Medical system for diagnosing the pathology and / or outcome of dementia

Abstract

A medical system that aids in determining a subject's future disease progression. More specifically, the present invention applies machine learning techniques to aid in predicting disease pathology and clinical outcomes in subjects symptomatic of cognitive decline, facilitating the clinical development of novel treatments.

Description

Technical Field

[0001] The subject matter of the present disclosure relates to a system for determining the progression of a disease in a subject. More specifically, the disclosed subject matter relates to the application of machine learning techniques for digital health tools to assist in clinical decision-making to facilitate the clinical deployment of novel therapies.

Background Art

[0002] Digital health technologies and their implementation in the clinical setting are matters of increasing interest. These technologies are becoming more powerful and sophisticated. In particular, machine learning is quite promising as a support to clinicians and other healthcare professionals who are striving to provide better treatment and care to patients.

[0003] Both the increasing availability of diagnostic tests and the need to recognize differences between abnormalities with nonspecific or overlapping symptoms make it possible to accumulate large amounts of data on an individual patient during their care. Some of these data may be redundant or it may be very important for the patient's outcome. If the tests performed were simply what was necessary to enable appropriate decision-making towards the desired outcome, it would be an advantage overall for the patent and the healthcare system.

[0004] Digital technologies can also be used in research environments that are particularly engaged in difficult research problems, for example, to more efficiently bring novel disease modifying drugs (DMDs) to market. Regulatory and reimbursement requirements are becoming increasingly stringent, and regulatory agencies such as the Food and Drug Administration (FDA) are implementing programs to trial and facilitate the development of drugs such as the Fast Track program, but the challenges faced by pharmaceutical companies during the clinical development of potential candidates are significant and remain a huge financial undertaking.

[0005] Bringing a therapy to market requires costly, large-scale, and lengthy clinical trials. In addition to the complexity, there is a high attrition rate caused by the recruitment of sickly subjects. The financial risk associated with taking drug candidates through clinical testing remains high. It is necessary to reduce it in order to support the pipeline by enabling drug candidates to move more efficiently through the pipeline, increasing their chances of success, and allowing other candidates to enter clinical testing.

[0006] When using dementia, particularly Alzheimer's disease (AD), as an example, there are no currently available disease-modifying drugs (DMDs). To date, many potential drugs have been tested, but none have shown any significant efficacy. This is partly because the trials have focused on subjects who are in the mid-stage rather than the early stage of the disease. To correct this, current research aims to recruit subjects with mild cognitive impairment (MCI) and identify those who are likely to convert to AD during the course of the trial where the drug is most likely to show an effect. More sensitive screening and stratification tools are needed to recruit subjects at this stage.

[0007] In other words, bringing a DMD to the Alzheimer's disease market is extremely complex due to the disease's heterogeneity and the need for large-scale and lengthy clinical trials. The principal confounder of DMD efficacy is the selection of appropriate subjects. More sensitive screening and stratification tools are needed to identify subjects suitable for a given cohort and endpoint.

[0008] Improvements in selecting patients for AD testing are possible by using amyloid positron emission tomography (PET) imaging. For example, Vizamyl™ (flutemetamol F-18 injection, GE Healthcare) is a radiopharmaceutical indicated for (PET) imaging of the brain to assess β-amyloid (Aβ) neuritic plaque density in adult patients with cognitive impairment being evaluated for AD or other causes of cognitive decline. See the GE Healthcare prescribing information for Vizamyl™ (flutemetamol F-18 injection). Other known agents include Neuraceq™ (florbetaben F18 injection, Piramal imaging), and Amyvid™ (florbetapir, Eli Lilly and Company). A negative Aβ scan indicates that neuritic plaques are sparse and does not correlate with the neuropathological diagnosis of AD at the time of image acquisition. The results of a negative scan reduce the likelihood that the patient's cognitive impairment is due to AD. A positive Aβ scan indicates that frequent amyloid neuritic plaques are present. Neurophysiological studies have shown that this amount of neuritic plaques is present in patients with AD, but there can also be patients with normal cognition, as well as other types of neurological conditions.

[0009] Aβ PET imaging can be considered as an adjunct to other diagnostic evaluations. A positive Aβ scan does not necessarily establish a diagnosis of AD or other cognitive abnormalities by itself. Amyloid PET-positive subjects may be eligible to enter clinical trials for amyloid-modulating therapies, but this test is insufficient as not all amyloid-positive subjects will progress to AD or will progress within the time frame relevant to the clinical trial. Taking the example of AD, this means the identification of subjects who are likely to have a positive amyloid scan.

[0010] As another example, the diagnosis of Parkinson's disease (PD) remains a challenge for patients with abnormal symptoms or those showing a lack of response to medication. PD is part of a group of diseases with common features named Parkinsonian Syndrome, including Progressive Supranuclear Palsy (PNP) and Multiple System Atrophy (MSA). Dopamine transporter (DAT) imaging can be used to obtain an accurate diagnosis by determining the loss of dopamine function. Several radiolabeled phenyltropane analogues are known for visualizing the dopamine transporter, and the approved product DaTscan (trademark) (GE Healthcare) with an 123 -I-labeled agent is one example. 123 -I-labelled agent) is an example.

[0011] Furthermore, in vivo imaging techniques that utilize radiopharmaceuticals such as PET and SPECT are relatively expensive and resource-intensive diagnostic methods. It is advantageous to be able to identify subjects with a high likelihood of having an outcome indicative of a pathological condition before performing such in vivo imaging techniques. Thus, for example, for PD, it is advantageous to identify subjects with a higher likelihood of having dopamine deficiency before performing molecular imaging. For AD, it is advantageous to identify subjects with a high likelihood of having amyloid plaques before performing molecular imaging.

[0012] Providing a prediction model based on an optimal combination of clinical biomarkers and imaging biomarkers has strong potential to improve the selection process. Common predictive biomarkers collected in clinical trials may include hippocampal volume obtained from MR imaging for prediction of brain atrophy and PET imaging for assessment of amyloid pathology. These biomarkers work better than using only typical inclusion criteria, but they cannot fully capture the complexity of the disease.

[0013] Recent studies have shown the ability to incorporate automation in quantifying brain volume and beta amyloid SUVr, along with the ability to use these means as an alternative to human visual image reads. See, for example, Thurfjell L, et al., Automated quantification of 18F-Flutemetamol PET Activity for Categorizing Scans as Negative or Positive for Brain Amyloid: Concordance with Visual Image Reads. Journal of Nuclear Medicine, 55, 1623-1628 (2014).

[0014] Attempts have been made to correlate clinical data with AD. For example, U.S. Patent No. 9,687,199 to Ithapu et al. entitled “Medical Imaging System Providing Disease Prognosis” discloses an artificial intelligence system for analyzing clinical data. This system utilizes multiple ranks of machine learning modules, each handling a separate portion of the clinical data to address the high dimensionality and low sample size of the data. However, the system disclosed in the '199 patent lacks any means for identifying subjects at risk of developing AD within a defined time frame. Summary of the Invention Problems to be Solved by the Invention

[0015] Therefore, there is a need for innovation to help address the deficiencies in current methodologies.

Means for Solving the Problem

[0016] The following presents a brief overview of one or more embodiments to provide a basic understanding of the presently disclosed subject matter. This overview is not an extensive overview of all contemplated embodiments, does not identify key or essential elements of all embodiments, and does not delineate the scope of any or all embodiments. Its sole purpose is to present some concepts of one or more embodiments in a simplified form as an introduction to the more detailed description that follows.

[0017] According to one aspect, a medical system for predicting the disease pathology or disease state of a subject having an uncertain cognitive state, the computer system being arranged to receive first medical data about the subject and at least partially configured as a trained learning machine trained with respect to second medical data, the trained learning machine being used to perform a prediction of disease pathology, and a display for displaying an indication of the prediction, the first medical data not including data obtained from a molecular imaging technique on the subject, the second medical data not including data obtained from one or more molecular imaging techniques, is disclosed. The disease pathology can be amyloid-beta (Aβ) positive in the subject's brain. The first medical data can include the results of a cognitive examination of the subject, the results of a cognitive examination of the subject, the age of the subject, the educational level of the subject, or some combination of these data.

[0018] According to another aspect, a medical system for predicting the clinical outcome of a subject having an uncertain cognitive state, comprising: a computer system arranged to receive first medical data about a subject and at least partially configured as a first trained learning machine trained with respect to second medical data and a second learning machine trained with respect to third medical data, the first trained learning machine and the second learning machine being used to perform a prediction of disease pathology; and a display for displaying an indication of the prediction, wherein the first medical data can include data obtained from a molecular imaging technique on the subject, and the first trained learning machine is at least partially trained with data obtained from one or more molecular imaging techniques. A medical system is disclosed.

[0019] According to another aspect, a method for predicting the disease pathology of a subject having an uncertain cognitive state, comprising: collecting cohort medical data in an electronic memory for a first set of subjects having known outcomes for the disease; using a computer system comprising a trained learning machine arranged to receive subject medical data about a subject and trained with respect to the cohort medical data, the trained learning machine being used to perform a prediction of disease pathology based at least in part on the subject medical data; and using a display to indicate the prediction. A method is disclosed.

[0020] According to another aspect, a method for predicting the clinical outcome of a subject having an uncertain cognitive state, comprising the steps of collecting first cohort medical data in an electronic memory for a first set of subjects having known outcomes for a disease, collecting second cohort medical data in an electronic memory for a second set of subjects having known outcomes for a disease, a computer system having a first trained learning machine arranged to receive subject medical data for a subject and trained with respect to the first cohort medical data, and a second trained learning machine trained with respect to the second cohort medical data, and using the first and second trained learning machines to make a prediction of disease pathology based at least in part on the subject medical data, and displaying an indication of the prediction. The first cohort medical data may include a data type different from the data type of the second cohort medical data. The first cohort medical data can include a data type that is at least partially the same as the data type of the second cohort medical data.

[0021] The medical system can be used for predicting the clinical outcome of a subject having an uncertain cognitive state and / or for predicting the disease pathology of a subject having an uncertain cognitive state. The ability to predict a specific disease pathology without molecular imaging as a prescreening tool makes it possible to enrich the cohort of subjects who go on to have a molecular imaging technique. There are also practical benefits in terms of health economics and reduction of subject exposure to radiation. Performing molecular imaging on subjects selected by prescreening enables stratification with respect to the rate of disease progression.

[0022] According to another aspect, there is disclosed a medical system for identifying a subject at risk of developing Alzheimer's disease (AD), which is arranged to receive medical data about one or more subjects with unknown outcomes regarding AD, and is at least partially constituted as a first trained learning machine that gives a first instruction and a second trained learning machine that gives a second instruction, and is configured to combine the first and second instructions to identify which subjects with unknown outcomes regarding AD are at risk of developing AD within a defined time frame. A medical system is disclosed that comprises a computer system and a display for displaying the identification. The first instruction can include the probability that the subject is Aβ positive. The second instruction can include the probability that the subject's mild cognitive impairment is rapidly progressive. The first trained learning machine and the second trained learning machine may be implemented on the same physical hardware. The medical system can comprise an electronic memory configured to store medical data and supply the medical data to the computer system. The medical data can include in vivo image data, cognitive function memory data, and genetic data. The in vivo image data may include standardized uptake value ratio (SUVR) or T1-weighted magnetic resonance (MR) volume measurements. The medical data may include demographic data and / or the electronic medical records of each subject. The subject may be a patient with mild cognitive impairment, and the defined time frame can be 3 years or less or 2 years or less. The first instruction may be the probability that the subject is Aβ positive, and the second instruction may be the probability that the subject's mild cognitive impairment is rapidly progressive, and the computer system may be configured to use the first probability and the second probability to identify subjects at risk of developing AD within the defined time frame.

[0023] According to another aspect, a method for identifying a subject at risk of developing Alzheimer's disease (AD), comprising the steps of collecting medical data in a memory-type electronic memory for a first set of subjects having known outcomes for the disease, identifying a subject at risk of developing AD within a defined time frame using a first trained learning machine and a second trained learning machine, and selecting a subject at risk of developing AD within a defined time frame for research on AD. The medical data may include in vivo image data, cognitive function memory data, and genetic data. The in vivo image data may include standard uptake value ratio (SUVR) or T1-weighted magnetic resonance (MR) volumetric measurements. The medical data may include demographic data and / or the electronic medical records of each subject. The subject may be a patient with mild cognitive impairment, and the defined time frame may be three years or less or several years or less.

[0024] According to another aspect, a method for determining the prognosis of a patient with Alzheimer's disease (AD), comprising the steps of administering an amyloid protein contrast agent to a patient in need thereof, imaging amyloid protein deposition in the patient, and correlating the imaging of amyloid deposition in the patient with a training set for the patient and other variables using known prognoses for AD.

[0025] According to another aspect, a method for treating a patient with Alzheimer's disease (AD), comprising the steps of comparing, in a computer, imaging data and other data obtained from the patient with a training set including imaging data and other data regarding the patient for known prognoses for AD to classify the patient into a disease cohort, identifying a drug therapy known to improve the outcome of the patient with AD within the disease cohort, and treating the patient with an effective amount of the identified drug therapy.

[0026] According to another aspect, there is provided a use which includes supplying medical data of one or more subjects having an unknown outcome with respect to Alzheimer's disease (AD) to a computer system that is at least partially configured as a first trained learning machine that gives a first instruction and a second trained learning machine that gives a second instruction, and which is configured to combine the first and second instructions to identify which subjects having an unknown outcome with respect to AD are at risk of developing AD within a defined time frame, in order to determine which subjects are at risk of developing AD.

[0027] According to another aspect, there is provided a medical system for classifying a subject as having mild cognitive impairment (MCI) or Alzheimer's disease (AD), which is arranged to receive medical data of one or more subjects having an unknown classification with respect to MCI or AD, and which is at least partially configured as a first trained learning machine that gives a first instruction and a second trained learning machine that gives a second instruction, and which is configured to combine the first and second instructions to identify which subjects have MCI and which subjects have AD, and a display for displaying the classification. The computer system may be configured to give a further classification to the subjects classified as MCI as to whether the MCI is early MCI or late MCI.

[0028] According to another aspect, a medical system for classifying a subject as having Alzheimer's disease (AD) or some other form of dementia, arranged to receive medical data about one or more subjects having an unknown classification for MCI or AD, and at least partially comprising a first trained learning machine that gives a first instruction and a second trained learning machine that gives a second instruction, and configured to combine the first instruction and the second instruction to identify which subjects have AD and which subjects have another form of dementia, a computer system, and a display for displaying the classification, a medical system is disclosed.

[0029] According to another aspect, a medical system for identifying additional instructions for a drug, arranged to receive medical data about one or more subjects taking the drug, and at least partially comprising a first trained learning machine that gives a first instruction and a second trained learning machine that gives a second instruction, and configured to combine the first and second instructions to determine whether the drug can be assigned to an instruction that is added to an existing instruction, a computer system, and a display for displaying the additional instruction, a medical system is disclosed.

[0030] Further features and advantages of the present invention, as well as the structure and operation of various embodiments of the present invention, will be described in more detail below with reference to the accompanying drawings. It should be noted that the present invention is not limited to the specific embodiments described herein. Such embodiments are merely shown herein for purposes of illustration. Further embodiments will be apparent to those skilled in the art based on the teachings contained herein.

[0031] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate the subject matter and, together with the description, further serve to explain the principles of the subject matter and to enable one skilled in the art to make and use the subject matter.

Brief Description of the Drawings

[0032]

Figure 1

Figure 2

Figure 3

Figure 4

DETAILED DESCRIPTION OF THE INVENTION

[0033] Next, various embodiments will be described with reference to the accompanying drawings. Like reference numerals are used throughout to refer to like elements. In the following description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of one or more embodiments. However, it may be apparent in some or all instances that any of the embodiments described below can be practiced without the specific design details described below. In other instances, well-known structures and devices are shown in block diagram form in order to assist in the description of one or more embodiments. The following presents a simplified summary of one or more embodiments in order to provide a basic understanding of the embodiments. This summary is not an extensive overview of all contemplated embodiments, nor does it identify key or essential elements of all embodiments, nor does it delineate the scope of any or all embodiments.

[0034] The described embodiments, and references in the specification such as "one embodiment", "an embodiment", "an example embodiment", etc., indicate that the described embodiments may include certain features, structures, or characteristics, but not all embodiments necessarily include the above - mentioned specific features, structures, or characteristics. Also, such phrases do not necessarily refer to the same embodiment. Further, when a particular feature, structure, or characteristic is described in relation to one embodiment, it should be understood that, whether explicitly described or not, it is within the knowledge of those skilled in the art to bring such features, structures, or characteristics in relation to other embodiments.

[0035] Embodiments of the present invention can be implemented in hardware, firmware, software, or any combination thereof. Embodiments of the present invention can also be implemented as instructions stored on a machine - readable medium that can be loaded and executed by one or more processors. A machine - readable medium can include any mechanism for storing or transmitting information in a form readable by a machine (e.g., a computing device). For example, a machine - readable medium can include read - only memory (ROM), random access memory (RAM), magnetic disk storage media, optical storage media, flash memory devices, electrical, optical, acoustic, or other forms of non - transitory machine - readable media. Further, firmware, software, routines, instructions can be described herein as performing some operations. However, such descriptions are for convenience only, and in fact, such operations result from computing devices, processors, controllers, or other devices that execute firmware, software, routines, instructions, etc.

[0036] To more clearly and concisely describe and point out the subject matter of the claimed invention, the following definitions are given for specific terms used throughout this specification and the claims. Any exemplification of specific terms in this specification should be considered as non - limiting examples.

[0037] The term "uncertain cognitive state" is used to apply to a subject exhibiting symptoms indicative of a disease or condition associated with cognitive decline. Symptoms may include confusion, poor motor coordination, identity confusion, impaired judgment, subjective memory loss, lack of concentration, and rambling speech. Non-limiting examples of known diseases and conditions associated with cognitive decline include AD, PD, MCI, TBI (traumatic brain injury), and Chronic Traumatic Encephalopathy (CTE).

[0038] The term "disease pathology" is used herein to refer to the pathological features typically associated with a disease or condition related to cognitive decline. Non-limiting examples of disease pathologies contemplated by the present invention include amyloid beta positivity (i.e., abnormal presence of amyloid), insufficient dopaminergic action, and the presence of neurofibrillary tangles (NFT: neurofibrillary tangle) (tau tangles).

[0039] The term "subject" is used herein to refer to any human or animal subject. In one aspect of an embodiment, the subject of the present invention is a mammal. In another aspect of an embodiment, the subject is an intact mammalian body in vivo. In another aspect of an embodiment, the subject is a human. The subject may be a Subjective Memory Complainer (SMC), or may suffer from mild cognitive impairment (MCI), and the likelihood of developing Alzheimer's disease (AD) is being investigated.

[0040] A "trained learning machine" is, for example, a computing system trained using machine learning with respect to a set of training data in order to have the ability to provide predictions, conclusions, and classifications when given new data. The computing system learns on its own without being explicitly programmed with the relationship between the data and the predictions, conclusions, and classifications.

[0041] "Display" can mean any device capable of displaying information in the form of alphanumeric characters or graphics, and typically includes a screen, circuitry, casing, and power source. Non-limiting examples of displays include computer monitors, tablet screens, and smartphone screens. A display may be local, i.e., directly connected to a local computer acting as the host of the system, or it may be remote, i.e., part of a user system where a computer acting as the host of the system communicates over a network, or, for example, the Internet having system services may be provided as a cloud-based utility.

[0042] As used herein, the term "prediction indication" is intended to refer to information displayed on a display in the form of alphanumeric characters or graphics. Non-limiting examples of indications include graphs, tables, bar graphs.

[0043] The term "molecular imaging technique" as used herein refers to an in vivo imaging technique that enables visualization (i.e., generation of molecular images) of cellular functions or molecular processes in a subject. "In vivo imaging technique" is a technique for non-invasively generating an image of all or part of the internal aspects of a subject. Non-limiting examples of molecular imaging techniques contemplated by the present invention include amyloid beta imaging, dopamine transporter imaging, and tau imaging.

[0044] The term "amyloid-beta (Aβ) positive" indicates, for example, that frequent amyloid neuritic plaques are moderate when observed with respect to an Aβ molecular image obtained by an Aβ molecular imaging agent. Prediction of the likelihood of amyloid positivity can be achieved by aspects of the present invention using data that does not include molecular imaging data. Non-limiting examples of such data include activities of daily living (ADL) baseline, digit span backward, logical memory II 30 min after story baseline, trail making part A-time (seconds), education, male gender, left caudate volume, right tonsil volume, and right caudate volume. In one embodiment, such data includes age, gender, mini-mental state score (MMSE), clinical dementia rating (CDR), clinical dementia severity rating scale (CDR-SB), ApoE gene test status, local left and right brain volumes of the hippocampus, parahippocampus, tonsil, olfactory cortex, medial temporal lobe, gyrus rectus, ventricle, and angular gyrus.

[0045] As used herein, the term "cognitive test" refers to a test performed on a subject to assist in determining a cognitive state. Typical tests are well known to those skilled in the art. Non-limiting examples of tests include various forms of IQ tests, memory, attention, and drawing concentration. Non-limiting examples of tests typically performed on non-human animal subjects include the mirror test and the T-maze test.

[0046] As used herein, the term "cohort medical data" is taken to mean the collection of data from a defined cohort of subjects (e.g., subjects potentially having a particular disease or condition and / or being considered for inclusion in a clinical study).

[0047] The term "mild cognitive impairment" (MCI) is used to refer to a condition that involves problems with cognitive functions such as memory, language, thinking, and judgment, which are often greater than normal age-related changes.

[0048] The term "rapidly progressive" refers to the relative transition from the initial symptoms of a disease or condition to the diagnosis of the disease or condition in a subject. Non-limiting examples are that the subject progresses from MCI to dementia within a relatively short time frame, e.g., within about 3 years. Some data can be useful for predicting rapid progression. Non-limiting examples of such data include ADL baseline, Mini-Mental State Examination (MMSE) baseline, semantic fluency task - animal category, composite amyloid standard uptake value ratio (SUVR) (pons), and hippocampal volume.

[0049] The present invention can be used by pharmaceutical companies at various points in the clinical trial workflow to help reduce time and costs in bringing new drugs to market. More specifically, by enriching the inclusion criteria (screening and stratification tools) or by identifying the appropriate cohort that is most likely to respond to treatment (market access tools).

[0050] For example, after an initial diagnostic examination has been performed, a trial may be considered by evaluating a patient's eligibility against inclusion criteria. Typically, this is done by a screening process that involves several inexpensive diagnostic tests and, in some cases, more expensive diagnostic tests, such as the use of positron emission tomography (PET) imaging. Phase 3 trials involve hundreds of subjects. For a trial, if the screening is highly inefficient, more subjects are screened than are needed to enter the trial. Thus, traditional screening methods are a source of economic burden for pharmaceutical companies and for the many subjects passing through this stage of the process in terms of the higher costs associated with some of the tests. The present invention uses clinical data collected by less expensive tests to predict the outcome of more expensive tests. In this way, patients are prioritized through the screening process. If necessary, additional features of the tool further stratify the subjects based on the rate of disease progression or the subjects most likely to respond to a therapy. In this way, there is enrichment of the inclusion criteria by prediction of the outcome.

[0051] Upon successful completion of a trial, the market access team requires data as evidence of drug efficacy to ensure that payers reimburse for the drug. Here, the present invention can help identify which patients will benefit from a drug, analyze existing real-world evidence from multiple sources, and create the outcome data that payers and service managers need for adoption.

[0052] By applying the present invention to data from the recruitment phase, the number of patients required to demonstrate drug efficacy can be reduced. Further, these techniques can predict the subjects most likely to show an effect of the drug, and thus improve the number of subjects and eligibility required to achieve a statistically relevant effect size. These efficiencies impact the overall cost of the trial by shortening the recruitment phase and lowering the overall cost associated with the screening test. This is just one example of a context in which the teachings of the present disclosure can be advantageously applied.

[0053] A general - purpose computer system 100 is shown in FIG. 1. The system 100 of FIG. 1 comprises a processor 110 connected to a bus 120. Also connected to this bus are a memory (e.g., firmware, RAM, or ROM) 130, a hard drive 140, a network interface 150, and an input / output (I / O) interface 160. The network interface 150 connects to a network 170. The I / O interface 160 can connect to one or more input / output devices such as an input device 180 and a display 190. The system 100 can have additional components such as ports, printers, CD / DVD ROM reader / writer, etc.

[0054] A general - purpose computer system 100, such as that shown in FIG. 1, can be configured for machine learning, i.e., machine learning that can perform tasks without specific programming. In one aspect, a set of training data ( "features") is given to the computer system 100, based on which the computer system 100 can learn to make predictions. Then, this computer system can make predictions based on new data.

[0055] One aspect of the disclosed subject matter involves the use of machine - learning - based analysis techniques that facilitate the identification of appropriate subjects based on their specific imaging data, genetic data, psychometric data, and demographic data, thereby determining optimal stratification of clinical - trial subjects with mild cognitive impairment (MCI). This technique involves the application of sequential or individual machine - learning models to exclude patients lacking appropriate biomarkers, rate of progression, or both.

[0056] Specifically, the process of selecting subjects for research is shown in FIG. 2. In step S20, subjects with mild cognitive impairment (MCI) are identified using standard test screening through psychological tests, collection of subject demographics, and acquisition of MRIs. In step S30, a machine learning model is used to predict the probability that the identified subjects are beta-amyloid (Aβ) positive. In step S40, amyloid PET images are obtained for subjects whose probability of being Aβ positive exceeds a certain threshold. In step S50, a machine learning model is used to predict the probability that a subject will progress to AD within the time frame of a clinical trial. In step S60, subjects whose probability of progression exceeds a certain threshold are selected for inclusion in the study.

[0057] Figure 2 shows a process in which two trained, i.e., machine learning models (one predicting Aβ positivity and one predicting transition probability) are used successively, i.e., in sequence, to select a subject. These models are also referred to herein as trained learning machines, and it will be understood that these trained learning machines may be implemented on separate hardware or on the same hardware. Alternatively, each model may be used individually, if desired, to customize the selection process. In this way, subjects who do not benefit from DMD under investigation do not need to undergo unnecessary tests, the cost of clinical trials can be dramatically reduced, and the overall test efficiency can be improved. Another possibility is to use the models in parallel, with each result being used as an input into subject selection. This is shown in Figure 3, where again at step S20, there is a step of identifying subjects with mild cognitive impairment using standard test screening through psychological tests, collection of subject demographics, and acquisition of MRIs. At step S70, Aβ positive markers are collected. At step S80, a machine learning model is used to predict the probability that the identified subject is Aβ positive. At step S90, this probability is used as an input into subject selection. At step S100, positive markers are collected for progressing to MCI. At step 110, a machine learning model is used to predict the probability that the identified subject has MCI, which is the fact of progressing. At step S90, this probability is used as another input into subject selection.

[0058] As an example, a model for predicting Aβ status and a model for predicting rapidly progressive MCI subjects defined as subjects who will transition to a probable AD (pAD: probable AD) status within a 36-month time frame were learned with respect to a Phase III clinical trial dataset including 232 MCI subjects, where 87 subjects were Aβ-positive and 81 subjects transitioned to AD within 36 months after baseline imaging examination. The primary objective of this Phase III clinical trial was to compare the pAD transition of MCI subjects with normal and abnormal 18 F] flutemetamol uptake. The long-term tracking of the subject transition status to pAD makes this a good dataset for the purpose of model learning.

[0059] The goal of the amyloid-positive model is to select subjects who are likely to have an amyloid-positive PET imaging scan before potential trial participants accept the cost and inconvenience of undergoing a PET scan. Phase III trial subjects lacking an amyloid PET scan were removed from the training set, leaving 227 subjects, and the amyloid positivity rate was 38%. The goal of the rapidly progressive MCI model is to select subjects who are likely to transition to pAD within the scope of a 3-year clinical trial. Phase III trial subjects lacking a conversion label or having a "not-converted" conversion label within less than 3 years were excluded from the training set, leaving 182 subjects, and the conversion rate was 45%.

[0060] Model generation was performed using features available from Phase III studies. Imaging data were quantified using automated quantification software, which generated quantified brain volumes from T1 MR images and SUVr values for the pons reference region from amyloid PET images. These quantified regions were then combined with the collected demographic, neuropsychological, and genetic data and selected using a feature selection algorithm for use in the model. An imputation method was utilized to handle subjects who deviated from the data outside the target level. Figure 2 graphically shows a generalized process for using this model.

[0061] The Aβ-positive model showed a 43% improvement in the test efficiency of selecting Aβ-positive subjects, a 79% improvement in model accuracy, and an 85% improvement in specificity when the improvement in test efficiency was measured as the positive predictive value (PPV) of the model compared to the original inclusion of Aβ-positive subjects. The rapidly progressive MCI model showed a 24% improvement in the selection of rapidly progressive MCI subjects, an 86% improvement in accuracy, and a 92% improvement in specificity when the improvement in test efficiency was measured as the PPV of the model compared to subjects identified as Aβ-positive by amyloid PET imaging alone.

[0062] Reporting of the performance metrics of the model brought about through tolerance verification is an effective method for representing the results of machine learning, but it is important to know that these can be adapted to independent datasets and are not unique to the training dataset.

[0063] To evaluate this ability, an Aβ-positive model was reconstructed using features common to both the Phase III dataset and the dataset of the Australian Imaging, Biomarker and Lifestyle Flagship Study of Ageing (AIBL). AIBL validation for Aβ-positive with n = 551 resulted in an accuracy score of 75% and a specificity of 87%, indicating extrapolation to other populations and disease states by incorporation of healthy and AD subjects within the AIBL dataset.

[0064] The use of machine learning for subject screening and stratification in clinical trials can increase the probability of success, indicating drug efficacy and the rate at which DMD therapies that produce the desired results become available. The performance of the models shown suggests that the incorporation efficiency can be improved by more than 50%. Furthermore, the validation of the Aβ-positive model using the AIBL dataset demonstrates the feasibility of extending a model learned on one population to a different population, and the ability to extend a positive risk score learned on MCI subjects only to both healthy and AD subjects.

[0065] Next, a method for constructing a model according to one aspect of the present invention will be described. This method can be considered to include two primary phases. The first is to prepare the data, i.e., "clean" the data. The second is to evaluate the data once it is prepared.

[0066] The features included for consideration can be obtained from any one or more categories such as psychometrics, demographics, genetics, AβPET, and T1 MRI. For psychometrics, the features can include data from CDR, CDR-SB, ADL, MMSE, semantic fluency task (animals), semantic fluency task (vegetables), ADAS-cog, digit span backward, digit span forward, digit symbol substitution test, logical memory II (30 minutes after the story), logical memory II (immediately after the story), trail making part A, and / or trail making part B. Demographic features can include education and / or age and / or gender. Genetic features can include ApoE. AβPET features can include frontal AβSUVR-PONS, anterior cingulate AβSUVR-PONS, precuneus post cingulate AβSUVR-PONS, parietal AβSUVR-PONS, temporal lateral AβSUVR-PONS, temporal mesial AβSUVR-PONS, occipital AβSUVR-PONS, sensorimotor AβSUVR-PONS, and / or composite AβSUVR-PONS. T1 MR can include hippocampal volume, thalamic volume, amygdala volume, putamen volume, caudate volume, parahippocampal volume, olfactory cortex volume, medial temporal lobe volume, ventricular volume, gyrus rectus volume, angular gyrus volume, total gray matter, and / or total white matter. Generally, a subset of these features such as ADL, MMSE, semantic fluency task (animals), composite AβSUVR-PONS, and hippocampal volume are used.

[0067] Regarding data preparation, it can also be considered as a group of techniques, such as feature selection (narrowing from a first set of potential features to a subset of potential features), feature engineering, subject selection, and complementing outlier features. Feature selection and feature engineering involve choosing a subset of features to be removed in a first instance based on prior knowledge about these features. For example, some features may be part of the inclusion criteria for a study and thus do not provide much variance for use in model building. Some features are known to be highly correlated with other features and are thus removed to prevent highly correlated features from biasing the model. Finally, some features may prove inappropriate for use as features in the model through experimentation.

[0068] The experimental phase of model building can also include using recursive feature elimination, random forest feature selection, or other similar methods known to those skilled in the art prior to the model building phase. In this way, features of low importance may be removed to simplify the model. This technique is useful when building the rapid progressive MCI model described below. After the experimental phase, the model features can be set in the model without further feature selection. Feature selection can be done using tools available from the scikit-learn machine learning library, which is available, for example, at scikit-learn.org.

[0069] Further feature engineering can include averaging features from MR scans of different cerebral hemispheres. This technique is also useful when building the rapid progressive MCI model described below. For example, the right hippocampal volume and the left hippocampal volume may be averaged to generate a hippocampal volume feature, and the left ventricle and the right ventricle may be summed to generate a ventricular volume feature.

[0070] When multiple datasets are used, it is useful to use the intersection across the feature sets to use only the features common to all datasets.

[0071] Regarding subject selection, in the case of the rapidly progressive MCI model, subjects who had a "No" transition status before the 3-year time point were excluded from the model building phase. For one reason or another, these subjects did not continue throughout the study, and thus their transition status was not reliable. In the case of the amyloid-positive model, subjects were given a negative (amyloid-negative) or positive (amyloid-positive) status based on their composite SUVr relative to a threshold of 0.62, and subjects lacking a transition label or PET scan that would allow for the generation of a transition label were removed from the model building population when their true status was unknown.

[0072] Regarding the complementation of outlying features, in order to account for subjects lacking values for the features used in the model, imputation of the median value may be performed, or the subject may be removed. Alternatively, the subject may be assigned the most common value for the feature. For example, if the ApoE status is a feature, a subject lacking an ApoE status may be given the modal status, i.e., the most common combination of alleles, rather than performing imputation regarding the median value of the data. Alternatively, if a feature is outlying, the variable is not substituted and the model is applied without that particular feature. Based on knowledge of important features, it is desirable for the modeling software to highlight to the user when a feature in an individual subject or subject cohort is unavailable. In some cases, this may allow for the addition of the corresponding value when it is known that such a test can be performed (e.g., to determine the ApoE status) and can improve the accuracy of the prediction.

[0073] Once the data is prepared, the data is evaluated. For constructing a rapid progression MCI model, stratified K-fold (5-fold) can be used. If both amyloid positivity and transition labels are present in the dataset used for model construction, the cohort can be stratified to have an equal distribution of amyloid positive converters, amyloid positive non-converters, amyloid negative converters, and amyloid negative non-converters. In this way, the splits can be stratified to better match the actual cohort of subjects than by purely stratifying based on transition status. The implementation of the stratified k-fold of the circuit run can be used for this purpose.

[0074] For constructing a rapid progression MCI model, logistic regression can also be used. For example, the circuit run implementation of logistic regression can be utilized for model construction using the implementation of the grid search CV of the circuit run to determine the optimal hyperparameters. The model can be reconstructed for each split to determine the mean statistics, and then the final model used can be reconstructed using all available data.

[0075] The model statistics for constructing a rapid progression MCI model can be calculated for each split, then averaged for the reported statistics and presented with 95% confidence intervals. The calculated statistics can include accuracy, f1-score, specificity, recall, pr-AUC, NPV, and precision.

[0076] To construct an amyloid-positive model, stratified K-fold (5-fold) can be used again. If both amyloid positivity and transition labels are present in the dataset used for model construction, the population can be stratified to have an equal distribution of amyloid-positive converters, amyloid-positive non-converters, amyloid-negative converters, and amyloid-negative non-converters. In this way, the split can be stratified to better match the actual population of subjects than by simply stratifying based on amyloid positivity. The transition labels are absent from the dataset used for construction, so the subjects can be stratified using only amyloid positivity to ensure that each split has an equal distribution of amyloid-positive and amyloid-negative subjects as in the true population of the dataset. Stratified k-fold of the circuit run can be used for this purpose.

[0077] Sequential forward feature selection can also be used to construct an amyloid-positive model. Forward feature selection can be performed within each split during the experimental phase to determine the optimal features for each split. After the optimal features are determined for each split, the features selected in three or more splits can be selected for use in the model construction phase. After the experimental phase, the forward feature selection can be removed from the processing steps, and the features used are not selected by the algorithm each time. The mlxtend (Machine Learning Extensions) Python library can be used for the implementation of forward feature selection.

[0078] A Gaussian Naive Bayes classifier can also be used to construct an amyloid-positive model. The circuit run implementation of Gaussian Naive Bayes can be used for this purpose with the default options maintained. The model can be reconstructed for each split to determine the mean statistics, and then the final model used can be reconstructed using all available data.

[0079] Model statistics can be calculated for each split, then averaged across the reported statistics, and presented with 95% confidence intervals. The calculated statistics can include accuracy, f1 score, specificity, recall, pr-AUC, NPV, and precision.

[0080] Machine learning models developed for the identification and stratification of ideal subjects for clinical trials significantly improve the ability of the automated process, thereby assisting clinical trials for the benefit of both the subjects and the institutions supporting the trials. The amyloid-positive model has shown the ability to improve test efficiency by up to 43%, while the rapid progressive MCI model has shown the ability to stratify subjects to increase transition efficiency by 17% over three years.

[0081] These models allow for the ability to fine-tune to meet the specific needs of clinical trials, and the models may be reconstructed to determine the priority of various statistical measures such as sensitivity, F1, or precision, and the subject risk threshold may be modified to target subjects showing the ideal pathology for a particular trial. Further, the models can be used individually or sequentially before expending resources on subjects who may not fit well into the trial, thereby demonstrating clinical trials in a stepwise manner to include subjects in the trial. For example, the amyloid-positive model can be used to select subjects to receive an amyloid PET scan, and then the rapid progressive MCI model can determine whether a subject is likely to progress at an ideal rate for a certain period of the trial before including the subject in a long and expensive protocol. The models have the ability to be combined in this way, but the models can also be used individually without sacrificing the performance of the models in any way.

[0082] More datasets may be included in the model building phase. Further, model building utilizing raw MR and PET imaging data can be used as an adjunct to, or in place of, the automated quantification algorithms described herein. The term "raw data" is intended to include direct pixel or voxel data.

[0083] The teachings of the present disclosure can be implemented as a web-based application. This is shown in FIG. 4. A client computer 400 including a web browser or a dedicated application 405 communicates with a virtual private cloud 410 through a distributed content information system 420 such as the Internet. The virtual private cloud 410 can include an application layer 430 to provide an interface with an execution framework 440. The execution framework 440 operates based on data supplied through the application layer 430 and stored in a database 460 using a trained model from a model storage unit 450, and gives, for example, the probability that a patient associated with the information given to the application layer 430 through the web browser 400 develops AD.

[0084] The foregoing example relates to using a system for selecting subjects for clinical research based on their prognosis, but this can also be used for other applications. As another example, the system may be used to predict the disease pathology of a subject having an uncertain cognitive state, and a learning machine trained by a computer system is used to perform the prediction of the disease pathology and display an indication of the prediction, and data obtained from molecular imaging techniques regarding the subject is not used. The disease pathology can be amyloid-beta (Aβ) positive in the subject's brain. The first medical data can include the results of a cognitive examination of the subject, the results of a cognitive examination of the subject, the age of the subject, the educational level of the subject, or some combination of these data.

[0085] As another example, a medical system can be used to predict the clinical outcome of a subject with an uncertain cognitive state. The system is arranged to receive first medical data about the subject and is at least partially comprised of a first trained learning machine trained with respect to second medical data and a second learning machine trained with respect to third medical data, the computer system being adapted to perform a prediction of disease pathology using the first trained learning machine and the second learning machine. The first medical data can include data obtained from molecular imaging techniques with respect to the subject, and the first trained learning machine is at least partially trained with data obtained from one or more molecular imaging techniques.

[0086] For example, as described above, the diagnosis of Parkinson's disease (PD) in patients with abnormal symptoms or who show a lack of response to medication can be difficult. DaTscan™ imaging can be used in an attempt to obtain an accurate diagnosis by determining the loss of dopaminergic function. Additionally, in vivo imaging techniques that utilize radiopharmaceuticals such as PET and single photon emission computed tomography (SPECT) are relatively expensive and resource-intensive diagnostic techniques. The use of trained learning as described above that uses in vivo imaging scan results as one of the characteristic techniques enhances the ability to identify subjects who are likely to have an outcome that implies a pathological condition. For PD, it is advantageous to identify subjects who are more likely to have dopaminergic deficiency before performing molecular imaging.

[0087] As another example, the teachings herein can be used to implement a method for predicting the disease pathology of a subject having an uncertain cognitive state, the cohort medical data is collected and stored in an electronic memory for a first set of subjects having known outcomes for the disease, the computer system is arranged to receive subject medical data for a subject, and comprises a trained learning machine trained with respect to the cohort medical data, and the computer system is adapted to use the trained learning machine to perform a prediction of disease pathology based at least in part on the subject medical data so as to provide a prediction of disease pathology.

[0088] As another example, the teachings herein can be used to implement a method for predicting the clinical outcome of a subject having an uncertain cognitive state, a first cohort medical data is collected and stored in an electronic memory for a first set of subjects having known outcomes for the disease, a second cohort medical data is collected and stored in an electronic memory for a second set of subjects having known outcomes for the disease, the computer system uses the subject medical data for the subject and comprises a first trained learning machine trained with respect to the first cohort medical data and a second trained learning machine trained with respect to the second cohort medical data. This computer system uses the first and second trained learning machines to perform a prediction of disease pathology based at least in part on the subject medical data and display an indication of the prediction. The first cohort medical data may include a data type different from the data type of the second cohort medical data. The first cohort medical data can include a data type that is at least partially the same as the data type of the second cohort medical data.

[0089] As another example, the teachings herein can be used to predict the clinical outcome of a subject having an uncertain cognitive state and / or to predict the disease pathology of a subject having an uncertain cognitive state.

[0090] The teachings of the present disclosure can also be applied to determine the prognosis of patients with AD. An amyloid protein contrast agent can be administered to a patient, and then the amyloid protein deposition in the patient can be imaged to obtain image data. The image data can be correlated with data from the patient's training set and other data using a prognosis known for AD.

[0091] The teachings of the present disclosure can also be applied to treat patients with AD. A computer can be used to compare imaging data and other data obtained from the patient with a training set including imaging data and other data regarding the patient and a prognosis known for AD to classify the patient into a disease cohort. The result of the comparison can be used to identify a known drug therapy, thereby improving the outcome of patients with AD within the disease cohort, and then the patient can be treated with an effective amount of the identified drug therapy.

[0092] The teachings of the present disclosure can also be applied to use medical data for one or more subjects with an unknown outcome for AD to determine which subjects are at risk of developing AD. The medical data is supplied to a computer system at least partially configured as a first trained learning machine that gives a first instruction and a second trained learning machine that gives a second instruction. The instructions are combined to identify which subjects with an unknown outcome for AD are at risk of developing AD within a defined time frame.

[0093] The teachings of the present disclosure can also be applied to classify subjects as having mild cognitive impairment (MCI) or Alzheimer's disease (AD). A computer is arranged to receive medical data about one or more subjects having an unknown classification for MCI or AD, and is at least partially configured as a first trained learning machine that gives a first instruction and a second trained learning machine that gives a second instruction, and may be configured to combine the first and second instructions to identify which subjects have MCI and which subjects have AD. The computer system may be further configured to give a further classification to subjects classified as MCI as to which MCI is early MCI or late MCI.

[0094] The teachings of the present disclosure can also be applied to a medical system for classifying subjects having Alzheimer's disease (AD) or some other form of dementia, wherein the computer system is arranged to receive medical data about one or more subjects having an unknown classification for MCI or AD, and is at least partially configured as a first trained learning machine that gives a first instruction and a second trained learning machine that gives a second instruction, and is configured to combine the first and second instructions to identify which subjects have AD and which subjects have some other form of dementia.

[0095] The teachings of the present disclosure can also be applied to a medical system for identifying additional instructions for a drug, wherein the computer system is arranged to receive medical data about one or more subjects taking the drug, and is at least partially configured as a first trained learning machine that gives a first instruction and a second trained learning machine that gives a second instruction, and is configured to combine the first and second instructions to determine whether the drug can be assigned to an instruction to be added to existing instructions.

[0096] This specification discloses the invention, including the best mode, using examples, and enables those skilled in the art to make and use any device or system and to practice the invention, including practicing any incorporated method. The patentable scope of the present invention is defined by the claims and may include other examples that occur to those skilled in the art. Such other examples are intended to be within the scope of the claims if they have structural elements that do not differ from the literal language of the claims or if they include equivalent structural elements that differ only slightly from the literal language of the claims. All patents and patent applications mentioned in the text are hereby incorporated by reference in their entirety as if individually incorporated herein.

Explanation of Reference Numerals

[0097] 100 General-purpose computer system, computer system, system 110 Processor 120 Bus 130 Memory 140 Hard drive 150 Network interface 160 Input / output (I / O) interface, I / O interface 170 Network 180 Input device 190 Display 400 Client computer, web browser 405 Web browser or dedicated application 410 Virtual private cloud 420 Distributed content information system 430 Application layer 440 Execution framework 450 Model storage unit 460 Database

Claims

1. A medical system for identifying whether a subject with mild cognitive impairment (MCI) is at risk of developing Alzheimer's disease, comprising: a memory for storing instructions; a first trained learning machine trained to predict, for each of a first set of subjects, a first probability that a patient with mild cognitive impairment (MCI) has a disease pathology associated with Alzheimer's disease based on first medical data related to the cognitive state of the patient, wherein the first medical data does not include data obtained from molecular imaging techniques, and the first trained learning machine is adjusted to identify, from the first set of subjects, a second set of subjects having a first probability of the disease pathology greater than a first threshold; a second trained learning machine trained to predict, for each of the second set of subjects, a second probability of progression to a diagnosis of Alzheimer's disease for a subject with mild cognitive impairment (MCI) based on (i) second medical data related to the cognitive state of the patient for each of the second set of subjects only, and (ii) image-related data obtained from molecular imaging techniques for each of the second set of subjects only, wherein the second trained learning machine is adjusted to identify, from the second set of subjects, a third set of subjects having a second probability of the progression greater than a second threshold; a display for displaying an output for identifying the third set of subjects; a processor for executing the instructions for implementing the same; and a medical system having the same.

2. The medical system according to claim 1, wherein the first trained learning machine and the second trained learning machine are adjusted for a clinical trial protocol related to a clinical trial of a drug.

3. The medical system according to claim 1, wherein the first trained learning machine and the second trained learning machine are adjusted for a clinical trial protocol associated with a patient's treatment plan. **Claim 4**: The medical system according to claim 1, wherein the image-related data includes magnetic resonance (MR) images or positron emission tomography (PET) images. **Claim 5**: The medical system according to claim 1, wherein the image-related data includes data related to at least one quantified region of a PET image or an MR image. **Claim 6** The medical system according to claim 1, wherein the disease pathology is a loss of brain cells that produce dopamine. **Claim 7** The medical system according to claim 1, wherein the first medical data and / or the second medical data includes demographic data. **Claim 8** The medical system according to claim 1, wherein the disease pathology is amyloid-beta (Aβ) positive in the brains of the subjects in the first subject set. **Claim 9** The medical system according to claim 8, wherein the first medical data includes activities of daily living (ADL) baseline, digit span backward, logical memory II 30 minutes after the story baseline, trail making part A time (seconds), education, gender, left caudate volume, right tonsil volume, and right caudate volume. **Claim 10** The medical system according to claim 8, wherein the first medical data includes age, gender, mini-mental state score (MMSE), clinical dementia rating (CDR), clinical dementia severity rating scale (CDR-SB), ApoE gene test status, local left and right brain volumes of the hippocampus, parahippocampal gyrus volume, tonsil volume, olfactory cortex volume, medial temporal lobe volume, gyrus rectus volume, ventricular volume, and angular gyrus volume. **Claim 11** The medical system according to claim 1, wherein the progression of the subject from MCI to a diagnosis of Alzheimer's disease is a rapid progression of MCI. **Claim 12** The first medical data includes an ADL baseline, a Mini-Mental State Examination (MMSE) baseline, a semantic fluency task - animal category, a composite amyloid standardized uptake value ratio (SUVR) (pons), and a hippocampal volume, for the medical system according to claim 11.

13. The first medical data includes the results of cognitive tests of the subjects in the first subject set, for the medical system according to claim 1.

14. The first medical data includes the ages of the subjects in the first subject set, for the medical system according to claim 1.

15. The first medical data includes the long-term education of the subjects in the first subject set, for the medical system according to claim 1.

16. The first medical data includes the results of ApoE gene tests, for the medical system according to claim 1.

17. The medical system according to claim 1, for use in a method of predicting a first probability of the disease pathology of a subject having an uncertain cognitive state.

18. The medical system according to claim 1, for use in a method of predicting the clinical outcome of a subject having an uncertain cognitive state.

19. The subjects in the first subject set are subjects for whom the possibility of developing Alzheimer's disease (AD) is being investigated, for the medical system according to claim 1.

Images