prepared datasets available to the training engine 120 or the prediction engine 130. The disclosed embodiments are not limited to any particular implementation of the data storage device 105. In accordance with the disclosed embodiments, the data storage device 105 may be implemented using one or more relational databases, object-oriented or document-oriented databases, tabular data stores, graph databases, distributed file systems, or other preferred data storage devices.

[0060] In accordance with the disclosed embodiments, the training engine 120 may be configured to train or create and train models. The training engine 120 may be configured to create models (for example, in response to a command to create a particular type of trained model using an input dataset) or to retrieve existing models from the model storage device 103. The training engine 120 may be configured to create or train models using a training dataset retrieved from the data storage device 105. In some embodiments, the training engine 120 may be configured to store trained models in the model storage device 103.

[0061] In accordance with the disclosed embodiments, the training engine 120 may include model training and model evaluation components. The training engine 120 may be configured to train a model using the model training components and then determine a performance metric value for the model using the model evaluation components.

[0062] In some embodiments, the training engine 120 may provide the model evaluation component with cross-validation or holdout portions of the model and training dataset. In some embodiments, the training engine 120 may specify one or more performance metrics as disclosed herein. In addition or alternatively, the model evaluation component may be configured with a predetermined or default set of such performance metrics. In some embodiments, the performance metric values ​​may be displayed to the user through the user device 199. The user may then interact with the training engine 120 through the user device 199 to update the model.

[0063] In some embodiments, the training engine 120 may automatically update the model being trained based on performance metrics. In various embodiments, the training engine 120 may update the model being trained in response to user input provided through the user device 199. Updating the model may include one or more of the following: performing additional training (e.g., using an existing training dataset or a different training dataset), modifying the model (e.g., changing the input features used by the model, changing the model's architecture, etc.), or changing the training environment (e.g., changing the training hyperparameters, changing the training, cross-validation, and holdout divisions of the training dataset, etc.).

[0064] In accordance with the disclosed embodiments, the prediction engine 130 may be configured to predict the brain tau or amyloid-beta state of a subject using a prediction model. In some embodiments, the prediction engine 130 may retrieve a trained model from the model storage device 103. In some embodiments, the prediction engine 130 may retrieve patient clinical data (e.g., subject data) of a subject from the data storage device 105. In some embodiments, the prediction engine 130 may retrieve subject data from another data storage location. This alternative data storage location may be associated with another entity or user. For example, the prediction engine 130 may receive or retrieve subject data from a healthcare system controlled by an entity different from the entity controlling the prediction engine 130.

[0065] In accordance with the disclosed embodiments, the target data may include demographic data, cognitive measurement data, genomic data, image data, biomarker data, or other suitable target data. The prediction engine 130 may apply the target data to a trained prediction model to provide a predicted brain tau or amyloid-beta state as an output. The output may be provided by the prediction engine 130 to a user device 199. In some embodiments, the output may be stored on a computing device associated with the platform 100 or provided to another system.

[0066] In accordance with the disclosed embodiments, the user device 199 may provide a user interface for interacting with other components of the platform 100. The user interface may be a graphical user interface. The user interface may enable the user to configure the ETL engine 110 to extract, transform, and load data according to user specifications. The user interface may enable the user to specify how the transformed data received by the dataset creation engine 115 should be transformed into labeled training data (or preferred patient data). In some embodiments, the user interface may enable the user to interact with the dataset creation engine 115 to manually or semi-manually label or annotate the training data. In some embodiments, the user interface may enable the user to provide data or a model to the training engine 120 for training, or to provide data or a model to the prediction engine 130 for identification and classification.

[0067] In some embodiments, the user interface may allow the user to interact with the training engine 120 to create or select a model for training, create or select a dataset to be used in training the model, or select training parameters or hyperparameters. In some embodiments, the user interface may allow the user to interact with the training engine 120 to display information related to training the model (e.g., performance metric values, loss function values ​​or their function during training, or other training information). In some embodiments, the user interface may allow the user to interact with the prediction engine 130 to select a training model and patient data (e.g., base images). In some embodiments, the user interface may allow the user to interact with the prediction engine 130 to display prediction data, store prediction data on a computing device, or send prediction data to another system.

[0068] The components of platform 100 may be implemented using one or more computing devices. Such computing devices may include tablets, laptops, desktops, workstations, computing clusters, or cloud computing platforms. In some embodiments, the components of platform 100 may be implemented using a cloud computing platform. For example, one or more of the ETL engine 110, dataset creation engine 115, training engine 120, and prediction engine 130 may be implemented on a cloud computing platform. In some embodiments, the components of platform 100 may be implemented using on-premises systems. For example, the record 101 or user device 199 may be an on-premises system or hosted thereon. As an additional example, the model storage device 103 or data storage device 105 may be an on-premises system or hosted thereon.

[0069] The components of platform 100 may communicate using any preferred method. In some embodiments, two or more components of platform 100 may be implemented as microservices or web services. Such components may communicate using messages transmitted over a computer network. Messages may be implemented using SOAP, XML, HTTP, JSON, RCP, or any other preferred format. In some embodiments, two or more components of platform 100 may be implemented as software, hardware, or a combined software / hardware module. Such components may communicate using data or instructions, function calls written to or read from memory (e.g., shared memory), or any other preferred method of communication.

[0070] As can be understood, the specific structure of platform 100 is not intended to be limiting. In accordance with the disclosed embodiments, two or more of the record 101, model storage device 103, or data storage device 105 may be combined or hosted on the same computing device. In accordance with the disclosed embodiments, the ETL engine 110 and the dataset creation engine 115 may be omitted from platform 100. In such embodiments, datasets formatted and configured for use by the training engine 120 or prediction engine 130 may be registered in the data storage device 105 by another system or by another method. In accordance with the disclosed embodiments, the ETL engine 110 and the dataset creation engine 115 may be combined. In such embodiments, data extraction, transformation, and loading may be combined with feature extraction, labeling, and sampling. In accordance with the disclosed embodiments, the training engine 120 and the prediction engine 130 may be combined.

[0071] Although shown with one user device 199, platform 100 may have multiple user devices. Different user devices may be associated with different entities or different users having different roles. For example, user device 199 may be associated with a software engineer or data scientist developing a predictive model, while another user device may be associated with a clinician using the predictive model.

[0072] The user device 199 may be combined with one or more other components of the platform 100. In some embodiments, the user device 199 and at least one of the ETL engine 110, dataset creation engine 115, training engine 120, or prediction engine 130 may be implemented by the same computing device. In various embodiments, the user device 199 and at least one of the model storage device 103 or data storage device 105 may be implemented by the same computing device.

[0073] As can be understood, platform 100 can be integrated into a method for treating a subject or conducting a clinical trial. The prediction engine 130 may use a trained prediction model and input data for the subject to predict the brain tau or amyloid-beta state of the subject. The predicted brain tau or amyloid-beta state can be used to determine a patient treatment plan for the patient. In some embodiments, the brain tau or amyloid-beta state for the subject can be used as a covariate in determining the therapeutic effect in a clinical trial. In some embodiments, the clinical trial acceptance decision may depend on the predicted brain tau or amyloid-beta level. For example, a patient with a predicted brain tau level that meets the inclusion criteria (e.g., Braak stage 3 or 4, but not earlier or later) may be eligible for inclusion in a clinical trial. Similarly, a patient with a predicted amyloid-beta centroid level that meets the inclusion criteria may be included in a clinical trial. In this way, the patient population can be enriched with patients who are likely to demonstrate a benefit from the investigational treatment.

[0074] Figure 2 shows a process 200 for predicting a target brain tau or amyloid-beta state, consistent with the disclosed embodiments. For convenience of explanation, the process 200 is described as being carried out using platform 100. However, the process 200 may also be carried out, at least in part, using a different computing system. The process 200 can be used to predict brain tau or brain amyloid-beta state. When predicting brain amyloid-beta state, patient medical record data may include or be derived from Aβ-SUVR levels or centroid levels. In some such embodiments, the predictive model may output predicted Aβ-SUVR levels or centroid levels, or Aβ categories (e.g., Aβ+, Aβ-, etc.). When predicting brain tau state, patient medical record data may include or be derived from tau-PET SUVR levels. In some such embodiments, the predictive model may output predicted tau-SUVR levels, or tau categories (e.g., tau-positive, tau-negative, etc.). Examples of the development of predictive brain tau and brain amyloid-beta models are provided herein.

[0075] Process 200 may include a dataset creation phase, a training phase, and a prediction phase. In the dataset creation phase, components of platform 100 (e.g., ETL engine 110, dataset creation engine 115, etc.) may acquire training data from a database (e.g., record 101, etc.). In the training phase, components of platform 100 (e.g., training engine 120, etc.) may create or refine a predictive model for predicting brain tau states from the target data. In the prediction phase, components of platform 100 (e.g., prediction engine 130, etc.) may use the predictive model generated in the training phase to predict brain tau state data from the target data. The predicted brain tau state data may be used to manage the treatment of the target, to include or exclude patients from clinical trials, or as a covariate when analyzing data from clinical trials.

[0076] In step 210 of process 200, components of platform 100 (e.g., ETL engine 110, dataset creation engine 115, etc.) may acquire training data in accordance with the disclosed embodiments. In some embodiments, the training data may include subject data consistent with the disclosed embodiments.

[0077] In some embodiments, the target data may include biomarker data as described herein. In some embodiments, the target data may include biomarker data and cognitive measurements, images, genomic data, or demographic data as described herein. For example, the target data may include biomarker data such as p-tau 217 levels, as well as demographic data such as the age, sex, race, and / or education level of the subject. In some embodiments, the target data may include biomarker data and cognitive measurement data or image data as described herein. Such image data may include measurements (e.g., for one or more brain regions identified as hubs) and composite values ​​for one or more clusters of brain regions identified as modules. In some embodiments, the target data may include biomarker data and genomic data. For example, the target data may include ApoE4 allele counts.

[0078] In some embodiments, if the target data includes image data, the image data may relate to measurements of brain regions (e.g., hubs), sets of regions (e.g., modules), or other parts (e.g., volume, surface area, cortical thickness, etc.). In some embodiments, such parts may be identified as having certain predictive values. For convenience, such parts are described herein as being identified by the dataset creation engine 115. However, these parts may also be identified by another preferred system. The dataset creation engine 115 may identify such parts using network analysis or multilevel clustering.

[0079] In some embodiments, the dataset creation engine 115 can identify such hubs and modules using MEGENA, described in U.S. Provisional Patent Application No. 63 / 561,285 and incorporated herein by reference, thereby reducing the dimensionality of the data to be trained and improving the robustness of the trained predictive model.

[0080] In such embodiments, the dataset creation engine 115 may obtain an MRI regionality scale for the training subjects. The MRI regionality scale may correspond to regions specified in a neuroanatomical atlas and may be normalized to reduce variability between subjects and account for variance due to head size. The dataset creation engine 115 may use the MRI regionality scale to construct a planar filtered network graph. The planar filtered network graph may include nodes corresponding to brain regions and edges corresponding to the relationships between brain regions. The planar filtered network graph may prioritize the inclusion of brain regions with higher correlations. Next, the dataset creation engine 115 may use the planar filtered network graph to perform a multilevel clustering analysis. Next, the dataset creation engine 115 may perform a multiscale hub analysis to identify modules.

[0081] In such embodiments, the dataset creation engine 115 may generate composite values ​​for the identified modules using MRI regionality measures for these modules. In some embodiments, a first principal component may be calculated for each measurement type (e.g., volume, surface area, cortical thickness, etc.) across all brain regions included in the module. This first principal component may then be associated with the module. As can be understood, principal component values ​​may be generated for multiple types of measurements (e.g., each of volume, surface area, cortical thickness, etc.), and the results for these types of measurements may be associated with the module. As can be understood, the disclosed embodiments are not limited to using principal component analysis to generate values ​​for modules. Regional values ​​for identified hubs and modules may be used as input data for training a predictive model, consistent with the disclosed embodiments.

[0082] In some embodiments, training data may be or include measured brain tau data for a subject, consistent with the disclosed embodiments. In some embodiments, the measured brain tau data may be continuous value data. As described herein, such continuous value brain tau data may be or include an assessment of brain tau load based on image data for a subject. Such image data may be or include tau-PET image data showing tracer incorporation, or image data acquired using another preferred image modality. The tau-PET tracer may be any preferred tracer, including but not limited to MK6240, RO948, PI2620, Flortaucipyl, etc. Other examples of preferred tau-PET tracers include arylquinoline derivatives (e.g., [18F]THK5317 and [18F]THK5351), or phenyl / pyridinyl-butadienyl-benzothiazone / benzothiazolium (PBB) derivatives such as [11C]PBB3. In some embodiments, the tau-PET tracer is [18F]-RO-948, [18F]-PI-2620, [18F]-JNJ-311, or [18F]-GTP1. In some embodiments, the brain tau data may be discrete data as described herein.

[0083] Similarly, when predicting the brain amyloid-beta state, the measured brain amyloid-beta data may be continuous data. As described herein, such continuous brain amyloid-beta data may be, or include, amyloid-PET image data showing tracer uptake, or image data acquired using another preferred imaging modality. The Aβ PET tracer may be any preferred tracer, including but not limited to 18F-NAV4694, florbetaben, florbetapyl, flutemetamol, etc. In some embodiments, the brain amyloid-beta data may be discrete data, as described herein.

[0084] In some embodiments, brain tau or amyloid-beta data may include continuous-value tau or amyloid-beta data. In some embodiments, continuous-value tau or amyloid-beta data may be metric data such as intensity data, detection amount data (e.g., the number of pixels that satisfy the detection criterion), or other metric data extracted from image data for the subject. For example, a component of platform 100 (e.g., ETL engine 110) may determine tau-PET SUVR values, amyloid-PET SUVR values, and / or amyloid-PET centroid values ​​from PET image data according to a preferred method. Alternatively, such a component of platform 100 may receive tau-PET SUVR values, amyloid-PET SUVR values, and / or amyloid-PET centroid values ​​from another component of platform 100 (e.g., record 101) or another system.

[0085] In some embodiments, the data in question may be or include repeated measures of tau or amyloid-beta data over time. For example, repeated measures may include Braak stage measures over time for the subject. Repeated measures may be implicitly or explicitly associated with elapsed time from baseline. For example, repeated measures may be a vector (or matrix) of values, where each position in the vector (or column in the matrix) is implicitly associated with elapsed time. As an additional example, repeated measures may be a set of tuples, each tuple containing an elapsed time and a set of predicted levels for that elapsed time.

[0086] In some embodiments, training data may relate to subjects meeting cognitive impairment criteria. Cognitive impairment criteria may specify that a subject has a diagnosis of neurological disorder, dysfunction, or injury (e.g., a diagnosis of AD, MCI, dementia, etc.) or has certain signs (e.g., positive amyloid on biomarker tests or PET scans, etc.; biomarker scores such as levels, scores, or ratios of p-tau 181, p-tau 217, p-tau 231, Aβ42, or Aβ40 in plasma, serum, or cerebrospinal fluid, etc.). In some cases, cognitive impairment criteria may be inclusion criteria for clinical trials. In some embodiments, at least a portion of the subjects may not meet cognitive impairment criteria. For example, such subjects may lack a diagnosis of neurological disorder, dysfunction, or injury or documented signs. In some embodiments, components of platform 100 may obtain at least a portion of the training data from a database (e.g., record 201, etc.) or another system. In some embodiments, components of platform 100 may generate at least a portion of the training data.

[0087] In step 220 of process 200, components of platform 100 (e.g., training engine 120) can train a predictive model for predicting brain tau states for a subject, consistent with the disclosed embodiments. In some embodiments, the training engine 120 may create a predictive model and subsequently store the predictive model in the model storage device 103. In some embodiments, the training engine 120 may retrieve the predictive model from the model storage device 103 or another database or system and subsequently refine the model.

[0088] In some embodiments, the training engine 120 may acquire hyperparameters for training a predictive model. The specific hyperparameters acquired may depend on the type of predictive model, and the disclosed embodiments are not limited to any particular set of hyperparameters. For example, a neural network model may have hyperparameters governing the arrangement and configuration of layers, batch size, dropout, etc. As an additional example, a gradient boosting model may have hyperparameters governing the learning rate, number of trees, bagging rate, tree depth, etc.

[0089] In some embodiments, a user may interact with a user device 199 to provide hyperparameters to the training engine 120. In some embodiments, the training engine 120 may receive or retrieve hyperparameters from another component of the platform 100. In some embodiments, the training engine 120 may generate suitable hyperparameters. For example, the training engine 120 may be configured to perform an iterative or adaptive search of a given hyperparameter space (e.g., through training a predictive model, evaluating the model's performance, and updating hyperparameters selected based on the model's performance).

[0090] In some embodiments, the training engine 120 may train a predictive model using hyperparameters and training data acquired in process 210. The disclosed embodiments are not limited to any specific code or instructions for training the model. In some embodiments, the training engine 120 may be configured to evaluate the performance of multiple model designs using the same training dataset (e.g., Monte Carlo logistic lasso model, Bayesian logistic elastic network model, Bayesian ordered logistic model, Bayesian linear lasso regression (BLLR), tree-based models (e.g., regularized random forest model), stochastic gradient boosting machine model, etc.). As can be understood, BLLR can reduce the complexity of the model by relaxing the weights of the predictor variables and shrinking less important variables toward zero. SGB can combine predictions from multiple decision trees to produce the final prediction and can automatically model the inherent nonlinearity and interactions between predictors without prior assumptions about the distribution or specific mathematical forms of relationships between predictors and outcomes.

[0091] The performance of a model design can be determined using k-fold cross-validation. In some embodiments, the training engine 120 may be configured to evaluate the performance of the best-performing model design by splitting the training dataset into training and validation subsets. The training engine 120 may evaluate the model design by performing k-fold cross-validation using the training subset. The training engine 120 may select a model design and evaluate its performance using the validation subset.

[0092] For example, if the training engine 120 uses the R statistics package and the predictive model is a gradient boosting model, the following code could be used to train the model: library(gbm) x = TrainingData[, xvar] y = TrainingData[, yvar] train.data=cbind(y,x) set.seed(263) gbm.fit<-gbm(y~.,data=train.data, verbose=FALSE,distribution=“gaussian”, shrinkage=0.01, interaction.depth=3, n.minobsinnode=100, n.trees=1000, cv.folds=5, bag.fraction=0.5, n.cores=1 )

[0093] The training engine 120 can execute this code to determine a predictive gradient boosting model using the training data and given hyperparameter values.

[0094] In some embodiments, the training engine 120 may be configured to evaluate the performance of multiple model designs using the same training dataset. The performance of the model designs may be determined using k-fold cross-validation. In some embodiments, the training engine 120 may be configured to evaluate the performance of the best-performing model design by splitting the training dataset into training and validation subsets. The training engine 120 may evaluate the model designs by performing k-fold cross-validation using the training subset. The training engine 120 may select a model design and evaluate its performance using the validation subset. As can be understood, the evaluation may depend on the outputs predicted by the model. For example, discrete output values ​​(e.g., tau or amyloid-beta positive status) may be evaluated based on AUROC, etc. (e.g., the training engine 120 may select a predictive model that yields a higher AUROC than a training model that yields a lower AUROC). As an additional example, continuous value outputs (e.g., tau or amyloid-beta SUVR levels) may be evaluated based on RMSE, for example (e.g., training engine 120 may select a predictive model that yields a lower RMSE rather than a training model that yields a higher RMSE).

[0095] In step 230 of process 200, components of platform 100 (e.g., ETL engine 110, dataset creation engine 115, prediction engine 130, etc.) may acquire target data for individual targets. In some embodiments, individual targets may satisfy cognitive impairment conditions. For example, individual targets and targets from which training data has been acquired may satisfy the same cognitive impairment conditions. In some embodiments, individual targets may satisfy similar or equivalent cognitive impairment conditions (e.g., a training target may have had a diagnosis of AD, while an individual target may have clinical findings suggestive of AD). In some embodiments, components of platform 100 may acquire at least a portion of the individual target data from a database (e.g., record 101, etc.) or another system. For example, platform 100 may be configured to accept prediction requests from other systems. In some embodiments, components of platform 100 may generate at least a portion of the individual target data.

[0096] In some embodiments, the target data for individual targets (e.g., predict target data) may be the same as the target data included in the training data (e.g., training data). For example, if the training data includes biomarker data, the predict target data may include the same biomarker data. In another example, if the training data includes a particular combination of demographic data and biomarker data, the predict target data may include the same demographic data and biomarker data. In a further example, if the training data includes biomarker data, demographic data, and image data, the predict target data may include the same biomarker data, demographic data, and image data.

[0097] As can be understood, obtaining the data to be predicted may involve reformatting or rearranging the data to match the format or arrangement of the training data. Similarly, obtaining the data to be predicted may involve handling missing or incorrect values ​​in the training data. Furthermore, if obtaining the training data involves generating certain values ​​(e.g., generating composite values ​​for a module), obtaining the data to be predicted may similarly involve generating these values.

[0098] In step 240 of process 200, components of platform 100 (e.g., prediction engine 130) can predict the presence or absence of cerebral tau or amyloid-beta status for a subject, consistent with the disclosed embodiments. In some embodiments, the prediction engine 130 may input data to be predicted into a trained prediction model. The output of the trained prediction model may be a prediction of cerebral tau or amyloid-beta status for a subject. For example, the prediction model may predict the absence or presence of cerebral tau or amyloid-beta status in the entire brain. In addition or alternatively, the prediction model may predict the absence or presence of cerebral tau or amyloid-beta status in a brain region or set of brain regions corresponding to Braak stages, or in a part of the brain such as an anatomical region (e.g., the medial temporal lobe, frontal lobe, or parietal lobe of the brain).

[0099] In some embodiments, the prediction of cerebral tau or amyloid-beta status for a subject may be a predicted cerebral tau or amyloid-beta status level for the whole brain. In some embodiments, the prediction of cerebral tau or amyloid-beta status for a subject may be a predicted cerebral tau or amyloid-beta status level for a part of the brain, such as a brain region or set of brain regions corresponding to a Braak stage (as described herein), or other anatomical parts. For example, a trained predictive model may predict a collective amount of tau, such as tau concentrates or tau depositions, or a collective amount of amyloid-beta depositions, in a region encompassed by a given Braak stage. The prediction may be a continuous value (e.g., SUVR level), a discrete value (e.g., high, low), a Boolean value (e.g., present, absent), or a categorical value, and the prediction may be for one or more parts of the brain.

[0100] As can be understood, the type of prediction can depend on how the model is trained. For example, if the training data includes class-value brain tau or amyloid-beta data, the output of the predictive model could be the predicted class, the predicted class(s), or the predicted class likelihood. In another example, if the training data includes tau or amyloid-beta level-metric data (e.g., tau-PET SUVR level data, amyloid-beta-PET SUVR, or centroid level data), the output of the predictive model could be tau or amyloid-beta level-metric data (e.g., tau-PET SUVR level data, amyloid-beta-PET SUVR, or centroid level data).

[0101] In accordance with the disclosed embodiments, platform 100 may provide predicted brain tau or amyloid-beta states. Platform 100 may provide the predicted states to a user of platform 100 (for example, by providing predicted output classes, class probabilities, or brain tau or amyloid-beta levels to a user device 199 for display), or store the predicted brain tau or amyloid-beta states in a component of platform 199, or provide the predicted brain tau or amyloid-beta states to another system (for example, the system that provided the prediction request).

[0102] In some embodiments, the output of the trained predictive model in step 240 may be a sequence of predicted brain tau or amyloid-beta levels (e.g., predicted brain tau or amyloid-beta progression data). The predicted brain tau or amyloid-beta progression data may be implicitly or explicitly associated with elapsed time from baseline. For example, the output may be a vector (or matrix) of values, where each position in the vector (or column in the matrix) is implicitly associated with elapsed time. As an additional example, the predicted brain tau or amyloid-beta progression data may be a set of tuples, each tuple containing an elapsed time and a set of predicted levels for that elapsed time. It will be understood that the disclosed embodiments may involve predicting repeated measures in an individual (e.g., predicting multiple measures for the same individual). For example, the predicted progression data may include predictions of Braak disease stages over time for a particular subject.

[0103] In accordance with the disclosed embodiments, platform 100 may provide predicted brain tau or amyloid-beta state data. Platform 100 may provide the predicted brain tau or amyloid-beta state data to a user of platform 100 (for example, by providing the predicted brain tau or amyloid-beta state to a user device 199 for display), store the predicted brain tau or amyloid-beta state data in a component of platform 100, or provide the brain tau or amyloid-beta state data to another system (for example, the system that provided the prediction request).

[0104] As can be understood, predicted brain tau or amyloid-beta status data can be manually, semi-automatically, or automatically evaluated as an indicator of the progression of neurological disease, dysfunction, or injury. In some cases, for example, based on brain tau or amyloid-beta status, a subject may be diagnosed with mild cognitive impairment. Similarly, predicted brain tau or amyloid-beta status may provide an indicator that a subject will progress to Alzheimer's disease. Platform 100 may be configured to automatically evaluate predicted brain tau or amyloid-beta status data (e.g., using baseline data, predicted tau or amyloid-beta levels, and biomarker information, demographic information, etc.) and provide indicators of such brain tau or amyloid-beta status data.

[0105] In some embodiments, subjects (or patients) described as having mild Alzheimer's disease dementia or mild AD dementia may meet the core clinical criteria of the National Institute of Aging-Alzheimer's Association (NIA-AA) for possible Alzheimer's disease dementia, as described in McKhann, GM et al., “The diagnosis of dementia due to Alzheimer's disease: Recommendations from the National Institute on Aging-Alzheimer's Association workgroups on diagnostic guidelines for Alzheimer's disease.” Alzheimer Dement. 2011;7:263-9. Subjects with a CDR score of 0.5–1.0 and a Memory Box score of 0.5 or higher at screening and baseline, as well as subjects showing changes in scores on the Revised Wechsler Scale of Memory – Logical Memory Subscale II (WMS-R LM II), are also included herein.

[0106] In some embodiments, subjects (or patients) described as having MCI (intermediate possibility) due to AD may be identified according to the NIA-AA Core Clinical Criteria for mild cognitive impairment (intermediate possibility) due to Alzheimer's disease (see McKhann (above)). For example, a subject may have evidence of a brain amyloid pathology that is symptomatic but not dementia, making the subject less heterogeneous in cognitive and functional decline and more similar to subjects with mild Alzheimer's dementia, as measured by the ADCOMS composite clinical score as defined herein. Subjects with a CDR score of 0.5 and a memory box score of 0.5 or higher at screening and baseline are also included. Furthermore, subjects reporting a history of subjective memory decline with progressive onset and slow progression over the last year prior to screening, as confirmed by the informant, are also included herein. Memory decline and / or episodic memory impairment may be assessed in subjects by changes in scores on the Revised Wechsler Scale for Memory – Logical Memory Subscale II (WMS-R LM II).

[0107] In some embodiments, subjects (or patients) described as having pre-symptomatic AD, pre-AD, or being asymptomatic for AD are cognitively normal individuals with moderate or elevated levels of amyloid in the brain, which may be identified by an asymptomatic stage with or without memory complaints and episodic memory and executive function deficits. Cognitively normal may include individuals with a CDR of 0, or individuals with scores within the normal range for cognitive measures as described herein. Pre-symptomatic AD occurs before significant irreversible neurodegeneration and cognitive impairment and is typically characterized by the appearance of in vivo molecular biomarkers of AD and the absence of clinical symptoms. Pre-symptomatic AD biomarkers that may suggest the future onset of Alzheimer's disease include moderate or elevated levels of amyloid in the brain as measured by amyloid PET (e.g., centiloid scale of approximately 20–40, e.g., centiloid scale of approximately 20–32), fluorodeoxyglucose (FDG) PET, or tau positron emission tomography (PET), cerebrospinal fluid (CSF) levels of Aβ1-42 and / or Aβ1-42 / 1-40 ratio, CSF levels of total tau, CSF levels of microtubule-binding region (MTBR)-tau, CSF levels of neurogranin, CSF levels of neurofilamentous light chains (NfL), and serum or blood Examples of blood biomarkers measured in plasma include, but are not limited to, one or more of the following: Aβ1-42 levels, the ratio of two forms of amyloid-β peptide (Aβ1-42 / 1-40 ratio, e.g., a ratio of approximately 0.092 to 0.094 or less than approximately 0.092), plasma levels of total plasma tau (T-tau), phosphorylated tau (P-tau) isoforms (including tau phosphorylated at 181 (P-tau181), tau phosphorylated at 217 (P-tau217), and tau phosphorylated at 231 (P-tau231)), glial fibrillary acidic proteins (GFAP), and neurofilamentous light chains (NF1)). For example, subjects treated with elenbecestat (E2609), a β-site amyloid precursor protein cleavage enzyme (BACE) inhibitor, with an amyloid baseline positron emission tomography (PET) standard uptake ratio (SUVr value) of 1.4 to 1.9, were found to exhibit the greatest slowing of cognitive decline during treatment.See Lynch, SY et al. “Elenbecestat, a BACE inhibitor: results from a Phase 2 study in subjects with mild cognitive impairment and mild-to-moderate dementia due to Alzheimer's disease.” Poster P4-389, Alzheimer's Association International Conference, July 22-26, 2018, Chicago, IL, USA. Similarly, subjects with baseline florbetapir amyloid PET SUVr levels less than 1.2 did not show sufficient cognitive decline to be detectable, while subjects with SUVr levels greater than 1.6 had amyloid levels that had reached saturation levels and appeared to correlate with a plateau effect where treatment did not result in changes in cognitive scales. See Dhadda, S. et al., “Baseline florbetapir amyloid PET standard update value ratio (SUVr) can predict clinical progression in prodromal Alzheimer's disease (pAD).” Poster P4-291, Alzheimer's Association International Conference, July 22-26, 2018, Chicago, IL, USA.

[0108] In some embodiments, subjects (or patients) described as having early-stage AD may exhibit AD severity ranging from mild cognitive impairment (intermediate possibility) to mild Alzheimer's disease dementia. Subjects having early-stage AD include subjects with mild Alzheimer's disease dementia as defined herein, and subjects with mild cognitive impairment (MCI) (intermediate possibility) as defined herein. In some embodiments, subjects having early-stage AD have an MMSE score of 22–30 and a Clinical Dementia Scale (CDR) global range of 0.5–1.0. Other methods for detecting early-stage AD disease may utilize the tests and assays specified below, including the National Institute of Aging-Alzheimer's Association (NIA-AA) core clinical criteria for possible Alzheimer's disease dementia, as described in McKhann, GM et al., “The diagnosis of dementia due to Alzheimer's disease: Recommendations from the National Institute on Aging-Alzheimer's Association workgroups on diagnostic guidelines for Alzheimer's disease.” Alzheimer Dement. 2011;7:263-9. Other methods include CDR-SB, ADCOMS, MMSE, ADAS-Cog, ADAS MCI-ADL, modified iADRS, WMS-IV LMI, and WMS-IV LMII. In some embodiments, subjects with early-stage AD have evidence of elevated amyloid in the brain or a positive amyloid load. In some embodiments, elevated amyloid levels or a positive amyloid load in the brain are indicated and / or confirmed by PET evaluation. In some embodiments, elevated amyloid levels or a positive amyloid load in the brain are indicated and / or confirmed by CSF evaluation of markers such as Aβ1-42 (e.g., soluble CSF biomarker analysis).In some embodiments, elevated amyloid or positive amyloid load in the brain is indicated and / or confirmed by measuring the level of p-tau 181. In some embodiments, elevated amyloid or positive amyloid load in the brain is indicated and / or confirmed by MRI. In some embodiments, elevated amyloid or positive amyloid load in the brain is indicated by retinal amyloid deposition. In some embodiments, two or more assessment methods are used.

[0109] In some embodiments, the control subject (or patient), untreated AD subject (or patient), or untreated control subject (or patient) is either untreated or treated for Alzheimer's disease. In some embodiments, the control subject has Alzheimer's disease. In some embodiments, the control subject has early Alzheimer's disease or pre-Alzheimer's disease. In some embodiments, the control subject has Alzheimer's disease and is not treated with anti-Aβ protofibril antibodies.

[0110] As can be understood, a trained predictive model consistent with the disclosed embodiments may be used to screen or select patients for inclusion in a clinical trial. Brain tau or amyloid-beta status data may be predicted for a candidate patient using baseline data obtained for that candidate patient. A candidate patient may be included in the trial if their predicted brain tau or amyloid-beta status meets the selection criteria. In various embodiments, the selection criteria may depend on predicted brain tau or amyloid-beta levels, fulfillment of certain demographic factors, biomarker measurements, or another preferred measure. For example, a patient may be included in a clinical trial if their predicted brain tau or amyloid-beta levels are above a threshold or within a specified range. Such a patient may have a greater need for treatment (otherwise, they may experience greater cognitive decline).

[0111] As can be understood, a trained predictive model consistent with the disclosed embodiments may be used in the design of a clinical trial. As described herein, the clinical trial population may be enriched with patients who are likely to benefit from the treatment. In particular, the trained predictive model may be used to screen or select patients for enrollment in a clinical trial. Selected patients may be those who are likely to have at least a minimum threshold amount of cerebral tau or amyloid-beta deposition, or at most a maximum amount of cerebral tau or amyloid-beta deposition. Such patients may be suitable targets for treatments suited to removing or reducing tau (or amyloid-beta) deposition. As can be understood, the therapeutic effect, trial size, and power of the trial may be related. By selecting patients who are likely to be suitable therapeutic targets, fewer patients may be enrolled, or the power of the trial may be increased, or the scale of the detectable therapeutic effect may be reduced, or any combination thereof.

[0112] Furthermore, the benefits of treatment for such patients may be more readily apparent than those for less affected patients. Screening or selecting patients using trained predictive models because the effects of treatment are more apparent (e.g., greater effects) may enable improved clinical trial designs: the number of patients enrolled may be reduced, the minimum detectable therapeutic effect may be increased, the power of the trial may be increased, or some combination of the above may be achieved.

[0113] As can be understood, predicted cerebral tau or amyloid-beta status may be used to evaluate the effectiveness of treatments for neurological disorders, dysfunctions, or injuries in clinical trials or in real-world clinical settings. Clinical trials may include multiple participants. Participants may be screened for compliance with demographic factors or biomarker data, and baseline data may be obtained for each participant. Participants may be assigned to either the control group or the treatment group of the trial.

[0114] Using a trained predictive model, cerebral tau or amyloid-beta status may be predicted for one or more participants in the treatment group of a clinical trial and may be used as a covariate when analyzing the results of the clinical trial.

[0115] For example, clinical trials may relate to the treatment of Alzheimer's disease. A trained predictive model may be used to predict cerebral tau or amyloid-beta status for at least some patients assigned to the treatment group in a clinical trial. The predicted cerebral tau or amyloid-beta status may be used as a covariate in determining the effectiveness of Alzheimer's disease treatment.

[0116] Figure 3 shows a process 300 for treating cerebral tau or amyloid-beta deposition consistent with embodiments of the present disclosure. In some embodiments, the process 300 is described as being carried out using platform 100. However, the process 300 may also be carried out, at least in part, using a different computing system. The process 300 may be used to predict cerebral tau or cerebral amyloid-beta status. When predicting cerebral amyloid-beta status, patient medical record data may include or be derived from Aβ-SUVR levels or centroid levels. In some such embodiments, the predictive model may output predicted Aβ-SUVR levels or centroid levels, or Aβ categories (e.g., Aβ+, Aβ-, etc.). When predicting cerebral tau status, patient medical record data may include or be derived from tau-PET SUVR levels. In some such embodiments, the predictive model may output predicted tau-SUVR levels, or tau categories (e.g., tau-positive, tau-negative, etc.). Examples of the development of predictive brain tau and brain amyloid-beta models are provided herein.

[0117] Process 300 may include a step 310 to acquire patient medical data (e.g., reference data for the patient). In some embodiments, the reference data may be acquired from the patient's medical record data. Components of platform 100 (e.g., ETL engine 110, dataset creation engine 115, etc.) may acquire the reference data in accordance with the disclosed embodiments. In some embodiments, the reference data may include biomarker data as described herein. In some embodiments, the reference data may include biomarker data and at least one of cognitive measurement data, genomic data, image data, demographic data, etc.

[0118] Process 300 may include a step 320 for predicting a brain tau or amyloid-beta state. Step 320 may include predicting a brain tau or amyloid-beta state for a patient by applying at least a portion of the acquired reference data (e.g., acquired in step 310) to a predictive model. In some embodiments, the predictive model may be trained to predict a brain tau or amyloid-beta state using a training dataset. The training dataset may include training examples. The training examples may include reference data and label information as described herein. In some embodiments, the reference data used to train the predictive model may include the same type of information as the reference data acquired in step 310. For example, if the predictive model is trained using reference data that includes p-tau 217R values ​​and demographic information, the reference data acquired in step 310 may include p-tau 217R values ​​and the same demographic information.

[0119] In some embodiments, at least a portion of the subjects being trained may meet disability criteria as described herein. It will be understood that the accuracy of such predictions can be improved by predicting the level of cerebral tau or amyloid-beta deposition in patients who meet disability criteria (e.g., those who are therefore more likely to have elevated levels of cerebral tau or amyloid-beta deposition).

[0120] In some embodiments, step 320 may be performed as described with respect to process 200. In addition or alternatively, step 320 may be performed using other suitable predictive models. As described herein, the predicted cerebral tau or amyloid-beta state (e.g., predicted in step 320) may indicate elevated levels of cerebral tau or amyloid-beta deposition. The prediction may relate to the whole brain, a brain region, a group of brain regions, or other parts of the brain. Such parts may be associated with a particular Braak stage or major cortical lobes of the brain (e.g., cortex, parietal, occipital, lateral temporal, medial temporal) and all cortical gray matter (e.g., all cortical gray matter or WCGM). Thus, the predicted cerebral tau or amyloid-beta state may enable the treatment and / or management of neurological disorders, disorders, or injuries. For example, the predicted cerebral tau or amyloid-beta state may include indicators that the patient requests further examination (e.g., cognitive assessment tests, additional biomarker tests, imaging) or monitoring (e.g., follow-up by a clinician). In some embodiments, the predicted cerebral tau or amyloid-beta state may include data for determining or predicting the progression of neurological disease, impairment, or injury in a patient.

[0121] In some embodiments, process 300 may proceed to step 330 based on the predicted cerebral tau or amyloid-beta state in step 320. For example, the predicted cerebral tau or amyloid-beta state may indicate that the patient has elevated levels of tau deposition in one or more brain regions. In response, process 300 may proceed to step 330. Otherwise (for example, if the patient is predicted to have low levels of cerebral tau or amyloid-beta deposition), process 300 may terminate.

[0122] Process 300 may be accompanied by an optional step 330 to confirm the predicted cerebral tau or amyloid-beta state. For example, the predicted cerebral tau or amyloid-beta state may be confirmed using a suitable imaging technique (e.g., a PET scan using a suitable tau radiotracer, or another suitable confirmation method). As can be understood, performing such imaging in response to a prediction of elevated levels of tau deposition may allow for a more efficient use of insufficient clinical resources.

[0123] Process 300 may include an optional step 340 that provides treatment. Step 340 may be based on a predicted brain tau or amyloid-beta state (e.g., step 320). For example, based on the predicted brain tau or amyloid-beta state, treatment may be provided to the patient, or instructions or recommendations for such treatment may be provided by platform 100 (e.g., to the user of platform 100). In some embodiments, treatment may be provided based on other factors in addition to the brain tau or amyloid-beta state, such as factors in the patient's medical history.

[0124] In some embodiments, providing treatment may involve administering a therapeutic agent to a patient. For example, a predicted cerebral tau or amyloid-beta state for a patient may contribute to determining which therapeutic agent to administer to the patient. It will be understood that different therapeutic agents may be administered to a patient based on the predicted cerebral tau or amyloid-beta state. A therapeutic agent may refer to any substance, treatment, or compound that can treat, cure, alleviate, or prevent a disease or condition. For example, a therapeutic agent may be a biologically active compound that can treat dementia-related conditions. In some embodiments, the therapeutic agent may be an anti-amyloid-beta protofibril antibody. For example, the antibody may be lecanemab. The sequence and other information relating to lecanemab can be found in PCT / U.S. Patent Application Publication No. 2022 / 073576, which is incorporated by reference as a whole. Exemplary dosing regimens for recanemab are disclosed in International Application No. PCT / U.S. Patent Application Publication 2024 / 033125, which is also incorporated herein by reference in whole. As another example, the therapeutic agent may be an anti-tau antibody such as E2814. Sequence information and exemplary dosing regimens for E2814, including as a combination therapy with recanemab, are disclosed in International Application No. PCT / IB2021 / 000937 and PCT / U.S. Patent Application Publication 2022 / 079509, each of which is incorporated herein by reference in whole.

[0125] In some embodiments, providing a treatment may involve monitoring the therapeutic effect over time. For example, such monitoring may include determining whether the treatment (e.g., recanemab) reduces predicted brain tau or amyloid-beta levels over time, or reduces the rate of increase in predicted brain tau or amyloid-beta levels over time.

[0126] In some embodiments, the steps of process 300 may be repeated periodically. For example, steps 310 and 320 may be periodically predicted to monitor the patient's cerebral tau or amyloid-beta state. For example, based on the cerebral tau or amyloid-beta state, treatment may not be initially prescribed to the patient. The patient may be monitored by repeating the steps of process 300. As the patient's condition progresses, the progression may be indicated in the patient's predicted cerebral tau or amyloid-beta state. Once the patient meets the treatment conditions, treatment may be offered to the patient. Similarly, once treatment has been offered to the patient, the steps of process 300 may be periodically repeated to monitor the effectiveness of the treatment (e.g., a reduction in the degree of neurofibrillary ensembles, or another indicator described herein). As can be understood, the disclosed embodiments may enable such monitoring strategies that may not be feasible with existing evaluation tools such as tau-PET imaging.

[0127] It will be understood that the disclosed embodiments provide improvements to predicting cerebral tau or amyloid-beta status and to providing treatments for elevated levels of cerebral tau or amyloid-beta deposition (including improved selection for patients likely to respond to treatment). As described herein, the disclosed embodiments may be provided as a screening tool (for example, for clinicians, etc.) to improve the ability to identify individuals with elevated levels of cerebral tau or amyloid-beta deposition, thereby providing an entry point for triage. The disclosed embodiments may assist in the early identification of neurological disorders, disabilities, or injuries and may alert clinicians to potential undetected risks in patients. The disclosed embodiments may also assist in the design and conduct of clinical trials of novel compounds intended to combat tau (or Aβ) deposition in patients' brains. The disclosed embodiments may also provide improvements in the monitoring and management of such neurological disorders, disabilities, or injuries by predicting data that may indicate that a patient should seek follow-up (e.g., with a specialist) and / or further examination. [Examples]

[0128] Several studies have been conducted on training and using predictive models consistent with the disclosed embodiments. These studies relate to predicting brain tau or amyloid-beta states using the data described herein.

[0129] The first study involved generating a predictive model for detecting tau neurofibrillary hierarchies along Braak disease stages. Neurofibrillary hierarchies may also be pathological features of Alzheimer's disease, and the disclosed embodiments may support the development of non-invasive methods for detecting tau hierarchies (e.g., to enable efficient screening for patients in AD clinical trials). The first study involved generating a predictive model for detecting tau hierarchies along Braak disease stages using demographics, ApoE4 status, and one or more of the following: cognitive function assessments, plasma p-tau 181, and brain region MRI measures (e.g., volume, area, and cortical thickness).

[0130] Figure 4 illustrates a diagram 400 of the progression of Alzheimer's disease along the Braak stages, consistent with the disclosed embodiments. Diagram 400 shows the various brain regions encompassed by Braak stages 1, 2, 3, 4, 5, and 6.

[0131] Data in the first study included baseline regional MK6240 tau-PET SUVR data for amyloid-positive patients from clinical trials. The tau-positive threshold for each Braak stage was defined by the one-sided upper 99% confidence interval of a subset of 70 subjects with amyloid-PET centiloid < 30 (representing negative controls). Cognitive function assessments included ADAS-COG-14, CDR-SB, MMSE, and their subscores. Plasma p-tau 181 was measured using the Simoa® assay. Subjects underwent 3.0 Tesla (T) structural MRI at baseline. Brain MRI data (volume, area, and cortical thickness) were generated for various brain regions of interest using the Desikan-Killiany atlas (Desikan et al., 2006), resulting in 207 regional scales. Cortical thickness values ​​are expressed in millimeters (mm). Volume (mm) 3 ) and area (mm 2 ) was normalized (divided) by intracranial volume to reduce inter-subject variability.

[0132] In the first study, to reduce the dimensionality of the data, structural brain network (SBN) modules and hub regions were derived using data from other clinical trials via the MEGENA algorithm. Such SBN modules and hub regions may represent clusters and regions with the strongest spatial correlation to adjacent regions. Figures 5A–5C illustrate exemplary SBN hubs and modules consistent with the disclosed embodiments. Figure 5A shows a module representing inferior parietal cortex thickness, and Figure 5B shows a module representing precentral cortex thickness. Figure 5C shows brain hubs among the top predictors for distinguishing Braak3–4 and 5–6.

[0133] To detect tau positivity in Braak disease stages 3–6, signatures were derived via stochastic gradient boosting (e.g., an ensemble tree-based machine learning algorithm). To further distinguish tau positivity in Braak disease stages 3–4 and 5–6, signatures were derived via Bayesian ordered logistic regression with Student's t prior distribution. Features for deriving signatures were derived from data types including clinical cognitive function assessments, plasma p-tau 181, MRI-based SBN and hubs, amyloid PET centroid, demographic data (e.g., age, sex, BMI), and genomic data (e.g., ApoE4 status). Data were randomly split into 70–30 training-test splits. Predictive performance was evaluated via 10-fold cross-validation on the training set, followed by evaluation on the test set.

[0134] Figure 6 illustrates a table summarizing patient data consistent with the disclosed embodiments. Figure 6 shows tau-positive status at various Braak stages for subjects according to clinical diagnosis (e.g., mild cognitive impairment or mild Alzheimer's disease), sex (e.g., female or male), ApoE4 status (e.g., non-E4 or E4 carrier), age, BMI, and MMSE score.

[0135] Figure 7 illustrates the distribution of SUVR values ​​for subjects with different tau-positive states (e.g., tau-positive in Braak stage 0, Braak 1-2, Braak 3-4, or Braak 5-6) consistent with the disclosed embodiments.

[0136] Figure 8 illustrates Table 800, which shows the performance of tau-positive prediction in accordance with the disclosed embodiments. Signatures including clinical cognitive assessment, ApoE4 status, and demographics achieved an accuracy of 76% for detecting tau positivity in Braaks 3–6 and 60.8% for identifying Braaks 3–4 for Braaks 5–6. Adding plasma p-tau 181 improved accuracy to 82.2% (p<0.05) in Braaks 3–6 and to 67.6% (p<0.05) for Braaks 3–4 for Braaks 5–6. When MRI data was combined with plasma p-tau 181, the accuracy of detecting tau positivity in Braaks 3–4 for expansion to Braaks 5–6 improved from 66.8% to 74.6% (p<0.05).

[0137] Figures 9A–9C illustrate top predictors for detecting tau-positive subjects in Braak3–6 using a stochastic gradient boosting machine model consistent with the disclosed embodiments. Figure 9A illustrates the relative impact of features using clinical (e.g., cognitive function) data and plasma p-tau-181 data. Figure 9B illustrates the relative impact of features using plasma p-tau-181 and MRI data. Figure 9C illustrates the relative impact of features using clinical (e.g., cognitive function), plasma p-tau-181, and MRI data. In Figures 9A–9C, HV.L refers to hippocampal volume (left), VCIPCR refers to inferior parietal cortex thickness (right), VVEL refers to entorhinal cortex, VCIPCL refers to inferior parietal cortex thickness (left), VCPRCR refers to precentral cortex thickness (right), ADCDRL refers to delayed word recall, and ADCRL refers to word recall.

[0138] Figures 10A–10C illustrate individual conditional expectation profiles for several features that predict tau positivity, consistent with the disclosed embodiments. Figures 10A–10C illustrate individual conditional expectation (ICE) profiles for each target (gray) and the average target (black). The vertical axis represents the normalized probability of tau positivity after normalizing for the probability at the minimum value of the feature. Heterogeneity between targets in the ICE profiles may be due to strong interactions between predictors. A stochastic gradient boosting algorithm can account for these nonlinear relationships and interactions without prior assumptions.

[0139] Figures 11A–11C illustrate heatmaps (with heatmap intensity corresponding to the probability of tau positivity) showing interactions between several features that predict tau positivity, consistent with the disclosed embodiments. Figure 11A illustrates a heatmap showing the interaction between plasma p-tau 181 and delayed word recall. Figure 11B illustrates a heatmap showing the interaction between plasma p-tau 181 and MRI inferior parietal cortex thickness (right). Figure 11C illustrates a heatmap showing the interaction between MRI inferior parietal cortex thickness (right) and delayed word recall. It will be understood that the interaction predictive profiles demonstrate a strong dependence between several key features for predicting tau positivity. For example, among patients with relatively weak delayed word recall, patients with high plasma p-tau 181 are more likely to be tau positivity. Similarly, patients with higher plasma p-tau 181 are more likely to be tau positivity if their inferior parietal cortex thickness is thinner.

[0140] Figures 12A–12B illustrate plasma p-tau 181 as a function of tau-positive status, consistent with the disclosed embodiments. Figure 12A illustrates that plasma p-tau 181 increases monotonically with the presence of tau along Braak stages (e.g., Braak 0–1–2, Braak 3–4, and Braak 5–6). Figure 12B illustrates plasma p-tau 181 as a function of tau-positive status, grouped by tertiles of inferior parietal cortical thickness. Figure 12B illustrates that, when combined with inferior parietal cortical thickness, plasma p-tau 181 levels are more strongly distinguished between the presence of tau at Braak 3–4 versus the presence of tau at Braak 5–6 for most patients, with accuracy increasing from 67.6% to 74.6% (p<0.05).

[0141] Figures 13A–13C illustrate predictors for identifying tau-positive subjects in Braak stage 3–4 and Braak stage 5–6 using a Bayesian ordered logistic model consistent with the disclosed embodiments. Figure 13A illustrates predictors using clinical data and plasma p-tau-181 data. Figure 13B illustrates predictors using plasma p-tau-181 data and MRI data. Figure 13C illustrates predictors using clinical data, plasma p-tau-181 data, and MRI data. In Figures 13A–13C, VCIPCR refers to inferior parietal cortex thickness (right), VCPRCR refers to precentral cortex thickness (right), VVPCCL refers to posterior cingulate cortex volume (left), ADCDRL refers to delayed word recall, ADCCMD refers to direction, ADCOF refers to item naming, ADCCP refers to constructive actions, and CDR0101 refers to memory.

[0142] Figures 14A–14E illustrate the top MRI predictors. In Figures 14A–14E, VCIPCR refers to inferior parietal cortex thickness (right), and VCPRCR refers to precentral cortex thickness (right). Figure 14A illustrates the odds ratio for predicted tau positivity using plasma p-tau 181, ApoE4 count, and MRI data (e.g., VCIPCR, VCPRCR). Figure 14B illustrates VCIPCR as a function of tau positivity in Braak stages 0–2, 3–4, and 5–6. As reflected by the odds ratio < 1, it will be recognized that tau-positive subjects in Braak stages 3–4 and 5–6 had progressively thinner thickness in the inferior parietal cortex. Figure 14C illustrates VCPRCR as a function of tau positivity in Braak stages 0–2, 3–4, and 5–6. Figure 14D illustrates the VCPRCR as a function of tau positivity in Braak stages 0–2, 3–4, and 5–6, grouped by VCIPCR tertiles. It will be understood that precentral cortex (PCC) thickness alone did not distinguish tau-positive Braak stages. When stratified by inferior parietal cortex (IPC) thickness, there was an increase in PCC thickness levels in tau-positive Braak stages 3–6. Such a trend may be attributable to the increased level of inflammation / gliosis due to tau reaching Braak stages 5–6 in the PCC. That is, atrophy in the IPC due to tau positivity in Braak stages 3–4 may precede encephalitis. Next, the presence of tau in Braak stages 5–6 in the inflamed brain may manifest as greater thickness in Braak 6 regions such as the PCC, cuneiform region, and postcentral cortex. The first cluster of subjects, represented by an IPC thickness of 1.724–2.217, may indicate a more advanced disease stage, with more atrophy in the parietal region and more cases with tau in Braak stage 5–6. The third cluster of subjects, represented by an IPC thickness of 2.376–2.691, may indicate a less advanced disease stage, with less atrophy in the parietal region and more cases with tau positivity in Braak stage 0–2.Figure 14E illustrates subjects represented by IPC thicknesses of 2.217–2.376 (e.g., the second cluster of subjects as illustrated in Figure 14D). In Figures 14D–14E, Braak stages 5–6 may be thicker than Braak stages 0–2 or 3–4 within a given inferior parietal cortical thickness cluster, which may be due to tau-related inflammation / gliosis. In addition, within a given inferior parietal cortical thickness cluster, Braak stages 0–2 may be thinner because neurodegeneration has already occurred or may be occurring.

[0143] Figures 15A–15B show similar trends in other Braak stage 6 regions. Figure 15A illustrates postcentral cortical thickness as a function of tau positivity in Braak stages 0–2, Braak 3–4, and Braak 5–6, grouped by VCIPCR tertiles, consistent with the disclosed embodiments. Figure 15B illustrates cuneiform cortical thickness as a function of tau positivity in Braak stages 0–2, Braak 3–4, and Braak 5–6, grouped by VCIPCR tertiles, consistent with the disclosed embodiments.

[0144] Figure 16 illustrates a table, consistent with the disclosed embodiments, showing how adding amyloid-PET centiloid data improved tau-positive prediction in a first trial. In some embodiments, adding amyloid-PET data to plasma p-tau 181 and / or clinical evaluation and / or MRI may result in more accurate prediction of tau positivity.

[0145] Figures 17A–17B illustrate how amyloid PET levels may play a complementary role in predicting tau positivity. Figure 17A illustrates amyloid PET centroid levels as a function of tau positivity in Braak stages 0–2, 3–4, and 5–6, grouped by plasma p-tau 181 levels. In this study, tau-positive subjects with low p-tau 181 levels in Braak stages 3–4 and 5–6 had high amyloid levels. Similarly, tau-negative subjects with high p-tau 181 levels had low amyloid levels. Figure 17B illustrates an interaction profile demonstrating how tau positivity in subjects with high p-tau 181 levels may be highly dependent on their amyloid levels.

[0146] Therefore, the first study demonstrated the detection of tau deposition throughout the brain along Braak disease stages using non-invasive measures such as plasma p-tau 181, cognitive ability, and brain structural MRI data. These signatures may be used for patient screening in clinical trials. In some cases, these signatures may be followed by confirmation with other assessments, such as imaging (e.g., tau-PET), in a subset of patients. These signatures may also be used in previous clinical trials that did not include tau-PET assessments for exploratory evaluations of the association between tau and clinical progression, subgroup effects, etc. Furthermore, it will be understood that in some cases, adding amyloid-PET data may result in better predictive of tau positivity.

[0147] The second trial investigated the efficacy of plasma p-tau-217R in predicting continuous values ​​of regional tau levels (e.g., tau-PET SUVR measurements) and identifying individuals with varying levels of tau accumulation. The trial further sought to identify individuals with varying levels of tau accumulation within a cohort diagnosed with amyloid-beta-positive (Aβ+) mild cognitive impairment (MCI) or mild Alzheimer's disease (collectively referred to as early Alzheimer's disease). Plasma p-tau-217 and non-phosphorylated tau-217 concentrations were quantified via immunoprecipitation-mass spectrometry. The MK6240 predictive model for tau-PET SUVR was developed and validated using 60–40 random splits of a clinical trial cohort containing 242 amyloid-beta-positive early Alzheimer's disease individuals. A stochastic gradient boosting algorithm was used to construct a model for simultaneously predicting SUVR values ​​across various brain regions. Further analysis explored whether integrating additional predictors into the model (e.g., cognitive assessments, fluid biomarkers, structural MRI, and amyloid PET) could improve the predictive performance of p-tau217R. Model performance was cross-validated within the training set and further evaluated in the holdout test set.

[0148] In this second trial, the p-tau217R-based model predicted tau-PET uptake across various brain regions, and R 2The values ​​ranged from 0.49 to 0.65. The maximum predicted SUVR values ​​across these brain regions fell within the range of 1.87 to 2.3. The area under the receiver operating characteristic curve (AUROC) for detecting the presence of tau ranged from 84% to 95% across six Braak stages and cortical regions, maintaining performance at higher tau accumulation levels. In this study, integrating additional predictors did not improve the performance of ptau217R. Using ptau217R alone to predict tau-PET SUVR reduced the need for tau-PET scans by up to 65%, particularly in identifying low tau concentrations in cortical gray matter, while maintaining a 5% false-negative rate. As can be understood, the techniques applied in this study allow for pathological disease staging (i.e., tau load) using plasma ptau217R measurements, potentially reducing the need for tau-PET scans.

[0149] The data used to build and validate the predictive model included Aβ+ early-stage Alzheimer's disease subjects with objective cognitive memory impairment and a wide range of tau-PET SUVR values ​​in different brain regions. The training and validation datasets for building the predictive model were derived from 60–40 random splits of 242 early-stage Alzheimer's disease subjects from the Clarity AD clinical trial (a trial to confirm the safety and efficacy of recanemab in participants with early-stage Alzheimer's disease; NCT03887455), where both tau PET and plasma biomarker data were available.

[0150] Figure 18A provides a summary of key demographic and clinical features of the subjects included in the data used to construct and validate the predictive model, consistent with the disclosed embodiments. As is evident in Figure 18A, the distribution of tau-PET SUVR values ​​in WCGM and MTL spanned a broad spectrum from low to moderate to high tau levels. With the exception of age, none of the features showed significant differences between the training and test sets. In Figure 18A, SD refers to the standard deviation.

[0151] All participants underwent structural MRI scans. Brain MRI data, including volume, area, and cortical thickness across various brain regions of interest, were derived using the Desikan-Killiany atlas, yielding 207 regionality measures. An imaging pipeline was used for image processing. Cortical thickness values ​​were expressed in millimeters (mm), while volume (cubic millimeters) and area (square millimeters) were normalized by intracranial volume to mitigate inter-subject variability and adjust for differences in head size across regions. To streamline the analysis and focus on valid data, predictive models utilizing the MRI data were constructed specifically targeting 18 structural brain network (SBN) modules and 45 hub regionality measures, as identified in U.S. Provisional Patent Application No. 63 / 561,285 and incorporated herein by reference, thereby reducing redundancy and dimensionality within the dataset.

[0152] Tau-PET SUVR values ​​from brain regions defined using the Hammers atlas in PMOD were combined into volume-weighted average composite cortical regions, including the frontal, occipital, parietal, lateral temporal, and medial temporal regions, as well as six Braak stage regions and WCGM. SUVR values ​​from amyloid PET scans (89% florbetaben, 10% florbetabil, and 1% flutemtamol) were standardized to centroid for consistency and comparability.

[0153] In this study, plasma levels of Aβ42, Aβ40, p-tau-217, and unphosphorylated (np) tau-217 were measured using a high-throughput mass spectrometry-based analytical platform. Ratios Aβ42 / Aβ40 and p-tau-217 / np-tau-217 were calculated. Blood samples were collected in K2 EDTA tubes, processed into plasma, and stored at -80°C until analysis. Plasma aliquots were shipped on dry ice to private laboratories for processing and analysis. The p-tau-217 / np-tau-217 ratio (p-tau-217R) was used for subsequent analysis to normalize inter-individual differences.

[0154] Plasma levels of p-tau-181, neurofilamentous light chains (NfL), and glial fibrillary acidic proteins (GFAP) were evaluated using commercially available single-molecule array (Simoa®) assay kits: the Simoa® pTau-181 Advantage V2 assay kit, the Simoa® NF-Light assay kit, and the Simoa® GFAP discovery kit, respectively.

[0155] In this study, clinical assessments encompassing both cognitive and functional domains were also evaluated to confirm their feasibility as predictors in predictive models incorporating p-tau 217R. These assessments included MMSE, ADAS-Cog-14, CDR-SB, ADCS-ADL, and their respective subscores.

[0156] A model for predicting regional tau-PET SUVR for each patient was initially constructed from a training set. The predictive model was a stochastic gradient boosting (SGB) model, using plasma p-tau 217R, age, sex, and tau brain regions as input data. The predictive model outputted predictions for selected cortical regions (frontal, parietal, occipital, lateral temporal, and medial temporal) along with regions depicted by six Braak stages. Additional predictive models were developed that accepted additional input data, including clinical assessments, volumetric MRI (vMRI) scales, amyloid PET centinloid, and other plasma biomarkers (p-tau 181, Aβ42 / Aβ40, GFAP, and NfL).

[0157] To mitigate overfitting, the models underwent internal refinement through holdout datasets and cross-validation techniques. In this trial, up to 1000 decision trees were constructed using up to three-way interactions between predictors. To confirm the significance of individual predictors within the prognostic framework, the impact of such individual predictors on reducing the mean squared error (MSE) when used as root nodes for tree splitting in the SGB algorithm was evaluated. These measures were then normalized, resulting in predictor ranks and relative impacts uniformly adjusted from 0 to 100%.

[0158] In this trial, the predictive performance of the model was evaluated. The first evaluation was performed through 10 iterations of 10-fold cross-validation within the training set. The second evaluation was performed using the validation set. This evaluation used the coefficient of determination (R) for observed SUVR values ​​versus predicted SUVR values ​​across each brain region. 2 The study involved measuring key measures, including SUVR, MSE, and MAE. The correlation between observed SUVR values ​​and predicted SUVR values ​​was compared between models via Hittner's two-tailed test for dependency correlation. In addition, the potential additional benefits of incorporating ApoE4 allele counts into the models and using p-tau-217 concentrations instead of p-tau-217R were evaluated within this framework.

[0159] To estimate the upper limit of reliable prediction for tau-PET SUVR across different brain regions, a Richards 5-parameter generalized logistic model was fitted using data from subjects within a validation set. This model incorporates five key parameters: lower and upper asymptotes, slope representing the steepness of the curve, inflection points marking where the curvature transitions, and the degree of asymmetry indicating the shape of the sigmoid curve. The lower asymptote was set to 1, reflecting the expected background tau level. The upper asymptote parameter was estimated to describe the maximum SUVR prediction limit.

[0160] To assess the versatility of the tau-PET SUVR prediction model across varying rates of tau accumulation along the spectrum of early Alzheimer's disease, the model's ability to predict tau status (i.e., whether the predicted SUVR value is below or above a threshold) was evaluated for each of the cortical and Braak stage regions, using two SUVR thresholds: a threshold for tau positivity that varied for each region, and a higher threshold of 1.5. This evaluation was performed within a validation set using receiver operating characteristic (ROC) curves and relevant metrics such as area under the ROC curve (AUROC), sensitivity, and specificity. This performance evaluation was then repeated for each demographic subgroup (i.e., age, sex, and ApoE4 status) within the validation set to determine whether the model performed consistently across all subgroups. For simplicity, only the WCGM and MTL regions were considered for this evaluation.

[0161] Determining the SUVR threshold for tau positivity in each region involved analyzing data from a subset of 50 subjects within a training cohort. These subjects had amyloid PET centiloids less than 15, ranging from minimal amyloid accumulation to no amyloid accumulation. This subset was selected to represent a typical tau population with background (none / low) tau levels. The threshold for tau positivity was established as the upper 95% confidence limit, calculated using the formula: median + 1.645 × 1.4826 × MAD. Here, 1.645 represents the upper 5% limit of the standard normal distribution, MAD means the absolute deviation of the median, and 1.4826 × MAD provides a good approximation to the standard deviation (SD). Considering the left-skewed distribution of SUVRs that deviates from normality, the use of the median and MAD, which are elastic to outliers and distribution shape, allows for the application of a normal approximation formula to determine the tau positivity threshold.

[0162] To evaluate the potential savings in tau-PET scans by utilizing blood-based biomarkers through SUVR predictive models to identify subjects with tau accumulation below a specific threshold, sensitivities of 95%, 90%, and 80% were selected, allowing for false negative rates of 5%, 10%, and 20%, respectively. The specificity of these models was then determined at the tau-positive threshold and a higher threshold of 1.5 SUVR. In scenarios where tau-PET scans are performed only on subjects predicted to have tau levels above a specific threshold of interest, the reduction in PET scans reflects the accurate identification of subjects with tau levels below that threshold. This alignment is based on the model specificity corresponding to 1 minus the false positive rate. For simplicity, these evaluations were performed only for the WCGM and MTL regions.

[0163] The training and test sets included 144 and 98 Aβ+ subjects, respectively. Of these, approximately two-thirds had MCI due to AD, while the remainder had mild Alzheimer's disease. The distribution of tau-PET SUVR values ​​across the WCGM and MTL did not differ significantly between the training and test sets, as expected considering that these were random splits from the same cohort. The interquartile range (25th to 75th percentile) of the tau-PET SUVR distribution in the WCGM was 0.98–1.55 for the training set and 0.96–1.76 for the validation set. For the MTL, the interquartile range was 1.04–1.97 for the training set and 1.05–1.98 for the validation set. Figure 18B shows a nonlinear pattern in the association between plasma p-tau 217R and tau-PET SUVR values ​​across selected cortical regions (lateral temporal 1802, parietal 1804, WCGM 1806, occipital 1808, MTL 1810, frontal 1812), consistent with the disclosed embodiments. Figure 18C shows a similar nonlinear pattern in the association between plasma p-tau 217R and tau-PET SUVR values ​​across six Braak stage regions (Braak IV 1814, Braak I 1816, Braak V 1818, Braak III 1820, Braak II 1822, Braak VI 1824), consistent with the disclosed embodiments.

[0164] Key demographic and clinical features, including diagnosis, age, sex, race, ApoE4 status, and MMSE scores, are summarized in Figure 18A. Since the training and validation sets were derived from randomized 60-40 splits of the clinical trial, no significant differences were observed between the two sets in any variable except age. The Wilcoxon test was used to compare age, MMSE, and Tau-PET SUVR. The chi-square test was used to compare diagnosis, sex, race, and ApoE4 status.

[0165] Figure 19 shows an overview of the predictive performance of the SUVR model across various brain regions via cross-validation, consistent with the disclosed embodiments. In particular, Figure 19 shows the R of the SUVR model across cortical regions and Braak disease stages, derived through 10 iterations of 10-fold cross-validation. 2 The values ​​are shown. Notably, the model utilizing only plasma p-tau 217R in combination with age, sex, and tau brain regions performed better than models using alternative predictors such as clinical assessments, MRI scales, amyloid-PET centiloid values, and other plasma biomarkers (p<0.05 according to Hittner's test for correlation). This model showed robust R values ​​ranging from 0.338 to 0.635 for cortical regions and 0.236 to 0.639 for the six Braak stage regions. 2 The value was achieved. Furthermore, the integration of these additional predictor types (e.g., clinical assessment, MRI scale, amyloid PET centiloid) into the p-tau217R-based model was successful. 2 This did not result in any improvement in the values, and this was a consistent trend across other performance metrics, including the root mean squared error (RMSE) and MAE.

[0166] As shown in Figures 20A–20E, p-tau-217R emerged as the dominant predictor in all multivariate models consistent with the disclosed embodiments. The relative influence of the predictor is shown for models incorporating p-tau-217R alone (Figure 20A), p-tau-217R with clinical assessment (Figure 20B), p-tau-217R with structural MRI scale (Figure 20C), p-tau-217R with amyloid-PET centinloid (Figure 20D), and p-tau-217R with other plasma biomarkers (Figure 20E), respectively. As shown, plasma p-tau-217R emerged as the dominant predictor, exerting at least a 50% relative influence across all scenarios. Tau brain regions followed closely behind, contributing approximately 20% of the relative influence and reflecting differences in tau-PET SUVR distribution across different regions.

[0167] In these figures, CDR-SB is the Clinical Dementia Rating Sum of Boxes score; ADCSADL is the Alzheimer's Disease Cooperative Study - Activities of Daily Living score; ADCDR is the delayed word recall score; ADAS.14 is the Alzheimer's Disease Assessment Scale - Cognitive Subscale 14 score; MMRL is the recall score; CDR0104 is the social activity score, MMLG is the language score; VCPRCL is the left precentral cortex thickness score; VCIPCL / R is the left / right inferior parietal cortex thickness score; VVPREL / R is the left / right precuneus volume score; VSSFR is the right superior frontal cortex area score; VSCACR is the right caudal anterior cingulate area score; APET.Centiloid is the amyloid PET centiloid value; Ab42d40 is the Aβ42 / Aβ40 ratio; NfL is the neurofilament light chain value; GFAP is the glial fibrillary acidic protein value.

[0168] Due to the superior performance of the p-tau217R-based model, subsequent evaluations were focused exclusively on this model. Figure 21 shows the prediction performance observed during extended cross-validation on the test set, consistent with the disclosed embodiments. The performance of the p-tau217R-based model in predicting tau-PET SUVR for selected cortical and Braak stage regions in the test set is shown. Performance metrics include root mean square error (RMSE), mean absolute error (MAE), and R that compare predicted tau-PET SUVR versus observed tau-PET SUVR 2 values. The R of this model 2 values ranged from 0.387 to 0.577 for cortical regions and from 0.378 to 0.585 for Braak stage regions. In addition, MAE values ranged from 0.269 to 0.347 for cortical regions and from 0.238 to 0.370 for Braak stage regions, further emphasizing the robustness of the model's predictions.

[0169] As shown in Figure 22, incorporating the ApoE4 allele count into the model did not improve predictive performance, consistent with the disclosed embodiments. The performance of the p-tau217R-utilizing model after integrating the ApoE4 allele count for predicting tau-PET SUVR in selected cortical regions and six Braak stage regions within the test set is shown. Performance metrics include the root-squared mean error (RMSE), mean absolute error (MAE), and R, which evaluate the comparison between predicted and observed tau-PET SUVR. 2 The value is included. Incorporating the ApoE4 allele count did not result in an improvement in predictive performance compared to the model without the ApoE4 allele count.

[0170] As shown in Figure 23, the performance of the model dependent on p-tau-217 concentration, rather than the model based on p-tau-217R, demonstrated significantly inferior performance (p<0.05) in predicting tau-PET SUVR in selected cortical and Braak stage regions within the test set, consistent with the disclosed embodiments. This discrepancy confirms the benefit of normalizing p-tau-217 concentration with non-phosphorylated tau-217 to obtain relative p-tau-217R. This normalization highlights its role in effectively reducing inter-individual variability and enhancing model performance.

[0171] Figures 24A and 24B show the differences in SUVR prediction profiles and boundaries for reliable tau-PET SUVR predictions using plasma ptau-217R-based models for WCGM (Figure 24A) and MTL (Figure 24B), consistent with the disclosed embodiments. Figures 25A–25D show boundaries in reliable tau-PET SUVR predictions using plasma ptau-217R-based models for additional cortical regions beyond those shown in Figures 24A and 24B, namely the lateral temporal region (Figure 25A), frontal region (Figure 25B), occipital region (Figure 25C), and parietal region (Figure 25D), consistent with the disclosed embodiments. Figures 26A–26F show boundaries in reliable tau-PET SUVR using plasma ptau-217R-based models for six Braak stage regions, consistent with the disclosed embodiments. In Figures 24A and 24B, Figures 25A–25D, and Figures 26A–26F, the upper prediction limit was determined by fitting a Richards 5-parameter logistic model to the predicted tau-PET SUVR values ​​versus the observed tau-PET SUVR values ​​in the test set.

[0172] Figure 27 shows the upper bounds for reliable prediction of tau-PET SUVR in the test set by a model using plasma p-tau 217R for selected cortical regions and six Braak stage regions. The upper bounds ranged from 1.68 to 2.3 for cortical regions and from 1.42 to 2.21 for Braak stage regions, consistent with the disclosed embodiments. To fit these regional prediction upper bounds to the context, the quartiles (first, second, and third quartiles) of the observed SUVR distribution are shown for each region based on pooled data from 242 Aβ+ early Alzheimer's disease subjects across training and test sets.

[0173] Figure 27 further shows the SUVR thresholds for tau positivity in each region, which, when combined with the provided upper predictive limits, can serve as an indicator of the dynamic range. The distribution of tau-PET SUVR encompassed a wide range across each region, from low to moderate and high tau levels. Notably, the upper limit of SUVR prediction significantly exceeded the tau positivity threshold for the cortical and Braak stage regions. The upper limit of SUVR prediction also exceeded the third quartile of the observed SUVR values ​​across all regions, except for the Braak 1–2 region.

[0174] Figure 28 shows an evaluation of tau status prediction, consistent with the disclosed embodiments, which distinguishes whether the predicted SUVR value is below or above a specified threshold. In particular, Figure 28 shows the ability of a tau-PET SUVR prediction model using plasma p-tau 217R to identify subjects in a test set as either below or above two SUVR thresholds (one indicating tau positivity and the other a higher threshold of 1.5 SUVR). Performance metrics include area under receiver operating characteristic curve (AUROC), sensitivity, and specificity, as well as their respective 95% confidence intervals. Depending on the relative importance of false negatives and false positives in a particular use case, SUVR prediction may be used to ensure optimal sensitivity or specificity.

[0175] Figures 29A–29D show ROC curves and corresponding AUROC values ​​for predicting tau positivity and a higher threshold of 1.5 SUVR in selected cortical and six Braak stage regions, consistent with the disclosed embodiments. The model demonstrated accuracy in predicting tau positivity across various brain regions, with AUROC values ​​ranging from 0.84–0.92 for cortical regions and 0.85–0.95 for Braak stage regions. Similarly, the model demonstrated accuracy in predicting a SUVR threshold of 1.5, with AUROC values ​​ranging from 0.85–0.9 for cortical regions and 0.79–0.87 for Braak stage regions.

[0176] Figure 30 shows the results of repeating the performance evaluations described with respect to Figures 28 and 29A-29D for each demographic subgroup (i.e., age, sex) and ApoE4 subgroup in the test set, consistent with the disclosed embodiments. For simplicity, only the WCGM and MTL regions were considered for this evaluation. Predictive performance was consistent across demographic and ApoE4 subgroups in MTL and WCGM (p>0.05), with the exception of certain age subgroups, which may be due to the small sample size (n<20) in those groups (n=14 subjects with age ≥ 70 and MTL tau SUVR < tau positive threshold 1.173, and n=19 subjects with age < 70 and WCGM tau SUVR ≥ 1.5).

[0177] A reduction in PET scans could be considered for a scenario where tau-PET scans are used for confirmatory and detailed assessment in Aβ+ early AD patients predicted to have tau accumulation above a specific threshold of interest using blood-based biomarkers. Such an approach presents an attractive screening strategy in certain Alzheimer's disease clinical trials. It could potentially correspond to real-world practice and significantly reduce cost and time expenditure while reducing the burden on patients. However, the success of this strategy depends on keeping the false negative rate (FNR) and false positive rate (FPR) for tau status prediction reasonably low.

[0178] In this study, the effect of reducing PET scans on the SUVR predictive model was determined by fixing the sensitivity to 95%, 90%, and 80%, corresponding to acceptable false negative rates of 5%, 10%, and 20%, respectively, and by investigating the model's specificity (1 minus FPR) for predicting tau status for the threshold corresponding to tau positivity and a higher threshold of 1.5 SUVR. For simplicity, this evaluation considered only the WCGM and MTL regions. The percentage of PET scan reduction reflects the model's accuracy in accurately identifying subjects with subthreshold tau levels and is therefore consistent with its specificity.

[0179] Figure 31 shows the significant potential reduction in PET scans achievable by utilizing a SUVR prediction model using p-tau 217R, consistent with the disclosed embodiments. In this example, the SUVR prediction model was used to exclude subjects with predicted tau accumulation below a specific SUVR threshold of interest, while maintaining an acceptable false negative rate (FNR). Results from the MTL and WCGM regions are shown in Figure 31 for simplification. In the MTL region, the reduction rates reached 54.3% for a tau positive threshold of 1.173 and 68% for a SUVR threshold of 1.5. Similarly, reductions of 65.4% and 43.6% were achieved in WCGM, respectively. At a 10% FNR, the PET scan reductions increased to 62.9% and 76% in the MTL and 75% and 75.4% in WCGM, respectively, for the two SUVR thresholds. Notably, substantial savings were achieved in 20% of FNRs, reaching 88.6% and 84% for MTL and 88.5% and 81.5% for WCGM, respectively, for the two SUVR thresholds.

[0180] In the third trial, a predictive model consistent with the disclosed embodiments was developed to detect whole-cortical and regional tau concentrates. The predictive model was trained using MK6240 tau-PET SUVR data from a clinical trial training cohort of 354 amyloid-positive early AD patients. Global and Braak stage-specific tau-positivity thresholds were established using 52 cognitively non-impaired individuals from the Lantheus / Cervaux cohort through robust one-sided 95% upper confidence limits. A Bayesian model was constructed to predict overall tau positivity and tau concentrates across Braak stages 1–6. Predictors included amyloid PET CL, cognitive measures (e.g., CDR-SB, ADAS-Cog-13, and MMSE subscores and composite values), and plasma p-tau 181, along with demographics and ApoE4 status. Predictive performance was evaluated through internal cross-validation (IV) within the training cohort and external validation (EV) using an independent ADNI cohort of 243 subjects with the Flortauspiru Tracer.

[0181] Overall tau positivity prediction using amyloid-PET CL achieved AUROC of 81% (IV) and 80% (EV). This was comparable to the performance achieved by combining cognitive assessment and plasma p-tau 181 (AUROC: IV 77%, EV 81%) and outperformed individual predictors. For Braak stage prediction, another model using amyloid-PET CL effectively distinguished early stages from late stages (Braak 0-2 vs. 3-6; AUROC: IV 90%, EV 87%) and performed well even in advanced Braak-6 stage (AUROC: IV 79%, EV 83%). The combination of p-tau 181 and cognitive assessment also accurately predicted tau across Braak 0-2 vs. 3-6 (AUROC: IV 88%, EV 80%) and Braak-6 (AUROC: IV 74%, EV 72%).

[0182] Therefore, models consistent with the disclosed embodiments can predict the overall presence of tau and Braak staging using amyloid-PET CL, plasma p-tau 181, and cognitive assessments. Because these models have been validated across different tau-PET tracers, they may enhance the efficiency of tau-PET screening and benefit clinical and research endeavors.

[0183] In a fourth trial, a predictive model consistent with the disclosed embodiments was developed for the quantification of cerebral tau or Aβ levels. This trial investigated whether plasma p-tau 217 could predict amyloid PET CL and regional tau-PET SUVR. The predictive model using plasma p-tau 217 predicted cerebral amyloid to a continuous degree over a range suitable for early AD patients and comparable in performance to that for CSF Aβ42 / Aβ40. Other predictive models using plasma p-tau 217 predicted tau-PET uptake.

[0184] Plasma p-tau 217 and amyloid beta Aβ42 / Aβ40 were measured using immunoprecipitation-mass spectrometry. CL predictive models were trained in a clinical trial screening cohort (TC) of 1,242 cognitively normal subjects who received the 18F-NAV4694 tracer and validated in two other clinical trial screening cohorts (VC-1 and VC-2) of 357 and 284 early-stage AD patients, respectively, of which over 90% received florbetaben. MK6240 tau-PET SUVR predictive models were trained and tested using 60-40 random splits of a clinical trial cohort of 242 early-stage AD subjects. Stochastic gradient boosting was used to construct SGB models across brain regions. Regional tau positivity thresholds were set to robust one-sided superior 95% confidence limits in the cohort of individuals without elevated amyloid (CL<30).

[0185] Figure 32A shows predicted amyloid-PET CL values ​​consistent with the disclosed embodiments. The p-tau217-based model best predicted up to 77 CL, with correlations of 0.81 and 0.69 for VC-1 and VC-2, respectively. AUROC was 92% and 97% accurate for detecting CL > 20 in VC-1 and VC-2, respectively, and 94% and 89% accurate for detecting CL > 50, respectively. Age was selected by the model, but adding Aβ42 / Aβ40 and ApoE4 status did not improve the model's predictive performance.

[0186] Figure 32B shows tau-PET uptake predicted by a p-tau217-based model in the MTL, consistent with the disclosed embodiments. The p-tau217-based model best predicted tau-PET uptake up to 2.1 SUVR in the medial temporal lobe, with an upper limit ranging from 1.9 to 3.2 across brain regions and a correlation ranging from 0.55 to 0.73. AUROC had an accuracy of 86% for detecting whole cortical tau, 83% to 93% in Braak stage 1–6 regions, and 93%, 83%, and 86% in the medial temporal, frontal, and parietal lobes, respectively.

[0187] The fifth trial investigated the effectiveness of p-tau217R and aβ42 / aβ40 in predicting continuous values ​​of regional amyloid levels (e.g., PET CL levels) and identifying subjects with different levels of amyloid accumulation. The resulting predictive model may support clinical screening for amyloid accumulation and reduce the need for PET scans.

[0188] The fifth trial developed and validated a predictive model using patient data spanning the AD continuum, from the pre-symptomatic stage to early AD. Patient data were divided into three separate cohorts. The training cohort (TC) for building the predictive model included 904 cognitively unimpaired (CUI) and early AD subjects. These subjects were aggregated from two clinical trial cohorts to ensure comprehensive presentation of both CUI and mildly impaired individuals. The TC included 620 CUI subjects drawn from a randomized subset of the AHEAD 3-45 screening cohort (AHEAD 3-45 trial: A trial to evaluate the efficacy and safety of recanemab treatment in participants with pre-symptomatic Alzheimer's disease and elevated amyloidosis, and in participants with early pre-symptomatic Alzheimer's disease and moderate amyloidosis; NCT04468659). The TC also included 284 amyloid-positive (Aβ+) early-stage AD subjects from the ClarityAD clinical trial, where both amyloid PET and plasma biomarker data were available (Study to confirm the safety and efficacy of recanemab in participants with early-stage Alzheimer's disease; NCT03887455).

[0189] The first validation cohort (VC-1) was used to evaluate the performance of the predictive model. VC-1 included the remaining 622 CUI participants from the AHEAD 3-45 screening cohort.

[0190] The second validation cohort (VC-2) included 357 subjects evaluated for early-stage Alzheimer's disease (AD) for selection in the clinical trial. These subjects had a clinical diagnosis of MCI or suspected mild cognitive impairment (MMSE 24–30, global CDR of 0.5 or 1, and impairment in delayed word recall during screening). Approximately 46% of these subjects tested positive for Aβ by visual reading of PET scans, while the remainder were amyloid-negative and therefore had cognitive impairment attributable to causes other than AD. These participants were drawn from screening cohorts of two identically designed clinical trials that were part of the elenbecestat Phase 3 program (a placebo-controlled, double-blind, parallel-group, 24-month trial with an open-label extension period to evaluate the efficacy and safety of elenbecestat [E2609] in subjects with early-stage Alzheimer's disease; NCT02956486, MissionAD1 and NCT03036280, MissionAD2).

[0191] Standardized uptake ratio (SUVR) levels from amyloid PET scans using various tracers were converted to CL units for subjects within the training and validation cohorts. Subjects from the AHEAD 3-45 screening cohorts included in TC and VC-1 received the 18F-NAV4694 tracer. Of the ClarityAD subjects in TC, 89% received florbetaben, while the remaining 10% and 1% received florbetapir and flutemetamol, respectively. In VC-2, 93% of subjects from the MissionAD1 / 2 trial received florbetaben, and the remaining 7% received florbetapir. The interquartile ranges (25th to 75th percentiles) of the CL distribution in TC, VC-1, and VC-2 were 2–74, -0.4–42, and -1–73, respectively, demonstrating a diverse spectrum of amyloid levels across the cognitive continuum from non-disability to early AD. The optimal CL threshold for detecting the earliest amyloid accumulation has been reported in various studies to be in the range of 15–30.

[0192] Plasma biomarkers were measured using high-throughput mass spectrometry (MS) based assays. Blood samples were collected in K2 EDTA tubes, processed into plasma, and stored at -80°C until analysis. Plasma Aβ42, Aβ40, p-tau 217, and np-tau 217 concentrations were quantified using a liquid chromatography-MS / MS (LC-MS) analytical platform with tandem MS. The Aβ42 / Aβ40 and p-tau 217 R ratios were then calculated. The p-tau 217 R ratio was used for subsequent analyses to normalize inter-individual variability.

[0193] Age and MMSE distributions were compared across training and validation cohorts using the Kruskal-Wallis test, while sex and APOE status were compared using the chi-squared test.

[0194] A separate model for predicting PET CL was formulated using the Aβ42 / Aβ40 ratio alone, p-tau 217R alone, and combinations of both biomarkers, along with demographic variables (e.g., age, sex) included in all models. BLLR and SGB were used in model construction. Overfitting was avoided by internally adjusting the models with holdout datasets and cross-validation. The predictive model was implemented by assembling up to 1000 decision trees with up to three-way interactions between predictors. The rank and relative impact of each predictor in the predictive model were derived by evaluating the decrease in MSE each time a predictor was used as a root node to split the decision tree in the SGB algorithm, and these were then normalized to range from 0 to 100%. Individual conditional expectation (ICE) profiles were used to investigate the association between predictors and outcomes.

[0195] The predictive performance of the predictive model was evaluated through 10 iterations of 10-fold cross-validation within the TC. Subsequently, the model was evaluated in two validation cohorts, VC-1 and VC-2. The evaluation was based on R of observed PET CL versus predicted PET CL. 2This included determining the RMSE and MAE. The correlation between observed CL levels and predicted CL levels was compared between predictive models via Hittner's test for dependency correlation. The effectiveness of incorporating the APOE E4 allele number into the model was evaluated within this framework.

[0196] The range of PET CL values ​​that could be reliably predicted by the predictive model was evaluated using a Richards 5-parameter generalized logistic model. The logistic model was fitted using data from CUI subjects in VC-1 and early AD subjects in VC-2. This pooled data facilitated the estimation of the predicted CL range across the spectrum from pre-symptomatic to early AD, and the lower and upper plateau parameters of the model defined the predicted range.

[0197] The applicability of predictive models along the disease continuum (e.g., from the early to late stages of amyloid accumulation) for pre-symptomatic and early-stage AD subjects was evaluated using VC-1 and VC-2. The predictive models were used to assess amyloid status using a wide range of CL thresholds (15, 30, 40, 50, 70, and 90). Predictive performance was summarized via receiver operating characteristic (ROC) curves and associated metrics: area under the ROC curve (AUROC), sensitivity, and specificity. Subsequently, AUROC values ​​were compared across models for each CL threshold via the DeLong test with Bonferroni multiplicity adjustment for further insight into their comparative performance. This predictive performance was then evaluated for each demographic (e.g., age, sex, race) and genotype (e.g., APOE E4 status) subgroup for 15 CL thresholds to determine whether the models performed consistently across all subgroups. The choice to focus solely on the 15 CL threshold was due to significant class imbalances and limited sample size observed at higher thresholds within certain subgroups.

[0198] The ability of a predictive model to potentially reduce PET scan use by identifying subjects without amyloid accumulation was evaluated. A sensitivity of 95% was selected, and a false-negative rate of 5% was tolerated. Next, the specificity of the predictive model (1 minus the false-positive rate) was evaluated over various CL thresholds. Assuming that PET scans are performed exclusively on subjects predicted to have amyloid accumulation (Aβ+), accurate identification of truly amyloid-negative subjects by the predictive model would theoretically lead to a reduction in PET scan use.

[0199] Figure 33 shows a table summarizing patient characteristics in the study dataset, consistent with the disclosed embodiments. Patient characteristics include APOE4 status, age, sex, and MMSE score. The dataset included 904 subjects in the TC, and 622 and 357 subjects in two validation cohorts (VC-1 and VC-2), respectively. Significant differences between TC, VC-1, and VC-2 are shown in the table (* indicates significance at the p<0.05 level, and SD indicates standard deviation). Within the training cohort, 68% (n=620) had CUI, 21% (n=186) had mild cognitive impairment (MCI), and 11% (n=98) had mild AD. VC-1 consisted only of CUI individuals, while VC-2 included 90% (n=320) MCI cases and 10% (n=37) mild AD cases. The amyloid PET CL distribution extended across a broad spectrum along the disease continuum, encompassing CUI, pre-symptomatic, and early AD populations. As can be understood, VC-2, which included subjects with MCI and mild AD, showed significantly lower baseline MMSE scores compared to the other cohorts (p<0.05). Furthermore, the distribution of APOE E4 status, age, and sex differed significantly across the three cohorts.

[0200] Figure 34 shows the performance of the SGB and BLLR models on the VC-1 and VC-2 datasets, consistent with the disclosed embodiments. The SGB model outperformed the BLLR model. Furthermore, the model including p-tau217R showed significantly better correlation than the model without p-tau217R (p<0.001). The predictive model using plasma Aβ42 / Aβ40 showed R values ​​of 0.3 and 0.16 for VC-1 and VC-2, respectively. 2 The values ​​were achieved. In contrast, the predictive model using plasma p-tau 217R showed improved predictive performance, with R values ​​of 0.63 and 0.66 for VC-1 and VC-2, respectively. 2 The values ​​were obtained. Predictive models using plasma Aβ42 / Aβ40 and plasma p-tau-217R maintained high predictive effectiveness. Similar results were evident for other performance metrics, including MAE and RMSE. Inter-individual variability was nominally reduced by using p-tau-217R instead of p-tau-217 concentration, but models relying solely on p-tau-217 concentration showed comparable predictive performance. In VC-1 and VC-2, R 2 The values ​​were 0.61 and 0.64 for the model based on p-tau-217 concentration, and 0.64 and 0.62 for the model incorporating both p-tau-217 concentration and Aβ42 / Aβ40, respectively. These values ​​were not significantly different from those reported for the p-tau-217R-based model in Figure 34 (p>0.05 based on Hittner's test for comparing dependency).

[0201] Figure 35A shows the percentage of relative influence of predictors in the SGB model, consistent with the disclosed embodiments. The SGB model included p-tau217R, Aβ42 / Aβ40, and demographic variables age and sex as inputs. The dominant predictor was p-tau217R, with a relative influence of over 80%.

[0202] Figures 35B and 35C show ICE profiles for individual subject-levels and mean outcomes, consistent with the disclosed embodiments. Figure 35B shows the ICE profile for p-tau 217R, while Figure 35C shows the ICE profile for Aβ42 / Aβ40. The nature of the association between these biomarkers and predicted PET CL levels is shown for each subject (gray) and mean subject (black) through these individual conditional expectation profiles. The predicted profile for each subject was centered by subtracting the predicted PET CL level corresponding to the lowest value of the predictor. These profiles reveal a strong nonlinear association characterized by a sigmoid pattern. The flexibility of the SGB algorithm makes it possible to model such complex associations without requiring explicit assumptions or predefined specifications.

[0203] Figures 36A–36C show the predicted CL ranges and observed CL values ​​for three predictive models using different combinations of biomarker inputs, consistent with the disclosed embodiments. The predicted CL ranges were estimated using a 5-parameter logistic model. The logistic model was fitted using pooled data from VC-1 and VC-2. The predictive model including p-tau217R (Figure 36B, demonstrating a reliable predicted range of 10.3–99.6 CL) and the predictive model including both p-tau217R and the Aβ42 / Aβ40 ratio (Figure 36C, demonstrating a reliable predicted range of 8–89.1 CL) showed larger predicted CL ranges than the predictive model including the Aβ42 / Aβ40 ratio but not p-tau217R (Figure 36A, demonstrating a reliable predicted range of 20.8–51.7 CL). The upper plateau predicted using the p-tau217R-based model was higher, but both models with p-tau217R predicted CL levels across a broad spectrum along the continuum from pre-symptomatic to early Alzheimer's disease. The CL scale is anchored to 0, the mean level in young, healthy individuals who are expected to be amyloid-free, and 100, for individuals with moderate AD. Because the anchor is mean, values ​​for individual subjects can be less than 0 or greater than 100.

[0204] Figure 37 shows the performance of a developed predictive model for predicting amyloid status across a range of CL thresholds (15–90 CL), consistent with the disclosed embodiments. This evaluation included both CUI subjects in VC-1 and early AD subjects in VC-2. Performance metrics (AUROC, sensitivity, and specificity) with 95% confidence intervals are shown for the wide range of CL thresholds. For a predictive model based on Aβ42 / Aβ40 (e.g., including the Aβ42 / Aβ40 ratio but not p-tau-217R), the AUROC values ​​ranged from 77.5%–83.5% in VC-1 and 68.3%–78.1% in VC-2, showing a progressive decline in performance with increasing CL thresholds. For the p-tau217R-based prediction model (e.g., including p-tau217R but not the Aβ42 / Aβ40 ratio), the AUROC values ​​ranged from 89.9% to 93.9% for VC-1 and 90.6% to 93.8% for VC-2 (p<0.05), achieving optimal performance at the 50 CL threshold. For the combined prediction model (e.g., including both p-tau217R and Aβ42 / Aβ40), the AUROC values ​​ranged from 91.9% to 94.7% for VC-1 and 89.3% to 93.2% for VC-2. The combined prediction model showed improved performance compared to the p-tau217R-based prediction model for the 15-40 CL threshold in VC-1, with the AUROC value increasing by up to 2.7 percentage points (p<0.05). In VC-2, a gradual improvement was observed in the CL thresholds of 15 and 30, although these improvements did not reach statistical significance.

[0205] Figures 38A–38H show ROC curves and AUROC values ​​for different combinations of CL levels and models for subjects in VC-1 and VC-2, consistent with the disclosed embodiments. In Figures 38A–38H, curve 3802 corresponds to the predictive model based on Aβ42 / Aβ40 and p-tau-217R, curve 3804 corresponds to the predictive model based on p-tau-217R, and curve 3806 corresponds to the predictive model based on Aβ42 / Aβ40. The predictive model based on p-tau-217R predicted amyloid status for a wide range of CL thresholds. The combined predictive model showed significantly improved prediction (p<0.05) for amyloid levels up to 40 CL in VC-1 and improved (but not significantly improved) prediction up to 30 CL in VC-2. Performance was well maintained even at higher CL thresholds. In Figure 38A, the AUROC for the prediction models based on Aβ42 / Aβ40 and p-tau-217R, the prediction model based on p-tau-217R, and the prediction model based on Aβ42 / Aβ40 were 92.7%, 89.9%, and 83.5%, respectively. In Figure 38B, the AUROC for the prediction models based on Aβ42 / Aβ40 and p-tau-217R, the prediction model based on p-tau-217R, and the prediction model based on Aβ42 / Aβ40 were 94.5%, 92.1%, and 83.5%, respectively. In Figure 38C, the AUROC for the prediction models based on Aβ42 / Aβ40 and p-tau-217R, the prediction model based on p-tau-217R, and the prediction model based on Aβ42 / Aβ40 were 94%, 93.9%, and 80.6%, respectively. In Figure 38D, the AUROC for the prediction models based on Aβ42 / Aβ40 and p-tau217R, the prediction model based on p-tau217R, and the prediction model based on Aβ42 / Aβ40 were 94%, 93.9%, and 80.6%, respectively. In Figure 38E, the AUROC for the prediction models based on Aβ42 / Aβ40 and p-tau217R, the prediction model based on p-tau217R, and the prediction model based on Aβ42 / Aβ40 were 92.9%, 91.6%, and 78.1%, respectively.In Figure 38F, the AUROC for the prediction models based on Aβ42 / Aβ40 and p-tau217R, the prediction model based on p-tau217R, and the prediction model based on Aβ42 / Aβ40 were 93.2%, 92.7%, and 74.4%, respectively. In Figure 38G, the AUROC for the prediction models based on Aβ42 / Aβ40 and p-tau217R, the prediction model based on p-tau217R, and the prediction model based on Aβ42 / Aβ40 were 92.8%, 93.8%, and 71%, respectively. In Figure 38H, the AUROC for the prediction models based on Aβ42 / Aβ40 and p-tau217R, the prediction model based on p-tau217R, and the prediction model based on Aβ42 / Aβ40 were 92.9%, 93.8%, and 68.6%, respectively.

[0206] Including the APOE ε4 allele count in a prediction model based on Aβ42 / Aβ40 can lead to CL level predictions (e.g., R 2 The values ​​increased from 0.3 to 0.35 in VC-1 and from 0.16 to 0.22 in VC-2, respectively, and significantly improved the prediction of amyloid status across various CL thresholds (e.g., acquisition of up to 3.1 and 4.7 percent points of AUROC in VC-1 and VC-2, respectively) (p<0.05).

[0207] Including the number of APOE ε4 alleles in the p-tau217R-based predictive model significantly improved the prediction of amyloid status in VC-1 for amyloid levels up to the 70 CL threshold (e.g., increasing AUROC by up to 2.2 percentage points) (p<0.05). While consistent trends were observed in VC-2, they lacked statistical significance across most CL thresholds.

[0208] Including the APOE ε4 allele count in the combined predictive models significantly improved predictive performance only in VC-2, particularly at high amyloid levels (CL≧40), resulting in a gradual increase in AUROC of less than 1 percentage point.15 The performance of these CL predictive models at CL thresholds was evaluated across various demographic subgroups (age, sex, race, APOE ε4 status).15 CL thresholds were selected due to significant class imbalance for higher thresholds within certain subgroups and limited sample size. Across VC-1 and VC-2, predictive performance remained consistent across demographic subgroups (p>0.05), except within VC-2 where one model showed significantly different performance across racial subgroups. This inconsistency may result from unreliable AUROC estimates arising from the small sample size of a certain racial subgroup (n=11 non-white subjects with CL>15). Within VC-1, which had a relatively large sample size (n=22 non-white subjects with CL>15), the model performance did not differ significantly across these subgroups.

[0209] Figure 39 shows the predicted effects of using a CL predictive model to screen subjects for confirmatory PET scans, consistent with the disclosed embodiments. This scenario assumes that the CL predictive model will be used as a preliminary screening tool to exclude amyloid-negative subjects. Confirmatory PET scans will be performed on subjects predicted to be amyloid-positive. This prediction was generated using two validation cohorts (VC-1 and VC-2). The criteria for excluding subjects were selected to maintain a 5% false-negative rate (95% sensitivity) across various CL thresholds. The specificity (1 minus FPR) of the predictive model for predicting amyloid status across CL thresholds ranging from 15 to 90 was evaluated. The reduction in PET scans, expressed as a percentage, reflects the accurate identification of true Aβ-subjects and is consistent with the model's specificity.

[0210] As shown in Figure 39, screening using a predictive model based on Aβ42 / Aβ40 would potentially achieve PET scan savings ranging from 40.5% to 47.4% and 28.7% to 37.7% across different CL thresholds in VC-1 and VC-2, respectively. Screening using a predictive model based on p-tau217R would potentially achieve PET scan savings of 46.9% to 77.6% and 46.8% to 81.6% in VC-1 and VC-2, respectively. Screening using a combined predictive model would potentially achieve PET scan savings of 61.7% to 78.6% and 62.7% to 70.9% in VC-1 and VC-2, respectively. At lower thresholds of 15 and 30 CL, the combined predictive model demonstrated significantly greater PET scan savings compared to the p-tau-217R-based predictive model (e.g., 66% and 78.6% savings for VC-1 compared to 46.9% and 56.6%, respectively, and 68.4% and 70.5% savings for VC-2 compared to 46.8% and 67.1%, respectively).

[0211] The sixth trial investigated the efficacy of p-tau217R and aβ42 / aβ40 in predicting amyloid-beta status. The predictive model can be used for community-based screening (CBS). The predictive results can be used to select subjects for different clinical programs. Selected subjects may undergo further evaluation / confirmation via amyloid PET or CSF. Symptomatic individuals may be identified via CogState CBB assessment.

[0212] Figure 40 shows the patient characteristics of a cohort for constructing a brain Aβ predictive model consistent with the disclosed embodiments. The cohort included 1,236 subjects from a trial-independent screening cohort: 784 subjects (63.4%) were amyloid-negative (AN) with amyloid PET centiloid (CL) < 20; 132 subjects (10.7%) had low amyloid (AL) with CL between 20 and 40; and 320 subjects (25.9%) had high amyloid (AH) with CL > 40. An algorithm for predicting the three-level ordinal amyloid status (AN, AL, AH) was constructed using Bayesian hierarchical ordinal logistic regression. The predictive model can predict the three-level ordinal status, rather than just a simple binary amyloid-positive or negative status.

[0213] The predictive model considered the following predictors: demographics (e.g., age, sex, and optional BMI); plasma Aβ42 / Aβ40 ratio; and plasma p-tau 217R. In addition to including both biomarkers, the effect of including either Aβ42 / Aβ40 or p-tau 217R was evaluated. Plasma markers were measured using the Simoa assay. Performance accuracy was first assessed through 20 replicates of 10-segment stratified cross-validation using 70% of the cohort (n=865) and further tested in the remaining 30% (n=371).

[0214] Figure 41 further shows the distribution of Aβ42 / Aβ40 ratios and p-tau217R values ​​in the training cohort, consistent with the disclosed embodiments. Between groups, a clearer separation of values ​​was observed between p-tau217R values ​​than between Aβ42 / Aβ40 values.

[0215] Figure 42 shows the odds ratios and significance of each predictor in a combined biomarker prediction model consistent with the disclosed embodiments.

[0216] Figure 43 shows the ROC curves and AUROC values ​​for a combined predictive model and three different CL values, consistent with the disclosed embodiment. Curve 4302 is for CL < 20 and AUC = 0.96, curve 4304 is for CL 20 - 40 and AUC = 0.8, and curve 4306 is for CL > 40 and AUC = 0.96.

[0217] Figure 44 shows predicted amyloid-beta PET status classified by the demographic characteristics of a cohort of patients obtained through community-based screening, consistent with the disclosed embodiments. Such screening included blood sampling and CogState CBB cognitive assessment. Of the 138 subjects enrolled, 26 were unpredictable as they failed screening and therefore blood samples were not collected. Of the 112 subjects, 42 were predicted to meet the CL20-40 or CL>40 criteria and could therefore be referred to clinical trials.

[0218] The foregoing description is provided for illustrative purposes only. It is not exhaustive and is not limited to the detailed forms or embodiments disclosed. Modifications and adaptations of embodiments will become apparent from the considerations herein and the practices of the disclosed embodiments. For example, while the described implementations include hardware, systems and methods consistent with this disclosure may be implemented using hardware and software. In addition, while certain components are described as being coupled together, such components may be integrated with each other or distributed in any preferred manner.

[0219] Embodiments as described herein include systems, methods, and tangible non-transient computer-readable media. A method may, for example, be performed by at least one processor receiving instructions from a tangible non-transient computer-readable storage medium, at least in part. Similarly, a system consistent with the disclosure may include at least one processor and memory, where memory may be a tangible non-transient computer-readable storage medium. As used herein, tangible non-transient computer-readable storage medium refers to any type of physical memory on which information or data readable by at least one processor may be stored. Examples include random-access memory (RAM), read-only memory (ROM), volatile memory, non-volatile memory, hard drives, CD-ROMs, DVDs, flash drives, disks, registers, caches, and any other known physical storage media. Singular terms such as “memory” and “computer-readable storage medium” may further refer to multiple structures, such as multiple memories or computer-readable storage media. As used herein, “memory” may include any type of computer-readable storage medium unless otherwise specified. A computer-readable storage medium may store instructions for execution by at least one processor, including instructions for causing a processor to perform a process or step consistent with the embodiments herein. In addition, one or more computer-readable storage media may be used when implementing a computer implementation method. The term “non-transient computer-readable storage medium” should be understood to include tangible articles, excluding carrier waves and transient signals.

[0220] Furthermore, while exemplary embodiments are described herein, the scope includes all embodiments having equivalent elements, variations, omissions, combinations (e.g., in various embodiments), adaptations, or modifications based on this disclosure. The elements in the claims should be interpreted broadly based on the words used in the claims and not limited to the examples described herein or in the course of the application, and such examples should be interpreted as non-exclusive. Furthermore, the steps of the disclosed methods may be modified in any way, including rearranging the steps or inserting or deleting steps.

[0221] The features and advantages of this disclosure are evident from the detailed specification, and therefore the appended claims are intended to encompass all systems and methods that fall within the true spirit and scope of this disclosure. As used herein, the indefinite articles "a" and "an" mean "one or more." Similarly, the use of plural terms does not necessarily mean plural unless evident in a given context. Furthermore, since numerous variations and modifications readily arise from examining this disclosure, it is undesirable to limit this disclosure to the exact configurations and operations illustrated and described, and therefore all suitable variations and equivalents may be used as falling within the scope of this disclosure. Accordingly, the embodiments and examples disclosed are intended to be considered merely examples, and the true scope of this disclosure is indicated by the following claims and their equivalents.

[0222] Embodiments may be further described using the following clauses: A1. A system comprising at least one processor and at least one non-temporary computer-readable medium containing instructions that, when executed by the at least one processor, cause the system to perform an action to predict the degree of neurofibrillary densities present in the brain of a human, wherein the action comprises: obtaining a machine learning model trained to predict a class label for a patient from patient input data (the class label being one of a set of class labels corresponding to the degree of neurofibrillary densities present in the brain of a human, and the input data including patient image data, cognitive function data, and / or biomarker data); generating the class label by applying the input data to the machine learning model; and providing an index of the class label. A2. A system comprising at least one processor and at least one non-temporary computer-readable medium containing instructions that, when executed by the at least one processor, cause the system to perform an operation for training a machine learning model to predict the degree of neurofibrillary densities present in the brain of a human subject, wherein the operation includes: acquiring classification image data for a training patient; acquiring input data for a training patient (the input data includes image data, cognitive function data, and / or biomarker data); determining a class label for a training patient using the classification image data (the class label is one of a set of class labels corresponding to the degree of neurofibrillary densities present in the brain of a human); generating a training sample that associates the class label for a training patient with the input data for a training patient; training a machine learning model using the training sample to predict a class label from the input data; and providing the trained machine learning model to enable prediction of the degree of neurofibrillary densities present in the brain of a human subject. A3. The system of clause A2, where the patient is amyloid-positive. A4. A system in any one of clauses A2-A3, where the classification image data includes tau-PET image data. A5. The system according to any one of clauses A1 to A4, wherein the set of class labels corresponds to Braak stage. A6. The system according to any one of clauses A1 to A5, wherein the cognitive function data comprises 14-item Alzheimer's Disease Assessment Scale (ADAS-cog-14), Clinical Dementia Rating Sum of Boxes (CDR-SB), or Mini-Mental State Examination (MMSE) data. A7. The system according to any one of clauses A1 to A6, wherein the biomarker data comprises plasma p-tau181 data. A8. The system according to any one of clauses A1 to A7, wherein the image data comprises structural brain network module values or hub region values. A9. The system according to clause A8, wherein the acts further comprise: obtaining structural MRI data; and generating structural brain network module values or hub region values using the structural MRI data. A10. The system according to any one of clauses A1 to A7, wherein the image data comprises or is dependent on cortical thickness values for brain regions. A11. The system according to any one of clauses A1 to A7, wherein the image data comprises or is dependent on cortical thickness values for brain regions. A12. The system according to any one of clauses A1 to A11, wherein the input data further comprises patient demographic information and / or ApoE4 status. A13. The system according to any one of clauses A1 or A4 to A12, wherein the class label is used to determine whether a patient is suitable for treatment with an anti-amyloid β (Aβ) protofibril antibody. A14. The system according to any one of clauses A1 or A4 to A12, wherein the class label is used to monitor therapeutic efficacy in a patient. A15. The system according to any one of clauses A1 or A4 to A12, wherein the class label is used to detect a reduction in the level of neurofibrillary tangles present in the human brain. A16. The system according to any one of clauses A1 to A15, wherein the neurofibrillary tangles comprise tau tangles. B1. A system for training a machine learning model, comprising at least one processor; and at least one non-temporary computer-readable medium containing instructions that, when executed by the at least one processor, cause the system to perform an action, wherein the action includes acquiring a training dataset comprising observations for a patient (each observation for a patient comprising biomarker data for the patient; and tau region values, where at least one of the brain tau region values ​​is based on image data for the patient); training a machine learning model to predict at least one of the brain tau region values ​​based on the biomarker data using the training dataset; and providing the trained machine learning model. B2. The system of clause B1, including biomarker data such as plasma p-tau 217 or plasma amyloid beta (Ab) 42 / Ab 40 biomarker data. B3. A system of clause B1 or B2 in which at least one of the brain tau region values ​​includes a tau-PET cortical vs. cerebellar standard uptake ratio. B4. A system in any of the categories B1 to B3, where the image data includes positron emission tomography data. B5. The system of clause B4, in which positron emission tomography data is acquired using a radioactive tracer. B6. The machine learning model is one of the systems described in clauses B1 to B5, including a stochastic gradient boosting model. B7. A system from any of clauses B1 to B6, in which the region includes the medial temporal lobe, frontal lobe, and parietal lobe. B8. A system for predicting at least one of cerebral amyloid or tau region values, comprising at least one processor; and at least one non-temporary computer-readable medium containing instructions, when executed by the at least one processor, causing the system to perform an action, the action comprising: obtaining observations for a patient (the observations including biomarker data); obtaining one of the trained machine learning models of any of the B1-B7; predicting at least one cerebral tau region value based on the biomarker data using the trained machine learning model; and providing at least one predicted cerebral tau region value. C1. A system comprising at least one processor; and at least one non-temporary computer-readable medium containing instructions that, when executed by the at least one processor, cause the system to perform an action for predicting a brain tau or amyloid-beta state, wherein the action includes obtaining a machine learning model trained to predict a patient's brain tau or amyloid-beta state from the patient's subject data (the subject data includes the patient's biomarker data, and the brain tau or amyloid-beta state includes one or more continuous brain tau or amyloid-beta levels; and / or relates to brain tau or amyloid-beta levels in multiple regions of the patient's brain); generating a prediction of the patient's brain tau or amyloid-beta state by applying the patient's subject data to the machine learning model; and providing a prediction of the patient's brain tau or amyloid-beta state. C2. The system of clause C1 in which the cerebral tau or amyloid-beta condition is related to cerebral tau levels in multiple regions of the patient's brain. C3. A system of clause C2 in which multiple regions correspond to one or more Braak stages. C4. A system in any one of the clauses C1 to C4 in which multiple regions include five or more, ten or more, twenty or more, or fifty or more regions. C5. Any one of the systems in clauses C1 to C5, wherein the cerebral tau or amyloid-beta state includes one or more continuous cerebral tau or amyloid-beta levels. C6. A system of clause C6 comprising one or more continuous cerebral tau or amyloid-beta levels, including one or more tau-PET SUVR levels. C7. A system of clause C6 comprising one or more continuous cerebral tau or amyloid-beta levels, including one or more amyloid-PET SUVR or centroid levels. C8. The machine learning model is one of the systems described in clauses C1 through C7, including an ensemble tree-based model. C9. Ensemble tree-based models, including stochastic gradient boosting models, are part of the systems of clause C8. C10. One of the systems in any of clauses C1-C9, where the biomarker data includes plasma or CSF biomarker data. C11. Any one of the systems in clauses C1 to C10, in which the biomarker data includes phosphorylated tau levels, non-phosphorylated tau levels, or a function or combination of phosphorylated and non-phosphorylated tau levels. C12. Biomarker data includes p-tau 217 levels in any one of the systems specified in clauses C1 to C11. C13. Biomarker data including the p-tau-217 / np-tau-217 ratio in any one of the systems in clauses C1-C12. C14. Biomarker data including the Aβ42 / Aβ40 ratio, from any one of the systems in clauses C1 to C13. C15. The target data further includes demographic data from any one of the systems in clauses C1 to C14. C16. The target data further includes genomic data, and is one of the systems specified in clauses C1 to C15. C17. The target data further includes cognitive scale data from one of the systems in clauses C1 to C16. C18. Any one of the systems described in clauses C1 to C16, where the target data further does not include cognitive scale data or image data. C19. The target data further includes image data, and the image data includes structural brain network module values ​​or hub region values, in any one of the systems described in clauses C1 to C17. C20. A system of clause C19 whose operation further includes acquiring structural MRI data of a patient; and using the structural MRI data to generate structural brain network module values ​​or hub region values. C21. The patient meets the cognitive impairment criteria in one of the systems from C1 to C20. C22. The system of clause C21, where the cognitive impairment condition is early AD. C23. Any one of the systems in any of clauses C1 to C22, wherein the operation further includes providing instructions for the patient to undergo PET imaging based on a prediction of the brain tau or amyloid-beta state. C24. Any one of the systems in any of the clauses C1 to C23, wherein the operation further includes determining whether the patient is suitable for treatment with an anti-tau antibody and / or an anti-amyloid-beta (Aβ) protofibril antibody, at least in part, based on a prediction of the cerebral tau or amyloid-beta state. C25. Any one of the systems in clauses C1-C24, whose operation further includes monitoring the effectiveness of treatment in patients using predictions of cerebral tau or amyloid-beta status. C26. One system from any of clauses C1 to C25, in which the prediction of cerebral tau or amyloid-beta status includes the prediction of future cerebral tau or amyloid-beta status. C27. A system comprising at least one processor and at least one non-temporary computer-readable medium containing instructions that, when executed by the at least one processor, cause the system to perform an operation for training a machine learning model to predict brain tau or amyloid-beta states, wherein the operation includes: acquiring brain tau or amyloid-beta states for a training patient; acquiring reference data for the training patient (the reference data including biomarker data); generating training samples that associate brain tau or amyloid-beta states with the reference data for the training patient; training a machine learning model using the training samples to predict brain tau or amyloid-beta states from the reference data; and providing the trained machine learning model to enable the prediction of brain tau or amyloid-beta states. C28. The system of clause C27 in which the cerebral tau or amyloid-beta condition is related to cerebral tau levels in multiple regions of the patient's brain. C29. A system of clause C28 in which multiple regions correspond to one or more Braak stages. C30. A system containing multiple domains, including two or more domains, as specified in any one of clauses C28-C29. C31. Any one of the systems described in clauses C28-C30, in which multiple regions include five or more, ten or more, twenty or more, or fifty or more regions. C32. Any one of the systems in clauses C28 to C31, wherein the cerebral tau or amyloid-beta state includes one or more continuous cerebral tau or amyloid-beta levels. C33. A system of clause C32 comprising one or more continuous cerebral tau or amyloid-beta levels, including one or more tau-PET SUVR levels. C34. The machine learning model is one of the systems described in clauses C27-C33, including an ensemble tree-based model. C35. A system of Clause 34, including an ensemble tree-based model and a stochastic gradient boosting model. C36. Biomarker data includes plasma or CSF biomarker data in any one of the systems specified in clauses C27-C35. C37. Any one of the systems in clauses C27 to C36, in which the biomarker data includes phosphorylated tau levels, non-phosphorylated tau levels, or a function or combination of phosphorylated and non-phosphorylated tau levels. C38. Biomarker data including p-tau 217 levels in any one of the systems in clauses C27-C37. C39. Biomarker data including the p-tau-217 / np-tau-217 ratio in any one of the systems described in clauses C27-C38. C40. Biomarker data including the Aβ42 / Aβ40 ratio in any one of the systems in clauses C27-C39. C41. The target data further includes demographic data from any one of the systems described in clauses C27-C40. C42. The system according to any one of clauses C27 to C41, wherein the subject data further comprises genomic data. C43. The system according to any one of clauses C27 to C42, wherein the subject data further comprises cognitive scale data. C44. The system according to any one of clauses C27 to C42, wherein the subject data further does not comprise cognitive scale data or image data. C45. The system according to any one of clauses C27 to C43, wherein the subject data further comprises image data, and the image data comprises structural brain network module values or hub region values. C46. The system according to clause C45, wherein the acts further comprise: obtaining structural MRI data from a training patient; and generating structural brain network module values or hub region values using the structural MRI data. C47. The system according to any one of clauses C27 to C46, wherein the training patient satisfies a cognitive impairment condition. C48. The system according to clause C47, wherein the cognitive impairment condition is early-stage AD. C49. A method of selecting a patient for treatment with anti-tau therapy and / or anti-amyloid therapy, comprising: identifying that the patient has AD or is at risk of AD according to the system of any one of clauses C1 to C26; and administering anti-tau therapy or anti-amyloid therapy to the patient. C50. The method according to clause C49, wherein the anti-tau therapy and / or anti-amyloid therapy comprises an anti-tau antibody and / or an anti-amyloid antibody. C51. The method according to clause C50, wherein the anti-amyloid antibody comprises lecanemab, or the anti-tau antibody comprises E2814. C52. A method of treating a patient having or suspected of having AD, comprising: identifying that the patient has AD or is at risk of AD according to the system of any one of clauses C1 to C26; and administering anti-tau therapy and / or anti-amyloid therapy to the patient. C53. The method according to clause C52, wherein the anti-tau therapy and / or anti-amyloid therapy comprises an anti-tau antibody and / or an anti-amyloid antibody. C54. The method of clause C53, wherein the anti-amyloid antibody comprises lecanemab or the anti-tau antibody comprises E2814. C55. A method for monitoring the therapeutic efficacy of AD, comprising: obtaining the level of p-tau 217 and applying it to a system according to any one of the clauses C1 to C26 to measure the number of brain regions containing elevated levels and / or neurofibrillary fibrous masses; and administering a therapeutic agent and repeating the measurement, wherein a reduction or delay in the progression of brain regions containing neurofibrillary fibrous masses indicates therapeutic efficacy.

[0223] As used herein, unless otherwise specifically stated, the term "or" encompasses all possible combinations, except in cases where it is not feasible. For example, if it is stated that a component may include A or B, then unless otherwise specifically stated or unless it is not feasible, the component may include A, or B, or A and B. As a second example, if it is stated that a component may include A, B, or C, then unless otherwise specifically stated or unless it is not feasible, the component may include A, or B, or C, or A and B, or A and C, or B and C, or A and B and C.

[0224] Other embodiments will become apparent from the considerations herein and the practices of the embodiments disclosed herein. This specification and the examples are intended to be illustrative only, and the true scope and spirit of the embodiments disclosed are shown by the following claims.

Claims

1. It is a system, at least one processor; and A non-transient computer-readable medium containing instructions that, when executed by the at least one processor, cause the system to perform an action to predict brain tau or amyloid-beta state. The operation includes, Obtaining a machine learning model trained to predict the brain tau or amyloid-beta state of a patient from patient reference data, wherein the reference data includes the patient's biomarker data, and the brain tau or amyloid-beta state is Includes one or more consecutive values ​​of brain tau or amyloid-beta levels; And / or to obtain information relating to brain tau or amyloid-beta levels in multiple regions of the patient's brain; To generate predictions of the patient's brain tau or amyloid-beta state by applying the aforementioned patient's target data to the machine learning model; and To provide the prediction of the patient's brain tau or amyloid-beta state. A system that includes this.

2. The system according to claim 1, wherein the cerebral tau or amyloid-beta state is related to the cerebral tau level in multiple regions of the patient's brain.

3. The system according to claim 2, wherein the plurality of regions correspond to one or more Braak disease stages.

4. The system according to any one of claims 1 to 4, wherein the plurality of regions include five or more, ten or more, twenty or more, or fifty or more regions.

5. The system according to any one of claims 1 to 5, wherein the cerebral tau or amyloid-beta state includes one or more continuous cerebral tau or amyloid-beta levels.

6. The system according to claim 6, wherein the one or more continuous cerebral tau or amyloid-beta levels include one or more tau-PET SUVR levels.

7. The system according to claim 6, wherein the one or more continuous cerebral tau or amyloid-beta levels include one or more amyloid-PET SUVR or centroid levels.

8. The system according to any one of claims 1 to 7, wherein the machine learning model includes an ensemble tree-based model.

9. The system according to claim 8, wherein the ensemble tree-based model includes a stochastic gradient boosting model.

10. The system according to any one of claims 1 to 9, wherein the biomarker data includes plasma or CSF biomarker data.

11. The system according to any one of claims 1 to 10, wherein the biomarker data includes a function or combination of phosphorylated tau levels, non-phosphorylated tau levels, or phosphorylated and non-phosphorylated tau levels.

12. The system according to any one of claims 1 to 11, wherein the biomarker data includes the p-tau 217 level.

13. The system according to any one of claims 1 to 12, wherein the biomarker data includes the p-tau 217 / np-tau 217 ratio.

14. The system according to any one of claims 1 to 13, wherein the biomarker data includes the Aβ42 / Aβ40 ratio.

15. The system according to any one of claims 1 to 14, wherein the aforementioned target data further includes demographic data.

16. The system according to any one of claims 1 to 15, wherein the aforementioned target data further includes genome data.

17. The system according to any one of claims 1 to 16, wherein the aforementioned target data further includes cognitive scale data.

18. The system according to any one of claims 1 to 16, wherein the target data further does not include cognitive scale data or image data.

19. The system according to any one of claims 1 to 17, wherein the target data further includes image data, and the image data includes structural brain network module values ​​or hub region values.

20. The aforementioned operation further, To obtain structural MRI data of the aforementioned patient; and Using the aforementioned structural MRI data, generate the structural brain network module values ​​or hub region values. The system according to claim 19, including the system described in claim 19.

21. The system according to any one of claims 1 to 20, wherein the patient satisfies the cognitive impairment criteria.

22. The system according to claim 21, wherein the cognitive impairment condition is early AD.

23. The aforementioned operation further, The system according to any one of claims 1 to 22, comprising providing an instruction for the patient to undergo PET imaging based on the prediction of the brain tau or amyloid-beta state.

24. The aforementioned operation further, The system according to any one of claims 1 to 23, comprising determining whether the patient is suitable for treatment with an anti-tau antibody and / or an anti-amyloid-beta (Aβ) protofibril antibody, based on the prediction of the cerebral tau or amyloid-beta state, at least partially.

25. The aforementioned operation further, The system according to any one of claims 1 to 24, comprising monitoring the therapeutic effect in the patient using the prediction of the cerebral tau or amyloid-beta state.

26. The system according to any one of claims 1 to 25, wherein the prediction of the brain tau or amyloid-beta state includes a prediction of the prognosis of the future brain tau or amyloid-beta state.

27. It is a system, At least one processor, and A non-transient computer-readable medium containing instructions that, when executed by at least one processor, cause the system to perform actions to train a machine learning model for predicting brain tau or amyloid-beta states. The operation includes, To obtain the brain tau or amyloid-beta state of the trained patient; To acquire target data for the aforementioned training patient, the acquisition of such target data including biomarker data; To generate training samples that associate the aforementioned brain tau or amyloid-beta state with the aforementioned target data for the training patient; Training a machine learning model using the training samples to predict the brain tau or amyloid-beta state from the aforementioned target data; and To provide the trained machine learning model that enables the prediction of the brain tau or amyloid-beta state. A system that includes this.

28. The system according to claim 27, wherein the cerebral tau or amyloid-beta state is related to cerebral tau levels in multiple regions of the patient's brain.

29. The system according to claim 28, wherein the plurality of regions correspond to one or more Braak disease stages.

30. The system according to any one of claims 28 to 29, wherein the plurality of regions include two or more regions.

31. The system according to any one of claims 28 to 30, wherein the plurality of regions include five or more, ten or more, twenty or more, or fifty or more regions.

32. The system according to any one of claims 28 to 31, wherein the cerebral tau or amyloid-beta state includes one or more continuous cerebral tau or amyloid-beta levels.

33. The system according to claim 32, wherein the one or more continuous cerebral tau or amyloid-beta levels include one or more tau-PET SUVR levels.

34. The system according to any one of claims 27 to 33, wherein the machine learning model includes an ensemble tree-based model.

35. The system according to claim 34, wherein the ensemble tree-based model includes a stochastic gradient boosting model.

36. The system according to any one of claims 27 to 35, wherein the biomarker data includes plasma or CSF biomarker data.

37. The system according to any one of claims 27 to 36, wherein the biomarker data includes a function or combination of phosphorylated tau levels, non-phosphorylated tau levels, or phosphorylated and non-phosphorylated tau levels.

38. The system according to any one of claims 27 to 37, wherein the biomarker data includes the p-tau 217 level.

39. The system according to any one of claims 27 to 38, wherein the biomarker data includes the p-tau 217 / np-tau 217 ratio.

40. The system according to any one of claims 27 to 39, wherein the biomarker data includes the Aβ42 / Aβ40 ratio.

41. The system according to any one of claims 27 to 40, wherein the aforementioned target data further includes demographic data.

42. The system according to any one of claims 27 to 41, wherein the aforementioned target data further includes genome data.

43. The system according to any one of claims 27 to 42, wherein the aforementioned target data further includes cognitive scale data.

44. The system according to any one of claims 27 to 42, wherein the aforementioned target data further does not include cognitive scale data or image data.

45. The system according to any one of claims 27 to 43, wherein the target data further includes image data, and the image data includes structural brain network module values ​​or hub region values.

46. The aforementioned operation further, To acquire structural MRI data of the aforementioned training patient; and Using the aforementioned structural MRI data, generate the structural brain network module values ​​or hub region values. The system according to claim 45, including the system described in claim 45.

47. The system according to any one of claims 27 to 46, wherein the trained patient satisfies the cognitive impairment condition.

48. The system according to claim 47, wherein the cognitive impairment condition is early AD.

49. A method for selecting a patient for treatment with anti-tau therapy and / or anti-amyloid therapy, comprising identifying the patient as having AD or being at risk of AD according to a system described in any one of claims 1 to 26, and administering the anti-tau therapy or anti-amyloid therapy.

50. The method according to claim 49, wherein the anti-tau therapy and / or the anti-amyloid therapy comprises an anti-tau antibody and / or an anti-amyloid antibody.

51. The method according to claim 50, wherein the anti-amyloid antibody comprises lecanemab or the anti-tau antibody comprises E2814.

52. A method for treating a patient who has or is suspected of having AD, comprising identifying the patient as having or being at risk of AD according to any one of claims 1 to 26, and administering anti-tau therapy and / or anti-amyloid therapy.

53. The method according to claim 52, wherein the anti-tau therapy and / or anti-amyloid therapy comprises an anti-tau antibody and / or an anti-amyloid antibody.

54. The method according to claim 53, wherein the anti-amyloid antibody comprises lecanemab or the anti-tau antibody comprises E2814.

55. A method for monitoring the effectiveness of AD treatment, Obtaining the level of p-tau 217 and applying it to the system according to any one of claims 1 to 26 to measure the number of brain regions containing elevated levels and / or neurofibrillary groves; and The therapeutic agent is administered, and the measurement is repeated, and a reduction or delay in the progression of brain regions including neurofibrillary groves indicates therapeutic efficacy. Methods that include...

Images