Dementia diagnosis

Abstract

Dementia diagnosis. A method for diagnosing dementia subtypes is described. The method includes obtaining brain imaging data related to one or more patients, analyzing the data using a deep learning model, and classifying one or more patients among a plurality of classes including a first patient class having a first dementia subtype and a second patient class having a second dementia subtype using the deep learning model. The deep learning model is trained using brain imaging data from patients, and the brain imaging data includes a first image set showing evidence of temporoparietal hypometabolism and a second image set showing evidence of hypometabolism in brain regions other than the temporal and parietal regions instead of or in addition to the temporal and parietal regions, where the first image set is labeled as being associated with the first dementia subtype and the second image set is labeled as being associated with the second dementia subtype. Associated systems and methods are also described.

Description

Technical Field

[0001] Field of the Invention The present invention relates to a method for diagnosing dementia subtypes, and more particularly, although not exclusively, to a method for classifying patients with Alzheimer's disease (AD) or mixed dementia, such as AD and cerebrovascular disease (CVD, mixed dementia), by analyzing brain images, particularly 18 (F)-fluoro-deoxy-glucose-positron emission tomography (FDG-PET) images.

Background Art

[0002] Background Alzheimer's disease (AD) may coexist with other brain pathologies that also cause cognitive decline, complicating both the diagnosis and treatment of AD. AD is frequently associated with cerebrovascular disease (CVD), and the presence of both of these pathologies has an additive effect on cognitive decline. CVD is associated with reduced cognitive performance and lowers the threshold for clinical findings of dementia in people with AD. The overlap between the two pathologies has given rise to the term "mixed dementia."

[0003] The coexistence and heterogeneity of pathologies in people with dementia make it difficult to distinguish typical AD from mixed pathologies. FDG-PET has been reported to be superior to other neuroimaging techniques, such as magnetic resonance imaging (MRI), computed tomography (CT), and blood flow single photon emission computed tomography (SPECT), in distinguishing AD from other pathologies. However, the visual interpretation of FDG-PET images requires intensive training of expert staff and is time-consuming. Furthermore, the visual interpretation of FDG-PET images of AD is subjective and depends on expertise (Morbelli et al., J Alzheimers Dis. 2015;44(3):815-26), and the agreement between expert visual analysis and clinical diagnosis is approximately 90% (Tripathi et al., Neuroradiol J 2014;27(1):13-21).

[0004] In addition, the visual interpretation of FDG-PET images may miss faint hypometabolism (Jo et al., Front. Aging Neurosci. 2019;11). This can be detected by semi-quantitative image analysis using a score or statistic calculated across the voxel values of the image. The sensitivity and specificity of semi-quantitative analysis have been reported to be as high as about 93% when differentiating almost certain AD from age-matched controls (Herholz et al., Neuroimage. 2002 Sep;17(1):302-16). However, semi-quantitative analysis is not without limitations as it can be prone to inaccurate results, especially when assessing brain regions that are either very small or adjacent to each other (Sarikaya et al., J Nucl Med Tech December 2018, 46 (4) 362-367). SUMMARY OF THE INVENTION PROBLEMS TO BE SOLVED BY THE INVENTION

[0005] Therefore, there is a need for an improved method for diagnosing AD, and specifically for differentiating AD from mixed dementia, without the drawbacks of the prior art. MEANS FOR SOLVING THE PROBLEMS

[0006] Summary of the Invention In a first aspect, there is provided a method for diagnosing a dementia subtype in one or more patients, the method comprising: obtaining brain imaging data relating to one or more patients; and providing the brain imaging data relating to one or more patients as an input to a deep learning model trained using brain imaging data from a plurality of patients, thereby classifying the one or more patients among a plurality of classes including a first patient class having a first dementia subtype and a second patient class having a second dementia subtype, wherein the brain imaging data includes a first image set showing evidence of temporo-parietal hypo-metabolism and a second image set showing evidence of hypo-metabolism in brain regions other than the temporal and parietal regions instead of or in addition to the temporal and parietal regions, wherein the first image set is labeled as being associated with the first dementia subtype and the second image set is labeled as being associated with the second dementia subtype.

[0007] The inventors have found that metabolic brain imaging data from patients with AD as well as mixed AD and CVD surprisingly exhibits significant differences in several brain regions that can be used to classify patients between these two subtypes using a semi-quantitative and deep learning image classification model, despite the coexistence and heterogeneity of pathologies, in people with dementia who are notoriously difficult to distinguish typical AD from mixed pathologies using existing methods.

[0008] Also, this specification describes a method for analyzing brain imaging data from a patient. The method includes obtaining brain imaging data related to the patient and analyzing the data using a deep learning model trained with brain imaging data from multiple patients. The brain imaging data includes a first set of images showing evidence of temporal-parietal hypometabolism, and a second set of images showing evidence of hypometabolism in brain regions other than the temporal and parietal regions, either instead of or in addition to the temporal and parietal regions. Here, the first set of images is labeled as being associated with a first subtype of dementia, and the second set of images is labeled as being associated with a second subtype of dementia. The method further includes classifying one or more patients among a plurality of classes including a first patient class having the first subtype of dementia and a second patient class having the second subtype of dementia using the deep learning model.

[0009] Obtaining brain imaging data related to one or more patients may include a processor that receives brain imaging data related to one or more patients. The steps of analyzing and classifying may be performed by the processor.

[0010] The first and second sets of images and their associated labels together form a training data set.

[0011] The deep learning model may provide, as an output, an indicator of the likelihood that the patient from whom the image was obtained belongs to the first class and / or the second class.

[0012] The method described in this specification is implemented on a computer unless otherwise indicated by context. In practice, the processes of image analysis using a deep learning model and training a deep learning model are complex and, in particular, require the analysis of large amounts of data through complex mathematical processing. Therefore, the method described in this specification goes far beyond human research capabilities.

[0013] In an embodiment, the dementia subtype is an Alzheimer's disease (AD) subtype. The first dementia subtype can be AD. The second dementia subtype can be AD in combination with cerebrovascular disease or vascular dementia. The second dementia type can be cerebrovascular disease or vascular dementia. The second dementia subtype can be mixed AD. The second dementia subtype can be a dementia other than AD.

[0014] Also described herein are methods and systems for diagnosing AD subtypes in one or more patients, the method comprising: obtaining brain imaging data related to one or more patients; and providing, as an input to a deep learning model trained using brain imaging data from a plurality of patients, the brain imaging data related to one or more patients, thereby classifying one or more patients among a plurality of classes including a first patient class having a first AD subtype and a second patient class having a second AD subtype, wherein the brain imaging data includes a first image set showing evidence of temporal-parietal hypometabolism, and a second image set showing evidence of hypometabolism in brain regions other than the temporal and parietal regions, instead of or in addition to the temporal and parietal regions, wherein the first image set is labeled as being associated with the first AD subtype, and the second image set is labeled as being associated with the second AD subtype.

[0015] In an embodiment, the classifying is between a first patient class having a first dementia subtype and a second patient class having a second dementia subtype. Thus, the classification can be a binary classification. For example, the classification can distinguish between patients with AD and patients with another dementia subtype, such as mixed AD or AD and CVD.

[0016] In an embodiment, the deep learning model is a deep neural network classifier. In an embodiment, the deep learning model includes a convolutional neural network (CNN). In an embodiment, the deep learning model includes a model pre-trained with unrelated image data. In an embodiment, the deep learning model includes a CNN pre-trained using a deep residual learning framework.

[0017] Convolutional neural networks have been shown to perform particularly well in image recognition tasks. Deep learning models can include models pre-trained on image recognition tasks against large image data collections such as the available ImageNet database. Such CNNs can be partially re-trained on new data by, for example, "freezing" (i.e., without re-training) lower-level layers (e.g., convolutional layers, etc.) trained to identify lower-level features of an image or by only fine-tuning said layers, and by training or re-training only higher-level layers (e.g., classification layers, etc.) to identify higher-level features specifically useful for the classification problem at hand. This partial re-training requires only determination by training of only a subset of the CNN's parameters (in the case of freezing), and / or requires only implementation of only fine-tuning of already optimized parameters, meaning that deep CNNs can be quickly trained using a limited amount of data. This can be particularly advantageous when it is difficult and / or laborious to obtain the amount of data available for training the classification task at hand. Deep residual learning is a learning framework developed for image recognition to address a problem known as "degradation" (the observation that accuracy saturates and then rapidly degrades as the depth of the network increases). Deep learning models can include pre-trained CNNs trained using deep residual learning, also known as ResNet. In an embodiment, the CNN is ResNet18. ResNet18 is a CNN trained on over one million images from the ImageNet database and, in its native form (before re-training), can classify images into 1000 object categories including, for example, keyboards, pencils, many animals, etc. The CNN was partially re-trained to perform various image classification tasks described herein (a process called "transfer learning").

[0018] In an embodiment, the deep learning model includes all layers of a CNN pre-trained in image recognition other than the classification layer, and a classification layer trained using the first and second image sets and associated labels. In an embodiment, the classification layer includes a fully connected layer and a softmax layer.

[0019] In an embodiment, the images of the first set and the second set show evidence of different metabolic activities in any one or more or all of the right prefrontal cortex, left prefrontal cortex, right temporal cortex, left temporal cortex, right parietal cortex, left parietal cortex, left cerebellum, and right cerebellum. In an embodiment, the images of the second image set show evidence of hypometabolism in brain regions including any one or more or all of the right prefrontal cortex, left prefrontal cortex, left cerebellum, and right cerebellum, instead of or in addition to one or more of the right temporal cortex, left temporal cortex, right parietal cortex, and left parietal cortex.

[0020] In an embodiment, hypometabolism means a lower glucose uptake rate and / or blood flow and / or FDG-PET-derived standardized uptake value ratio (SUVR) than expected for a control. The control can be a standard representing healthy patients.

[0021] In an embodiment, the brain imaging data is imaging data obtained using any functional brain imaging modality that provides information about the metabolic activity of the imaged brain regions. In an embodiment, the information about the metabolic activity of the imaged brain regions is obtained by detecting glucose uptake by the imaged brain regions and / or blood flow to the imaged brain regions.

[0022] In an embodiment, the brain imaging data is FDG-PET data or ASL data. In an embodiment, the brain imaging data is baseline functional brain imaging data.

[0023] In an embodiment, analyzing data using a deep learning model includes analyzing a single section of brain imaging data for each patient. In an embodiment, the method includes selecting a single section of the brain imaging data set for each patient. In an embodiment, the single section is a single axial section at the thalamus level, and / or the single section is a section including at least a part of the hippocampus and olfactory cortex.

[0024] In an embodiment, the brain imaging data used to train a deep learning model includes a single section of the brain imaging data set for each of a plurality of patients. In an embodiment, the method includes selecting a single section of the brain imaging data set for each of a plurality of patients. In an embodiment, the single section is a single axial section at the thalamus level. In an embodiment, the single section is a section including at least a part of the hippocampus and olfactory cortex. The single section means a single image, and the two terms can be used interchangeably. The inventors have found that a patient with AD or mixed AD can be accurately classified using a single section of the brain imaging data set. Further, it has been found that this section should preferably be an axial section at the thalamus level.

[0025] In an embodiment, the method can be repeated in one or more additional sections, and the classifications from each of the analyzed sections can be combined to obtain a classification of the patient. Alternatively, the deep learning model can be configured to take in a plurality of images as inputs and generate classifications of the plurality of images as outputs. For example, the deep learning model can include a plurality of instances of the deep learning model described herein that each take in a single image as an input and a function to combine the outputs of the model to obtain classifications of the plurality of images.

[0026] In an embodiment, the first image set includes one or more images of each of at least 30, at least 40, at least 50, at most 500, at most 200, or at most 100 patients. In an embodiment, the second image set includes one or more images of each of at least 30, at least 40, at least 50, at most 500, at most 200, or at most 100 patients. In an embodiment, the first image set includes one or more images for each of within 10% or within 20% of the patients or each of a plurality of patients, and one or more images for each thereof are included in the second image set. In an embodiment, the first and second image sets include one or more images of each of a plurality of patients, where the plurality of patients in the first and second image sets have matching age and gender.

[0027] The inventors have surprisingly discovered that it is possible to train a deep learning network to accurately distinguish dementia subtypes using a relatively small amount of training data having the characteristics described herein (i.e., specific hypometabolic patterns).

[0028] A group of patients with matching age and gender can mean a group of patients with approximately the same average and / or distribution of age and gender. For example, the group of patients can have an average age within 10% of each other, and / or a ratio of female to male patients within 10% of each other, and / or a ratio of patients in one or more age categories within 10% of each other.

[0029] In an embodiment, through expert review of the first and second image sets, the first image set is labeled as being associated with a first dementia subtype, and the second image set is labeled as being associated with a second dementia subtype. Optionally, each image in the first image set and the second image set is assigned the same label by at least two experts. Thus, an image in the first image set can be assigned a label associated with the first dementia subtype if at least two independently obtained expert-derived labels are the same. Similarly, an image in the second image set can be assigned a label associated with the second dementia subtype if at least two independently obtained expert-derived labels are the same. Images not assigned the same label by at least two experts can be excluded from the first and second image sets.

[0030] In an embodiment, the brain imaging data used to train the deep learning model includes images from multiple patients and images obtained from images from multiple patients by image augmentation. In an embodiment, image augmentation includes creating an inverted version of one or more of the images and / or creating a randomly rotated version of one or more of the images. For example, an image set (e.g., one or more or all of the original images) can be rotated by a randomly selected amount (e.g., -30° to +30°, etc.) between predetermined boundaries. As another example, an image set (e.g., one or more or all of the original images) can be flipped horizontally.

[0031] In an embodiment, the deep learning model classifies images of the first and second classes with at least 90% or at least 95% accuracy. In an embodiment, the deep learning model classifies images of the first and second classes with at least 90% or at least 94% sensitivity. In an embodiment, the deep learning model classifies images of the first and second classes with at least 90% or at least 95% specificity. The accuracy, specificity, and / or sensitivity of deep learning model classification can be measured by performing cross-validation, such as 5- or 10-fold cross-validation, to quantify the accuracy, specificity, and / or sensitivity at each split of the cross-validation, and optionally as an average or other summary metric across multiple splits.

[0032] In an embodiment, the method further comprises training a deep learning model using brain imaging data from a plurality of patients, the brain imaging data including a first set of images showing evidence of temporoparietal hypometabolism and a second set of images showing evidence of hypometabolism in brain regions other than or in addition to the temporal and parietal regions, wherein the first set of images is labeled as being associated with a first dementia subtype and the second set of images is labeled as being associated with a second dementia subtype.

[0033] According to a second aspect, there is provided a method of providing a tool for diagnosing a dementia subtype in one or more patients, the method comprising: obtaining training image data comprising a first image set showing evidence of temporoparietal hypometabolism and a second image set showing evidence of hypometabolism in brain regions other than the temporal and parietal regions instead of or in addition to the temporal and parietal regions, wherein the first image set is labeled as being associated with a first dementia subtype and the second image set is labeled as being associated with a second dementia subtype; and training a deep learning model using the training image data to classify patients among a plurality of classes including a first patient class having the first dementia subtype and a second patient class having the second dementia subtype.

[0034] The method of this aspect may have any of the features described in connection with the first aspect. The method of this aspect is preferably implemented on a computer. As described above, at least the step of training the deep learning model is implemented on a computer in any practical application. Thus, the steps of the method may include a processor that executes instructions for performing the steps. For example, obtaining training image data may include a processor that executes instructions for obtaining training image data from a data source (such as a database, computer memory, etc.). Similarly, training the deep learning model may include a processor that executes instructions for training the deep learning model.

[0035] Training a deep learning model can include at least partially retraining a pre-trained deep learning model. Partially retraining a deep learning model can include fixing the parameters for one or more of the lower layers of the model and determining the parameters for the remaining (higher level) layers of the model. Partially retraining a deep learning model can include fine-tuning the weights of multiple layers of the deep learning model and training the weights of one or more additional layers. The one or more additional layers can be classification layers. The classification layer can include a fully connected layer and a softmax layer.

[0036] In any of the aspects described herein, obtaining brain imaging data can include receiving brain imaging data from a computing device, imaging data acquisition means, data store, or user interface. The method of any aspect can include obtaining brain imaging data from a patient. In some cases, the method includes administering an imaging tracer to the patient and obtaining brain imaging data from the patient. This step may not be implemented on a computer and can be performed prior to any computer-implemented steps performed on the acquired data. Alternatively, all steps of the method can be implemented on a computer and can include receiving previously acquired brain imaging data.

[0037] Any of the methods may include providing the user, for example via a user interface, with the results of the classification, the trained deep learning model, and / or any information derived therefrom. The data store may be a public or private database. The results of the classification may include the likelihood of belonging to a first and / or second class obtained using the deep learning model, the classification label of one or more images, the classification label of one or more patients, the trained deep learning model, and / or one or more of the parameter values (e.g., architecture and weights) of the trained deep learning model. The information derived from the results of the classification may include one or more of a prognostic indicator derived from the classification obtained using the deep learning model, a treatment indicator derived from the classification obtained using the deep learning model, and a suitability indicator for participation in a clinical trial derived from the classification obtained using the deep learning model.

[0038] According to a third aspect, there is provided a method of selecting a subject having or diagnosed as likely to have AD for participation in a clinical trial, the method comprising: providing brain imaging data relating to the subject as an input to a deep learning model trained using brain imaging data from a plurality of patients, thereby classifying the subject among a plurality of classes including a first patient class having a first subtype of dementia and a second patient class having a second subtype of dementia, wherein the brain imaging data includes a first set of images showing evidence of temporoparietal hypometabolism and a second set of images showing evidence of hypometabolism in brain regions other than the temporal and parietal regions, either instead of or in addition to the temporal and parietal regions, wherein the first set of images is labeled as being associated with the first subtype of dementia and the second set of images is labeled as being associated with the second subtype of dementia, and selecting or excluding the subject for participation in the clinical trial depending on whether the subject is classified as having the first subtype of dementia or having the second subtype of dementia.

[0039] The method of this aspect may have any of the features described in relation to the first aspect. Specifically, the method may include selecting a subject for participation in a clinical trial if the subject is classified into the first class. The method may include excluding a subject from participation in a clinical trial if the subject is classified into the second class. The clinical trial may be a trial for treating AD. The method of this aspect may include any combination of some, all, or none of the preferred and optional features described above.

[0040] According to a fourth aspect, there is provided a method of providing a prognosis for a subject diagnosed as having or likely to have AD, the method comprising: providing brain imaging data relating to the subject as an input to a deep learning model trained using brain imaging data from a plurality of patients, thereby classifying the subject among a plurality of classes including a first patient class having a first dementia subtype and a second patient class having a second dementia subtype, wherein the brain imaging data includes a first image set showing evidence of temporoparietal hypometabolism, and a second image set showing evidence of hypometabolism in brain regions other than the temporal and parietal regions, instead of or in addition to the temporal and parietal regions, wherein the first image set is labeled as being associated with the first dementia subtype and the second image set is labeled as being associated with the second dementia subtype; classifying; determining the prognosis of the subject based on whether the subject is classified as having the first dementia subtype or having the second dementia subtype, optionally, determining the prognosis includes determining that the subject is likely to have a faster rate of cognitive decline if the subject is classified into the second class than if the subject is classified into the first class; determining. The method of this aspect may include any combination of some, all, or none of the preferred and optional features described above.

[0041] According to a fifth aspect, a system is provided for diagnosing a dementia subtype and / or for providing a tool for diagnosing a dementia subtype, the system comprising: one or more processors; and a computer-readable memory storing instructions for causing the processor to perform the method of any embodiment of any preceding aspect, including in particular any embodiment of either the first or the second aspect. The system may further include data acquisition means configured to obtain brain imaging data related to one or more patients. In some embodiments, the system may include, for example, one or more computers, servers, or cloud-based devices.

[0042] According to a sixth aspect, there is provided a non-transitory computer-readable storage medium containing machine-executable instructions for causing a processor to perform the method of any embodiment of any of the first to fourth aspects, including any one or any combination of any features specified in relation as far as applicable.

[0043] According to a seventh aspect, there is provided a computer program including executable code for causing a computer to perform the method of any embodiment of any preceding aspect, including in particular any embodiment of any of the first to fourth aspects, including any one or any combination of any features specified in relation as far as applicable when executed on the computer.

[0044] The present invention includes the described aspects and combinations of preferred features, except where such combinations are clearly unacceptable or are explicitly avoided.

[0045] Summary of the Figures Next, embodiments and experiments illustrating the principles of the present invention will be considered with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0046]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Mode for Carrying Out the Invention

[0047] Detailed Description of the Invention Next, aspects and embodiments of the present invention will be considered with reference to the accompanying drawings. Further aspects and embodiments will be apparent to those skilled in the art. All documents cited in the text are hereby incorporated by reference into this specification.

[0048] The inventors analyzed brain images (specifically FDG-PET scans) from patients with two different AD types (AD and mixed AD and CVD) and showed for the first time that these two subtypes can be identified using semi-quantitative image analysis, are clinically different, and can be distinguished with very high accuracy using automated image analysis with a deep neural network model trained by transfer learning.

[0049] Method for diagnosing a dementia subtype Figure 4 shows a flow diagram of a method for diagnosing a dementia subtype in a target patient and a method for providing a tool for diagnosing a dementia subtype. This method can be implemented on one or more computing devices. In step 400, training brain imaging data is obtained from a plurality of patients. The training brain imaging data can be baseline FDG-PET scan data. At any step 410, a single section (image) is selected for each of the plurality of patients. In step 420, if each of the single sections shows evidence of temporal-parietal hypometabolism, a first label is assigned, and if it shows evidence of hypometabolism in brain regions other than or in addition to the temporal and parietal regions, a second label is assigned. Step 420 can be performed by expert visual analysis and preferably includes obtaining a consensus label from at least two different experts in step 425. At any step 430, a subset of the training brain imaging data can be selected, for example, to include a similar number of images with the first and second labels assigned, and / or images with the first and second labels assigned that match in age and gender. At any step 435, the training brain imaging data or a subset thereof can be subjected to image augmentation, for example, by flipping and / or rotating one or more of the images. In step 440, the training brain image data (optionally selected and / or augmented) and the labels are used to train a deep learning model. Training the deep learning model can include at least partially retraining a pre-trained deep learning model. The pre-trained deep learning model can be one that was previously trained to perform an unrelated image recognition task. Thus, step 440 can include obtaining the pre-trained deep learning model from, for example, memory, a data store, a user interface, or a computing device in step 445.At any step 450, the performance of the deep learning model can be evaluated by determining its accuracy, sensitivity, and / or specificity, for example, when distinguishing between an image assigned the first label and an image assigned the second label. This can be performed using cross-validation, such as five-fold cross-validation for example. For this purpose, the training image data can be split between the training dataset and the validation dataset for each iteration of the training and validation processes. At any step 460, the trained deep learning model can be provided to the user. At step 470, brain imaging data related to one or more patients is obtained. At any step 480, a single section (image) is selected for each of the one or more patients. At step 490, a deep learning model obtained through steps 400 - 460 is used to classify one or more patients among a plurality of classes including a first patient class having a first dementia subtype and a second patient class having a second dementia subtype. At any step 500, the result of the classification or information derived therefrom is provided to the user. The information derived from the result of the classification can include one or more of diagnosis, prognosis, treatment, and selection for clinical trials, as further described below. The method described herein can include any of steps 400 - 460 and / or any of steps 470 - 500. Selecting a single image at step 410 and / or 480 can be performed manually or automatically. For example, a predetermined section of any brain image dataset can be selected (for example, a single-axis section including one or more predetermined target regions, such as the thalamus and / or hippocampus and / or olfactory cortex, etc. can be selected).

[0050] The terms "subject" and "patient" are used interchangeably throughout the present disclosure.

[0051] Use of the analysis results The results of such an analysis can be used to diagnose patients with AD or mixed AD, for example, a combination of AD and cerebrovascular disease (CVD). The results of such an analysis can be used to select patients participating in clinical trials. For example, a clinical trial can be designed to exclude patients with mixed AD and / or to include only patients with "typical" (i.e., non-mixed) AD. As another example, a clinical trial can be designed to exclude patients with "typical" AD and / or to include only patients with mixed AD. Also described herein is a method of selecting subjects having AD or diagnosed as likely to have AD for participation in a clinical trial, the method comprising: analyzing one or more brain images from a patient using the method described herein; and selecting or excluding a subject for participation in a clinical trial depending on whether the patient is classified as having a first dementia (or AD) subtype or a second dementia (or AD) subtype.

[0052] The results of such an analysis can be used to predict the level of cognitive impairment in patients diagnosed with or likely to have AD, where patients classified as having mixed AD and CVD are more likely to have severe cognitive impairment than patients with AD. The results of such an analysis can be used to provide a prognosis for patients diagnosed with or likely to have AD. In fact, patients with mixed AD in which AD coexists with CVD exhibit a faster rate of cognitive decline than patients with AD alone (see, for example, Zekry et al., Acta Neuropathol 2002;103:481-7, and Kapasi et al., Acta Neuropathol 2017;134:171-86). Thus, the present method can be used to predict whether a patient is likely to have a faster rate of cognitive decline than expected for an AD patient. Further, when treatments become available that can address vascular pathology, the present method could be used to aid in the selection of patients suitable for treatment. For example, Rodriguez et al. (Brain Res 1588(2014):144-149) reported that methylene blue can reduce the extent of hypoxic injury to brain tissue resulting from occlusion of the carotid artery supplying the brain. Thus, the present specification also describes a method for providing a prognosis for a subject diagnosed with or likely to have AD, the method comprising: analyzing one or more brain images from a patient using the method described herein; and identifying the prognosis of the subject based on whether the patient is classified as having a first dementia subtype or classified as having a second dementia subtype. For example, patients classified into the patient class of the second AD subtype may be associated with a poor prognosis compared to patients classified into the patient class of the first AD subtype. A poor prognosis may mean a faster rate of cognitive decline than expected for an AD patient. The rate of cognitive decline expected for an AD patient may be the average rate of cognitive decline observed across a cohort of AD patients. The cohort of AD patients may be patients of the same age and gender diagnosed as having AD. The cohort of AD patients may be patients diagnosed as having AD in the absence of CVD.The patient cohort can be patients diagnosed as having AD by analysis of FDG-PET images. The patient cohort can be patients having only temporoparietal hypometabolism in FDG-PET images.

[0053] The results of such an analysis can be used to treat or identify the treatment of patients with dementia. For example, patients diagnosed with "typical" dementia may be treated differently from patients diagnosed with mixed dementia, including AD and vascular dementia / CVD. For example, patients with mixed dementia may be treated with compounds for treating hypertension, reducing cholesterol, and / or preventing blood clots, instead of or in addition to compounds for treating AD. As another example, patients with mixed dementia may be recommended to follow lifestyle changes, such as changes in diet, activity regime, alcohol consumption, or tobacco consumption, instead of or in addition to the treatment of AD. As described above, patients with mixed dementia may be recommended or selected for treatment with therapeutic agents for treating vascular pathology, such as methylene blue. As another example, a patient with dementia may be treated with a compound for treating AD only after vascular dementia has been excluded and / or after AD has been diagnosed. For this reason, this specification also describes a method for identifying a therapeutic agent for a subject having or diagnosed as likely to have dementia, the method comprising: analyzing one or more brain images from a patient using the method described herein; and identifying the subject for treatment with a first therapeutic agent or a second therapeutic agent depending on whether the patient has been classified as having a first dementia (or AD) subtype or having a second dementia (or AD) subtype. The first therapeutic agent can be a therapeutic agent for treating AD. The second therapeutic agent can be a therapeutic agent for treating vascular dementia. The second therapeutic agent can include a therapeutic agent for treating vascular dementia and a therapeutic agent for treating AD. This specification also describes a method for selecting a subject having dementia for treatment with a therapeutic agent for vascular dementia, the method comprising: analyzing one or more brain images from a patient using the method described herein; and selecting the subject for treatment with a therapeutic agent for vascular dementia if the patient has been classified as having a second dementia subtype.This specification also describes a method of selecting a subject having dementia for treatment with a therapeutic agent for AD, the method comprising analyzing one or more brain images from a patient using the method described herein, and selecting the subject for treatment with a therapeutic agent for AD if the patient is classified as having a first dementia subtype. This specification also describes a method of treating a subject having dementia, the method comprising analyzing one or more brain images from the subject using the method described herein, and administering to the subject a therapeutically effective dose of a therapeutic agent for treating vascular dementia if the subject is classified into a second patient class having a second dementia subtype, and / or administering to the subject a therapeutically effective dose of a therapeutic agent for treating AD if the subject is classified into a first patient class having a first dementia subtype.

[0054] As used herein, "treatment" and "therapy" mean reducing, alleviating, or eliminating one or more symptoms of the disease being treated as compared to the symptoms before treatment.

[0055] System Figure 5 shows an embodiment of a system for diagnosing AD subtypes, and / or for classifying subjects as having AD or mixed dementia, and / or for analyzing brain images, and / or for providing a tool for diagnosing AD subtypes. The system includes a computing device 1 including a processor 101 and a computer-readable memory 102. In the illustrated embodiment, the computing device 1 also includes a user interface 103 exemplified as a screen, but may include any other means for communicating information to the user, such as via audible or visual signals. The computing device 1 is communicatively connected to data acquisition means 3 (also referred to as "brain image data acquisition means"), such as a PET machine or an MRI machine or a computing device associated therewith, and / or one or more databases 2 storing brain imaging data, for example, via a network. The one or more databases 2 may further store one or more of: one or more deep learning algorithms, training data, parameters (such as parameters of a deep learning model used for diagnosing AD subtypes), clinical and / or sample-related information, and the like. The computing device may be a smartphone, a tablet, a personal computer, or other computing device. The computing device is configured to implement a method for diagnosing AD subtypes, for analyzing brain images, and / or for classifying subjects as having AD or mixed dementia, as described herein. In an alternative embodiment, the computing device 1 is configured to communicate with a remote computing device (not shown) configured itself to implement the method described herein. In such a case, the remote computing device may also be configured to send the results of the method to the computing device.Communication between the computing device 1 and the remote computing device can be via a wired or wireless connection and can occur over a local or public network 6, such as over the public Internet. The data acquisition means 3 can be connected to the computing device 1 by a wired connection or, as illustrated, can communicate via a wireless connection, such as via WiFi and / or over the public Internet, etc. The connection between the computing device 1 and the data acquisition means 3 can be direct or indirect (e.g., via a remote computer, etc.). The data acquisition means 3 is configured to acquire brain imaging data from a patient. Any imaging protocol suitable for use in obtaining information regarding the metabolic activity of a patient's brain (e.g., PET, MRI, etc.) can be used within the scope of the present invention. The data acquisition means preferably includes a PET scanner and is preferably configured to collect FDG-PET images. The data acquisition means can include arterial spin labeling (ASL).

[0056] The following are presented as examples and should not be construed as limitations to the claims.

Example

[0057] Example As described above, by exploring findings using semi - quantitative methods and using visual interpretation of FDG - PET images, the limitations of traditional methods for detecting and monitoring AD from such images can be minimized. Recently, automated technologies such as artificial intelligence (AI) have been proposed as promising alternatives (Jo et al., Front. Aging Neurosci. 2019;11). Deep learning, a type of machine learning technology, can be trained directly using images, text, or sound to learn the classification patterns of given inputs. The inventors assume that such technologies could be applied using expert classification of AD subtypes as the gold standard, and also assume that the classification algorithm could be trained to distinguish between AD and mixed dementia. Deep learning algorithms such as convolutional neural networks (CNNs), which capture input images, analyze them according to a training algorithm, and classify them into a specific category, have been used to analyze medical images (Yadav et al., J Big Data 2019;6(1):113). However, fully training a new CNN is difficult and computationally demanding. Since a large dataset with a sample size of thousands is required to obtain acceptable accuracy, such methods are impractical for validating models with a medium - sized dataset having a sample size of hundreds. To address these problems, it is possible to train an image classification algorithm using a large generic image dataset and then "fine - tune" it using appropriate medical image data. Such an approach is called transfer learning and is a refinement of the CNN applicable to smaller datasets.

[0058] By introducing advanced methods for classifying AD into subtypes, diagnostic reliability will be improved. Also, there is a possibility of assisting clinical judgment in the management of people with AD. Here, the present inventors used visual classification of FDG-PET images to group AD participants into those having a typical AD pattern of FDG-PET hypometabolism (referred to as "typical type") and those having a mix of a typical AD pattern of FDG-PET hypometabolism and FDG-PET hypometabolism typical of CVD (referred to as "mixed type"). Next, the differences between these groups were assessed by region-of-interest-based analysis of SUVR (standardized uptake value ratio). Further, treating visual classification as the gold standard, a classification model based on transfer learning of the residual network-18 (ResNet-18) architecture (He et al. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR); 2016:770-778) was constructed to classify the images. Using data from two large-scale clinical trials of well-characterized people who met the clinical criteria for AD, the present inventors aimed to classify people as having typical or mixed AD by using conventional visual classification and then comparing it with classification using transfer learning of convolutional neural networks.

[0059] Method Participant selection and recruitment. In this example, baseline FDG-PET data collected as part of two large-scale Phase III clinical trials of the novel tau aggregation inhibitor drug leuco-methylthioninium (LMTX®) in 1,690 participants under 90 years of age who met the research criteria for mild to moderate AD were used. 794 participants were included in this study, and 896 participants were excluded due to either lack of FDG-PET imaging data or incomplete images. Details of the clinical trial approvals and methods are described in previously published works (see Gauthier et al. The Lancet. 2016;388(10062):2873-2884, Ilcock et al., J Alzheimers Dis. 2018;61(1):435-457, Chelter et al., J Alzheimers Dis. 2019;72(3):931-946). Written informed consent was obtained from all participants prior to registration in the trials TRx237-015 (clintrials.gov NCT01689246) and TRx237-005 (clintrials.gov NCT01689233). Consent for patients lacking decision-making capacity was provided by their legal representatives.

[0060] FDG PET imaging protocol. Brain images were obtained by positron emission tomography-computer tomography (PET CT) using high-resolution PET devices such as the Siemens high-resolution research tomograph (HRRT) system using a transmission source or low-dose CT for attenuation correction, or the GE Healthcare Systems Discovery STE. PET sites had to withstand strict quality criteria set and checked by Molecular NeuroImaging (mnimaging.com) prior to participant recruitment. A standard dose of 5mCi / 185MBq (±10%) FDG was injected intravenously into the antecubital vein over 1 minute via an indwelling catheter. PET images were acquired 30 minutes (±5 minutes) after FDG administration.

[0061] Image reconstruction and processing. The images were reconstructed with a 128×128 image matrix at a slice thickness of 4 mm. The images were zoomed in a 350 mm field of view (FOV) with 5.0 mm Gaussian full width at half maximum (FWHM) smoothing. Standard low-dose non-diagnostic CT acquisitions of the head were obtained for attenuation correction. The images were assessed for artifacts, patient motion, excessive noise, low counts, and patient positioning.

[0062] Visual classification of FDG-PET images. FDG-PET images were classified, using visual inspection, into those having a typical AD pattern of temporoparietal hypometabolism (see, e.g., Marcus et al. Clin Nucl Med 2014; 39:e413-26, Milke et al., Eur J Nucl Med 1994;21:1052-60, Kerrouche et al., J Cereb Blood Flow Metab 2006;26:1213-21, Garibotto et al., Neurobiol Aging 2017;52:183-95) and those having a mixed AD and CVD pattern (mixed pattern) of temporoparietal hypometabolism and deficits in one or more vascular regions (i.e., known blood supply regions of the brain such as the middle cerebral artery). Scans with a typical AD FDG-PET profile had a reduction in glucose uptake restricted to the temporoparietal region (see Fig. 1A). Those with a mixed AD / CVD profile had, in addition to typical temporoparietal hypometabolism, a reduction in FDG uptake and / or patchy uptake in specific vascular regions such as the middle cerebral artery (see Fig. 1B). The PMOD Alzheimer's disease discrimination analysis tool (PALZ, Haense et al., Journal of Nuclear Medicine. 2008;49(supplement 1):34P-34P) was used for visualization of FDG-PET images. Classification was based on visual review of scan images displayed in three planes using a standard color scale representing FDG uptake. To determine the level of inter-rater variability, a subset of the data was classified by a second trained observer and Cohen's kappa was calculated. There was a moderate level of agreement between raters. The inter-rater reliability measured using Cohen's kappa was 0.55, suggesting an acceptable level of agreement. In cases of disagreement, the images were reviewed jointly and considered until a consensus was reached. The final consensus classification obtained by two raters was used as the gold standard for the purpose of training a machine learning model (see below).Figure 1 shows examples of FDG pet images of the brains of participants classified into A) a typical AD pattern of glucose metabolism in the temporoparietal hypometabolic region 10 and B) a mixed pattern of patchy hypometabolism in other vascular regions of the brain such as the frontal and cerebellar regions 12 along with the temporoparietal hypometabolic region 10.

[0063] Analysis of visually classified FDG-PET images. Quantification of FDG-PET metabolism was achieved by determining the standardized uptake value ratio (SUVR) of different brain regions, namely, the right and left frontal cortices, the right and left temporal cortices, the right and left parietal cortices, the right and left occipital cortices, and the right and left cerebellar cortices, at intensities normalized to a bridge (Nugent et al., Scientific Reports. 2020;10(1):9261).

[0064] Explanation of the transfer learning method. For the purpose of transfer learning, 50 age- and gender-matched participants were selected from each of the visual classification AD group and the mixed group using variable adaptation randomization. Note that the learning dataset does not need to contain an equal number of participants in both groups, nor does it need to contain age- and gender-matched participants. However, the use of age- and gender-matched participants can increase the reliability of the resulting classifier with respect to how well it performs in the validation test (i.e., how well its performance generalizes to other datasets other than the ones it was trained on). The FDG-PET images of the participants were normalized to the PET template (the PET template included in SPM12) in the standard Montreal Neurological Institute (MNI) space using SPM12 implemented in MATLAB R2020a (available at www.fil.ion.ucl.ac.uk / spm / software / spm12). A single axial section at the thalamus level was used as the input to the transfer learning of ResNet-18. Note that any image containing the surrounding area of the hippocampus and olfactory cortex, specifically any axial section, could be used. Image augmentation was performed by creating a flipped version (horizontal flip) of each axial section and then randomly rotating the section by -30 to 30 degrees. Image augmentation advantageously reduces the risk of overfitting the network to the training images. 20 datasets (i.e., single images from 20 patients) were saved for testing, while the remaining 80 were split into a training dataset and a validation dataset at an 80% / 20% ratio. The final classification layer of ResNet-18 was replaced with a new fully connected layer containing two classes representing the mixed group and the AD group. ResNEt-18 is an 18-layer deep convolutional neural network (CNN). The version used was pre-trained on over 1 million images from the ImageNet database (www.image-net.org) and implemented in MatLab. Figure 2 shows the stages of the data flow of the machine learning algorithm used in this test.Specifically, Figure 2 shows the basic architecture of ResNet-18, which shows various layers of the CNN. The figure shows skip connections in the form of dotted lines (skipping some layers of the neural network and feeding the output of one layer as the input to the next layer) and numbers representing the output size. Average pooling is a global average pooling layer that reduces the spatial size of the representation, computational complexity, and the number of parameters. FC is a fully connected layer of 1000 units, which is a feedforward neural network with 1000 FC sub-layers and full connections to all previous sub-layers. The last layer is a softmax layer, which is the last activation function of the neural network and is used to normalize the output of the network to provide an output between 0 and 1. The average accuracy of five splits was reported using five-fold cross-validation. The network was trained using the trainNetwork function in MatLab, thereby fine-tuning the weights of the pre-trained network using the training images instead of training the network "from scratch" with randomly initialized weights. As a result, advantageously, the training was faster and very good performance was achieved despite the relatively medium size of the available training dataset.

[0065] Statistical analysis. All statistical analyses were performed in SPSS version 26. Descriptive statistics are presented with comparisons of means as appropriate. Student's t-test was used to test for differences in means for normally distributed continuous data, and the chi-square test was used to find differences in binary data. The relationships between variables were further explored with general linear modeling as appropriate. A p-value <.05 was considered significant, and the convention for representing the significance level was * <.05, ** <.01, *** <.001. Furthermore, sensitivity, specificity, and accuracy were calculated between the true and predicted subtypes of AD.

[0066] Results Table 1 shows the demographic and clinical characteristics of the participants in this study. The mean age of the participants was 70.56 years, and more women participated (55.16% vs. 44.83%). Of the total 794 participants (438 women), 533 (284 women) were classified as having typical AD, and 261 participants (154 women) were classified as having the mixed type (Table 1). Furthermore, for the purpose of transfer learning, 100 age- and sex-matched participants (50 each from typical AD and the mixed type) were selected. Participants classified as having mixed hypometabolism were younger and had greater cognitive impairment compared to typical AD participants.

[0067]

Table 1

[0068] Table 2 shows the mean SUVR differences in various regions of interest (ROIs) for participants with typical AD and those with the mixed subtype. The ROIs used were regions that exist in the Montreal Neurological Institute (MNI) standard space created in standard space and were copied to the normalized version of the patient data. By comparing the mean SUVRs of the AD and mixed pattern subtypes in the right and left frontal cortices, the right temporal cortex, and the right and left parietal cortices, significant differences were found. However, no significant differences were found in the left temporal cortex, the right and left occipital cortices, and the right and left cerebella.

[0069]

Table 2

[0070] Furthermore, by controlling for weighted age and sex and using these in the contrasts of the statistical analysis (Table 3), when comparing AD subtypes in various regions of interest (ROIs), it was observed that the SUVR was lower in those with the mixed pattern in all ROIs except the right and left occipital cortices. The SUVR in the right frontal cortex was statistically significant when correlated with the dementia subtype.

[0071]

Table 3

[0072] The ResNet-18 based classification model trained with transfer learning was found to have sensitivities of 94.73%, 95.23%, and 95%, specificities of 94.73%, 95.23%, and 95%, and accuracies of 94.73%, 95.23%, and 95% respectively in one randomly selected cross-validation loop. The average accuracy after 5-fold cross-validation was found to be 97.5%.

[0073] Figures 3A and 3B show occlusion sensitivity maps of six participants each from the typical AD (A) group and the mixed pattern (B) group. The occlusion maps show that region 30 of the image makes a positive contribution to group classification. Occlusion measures the network's sensitivity to occlusion (decrease in the probability score of a specific class) in different regions of the image by replacing small regions of the image with an occlusion mask (e.g., a gray square). Figures 3A and 3B show that the most informative regions for classifying patients between AD and the mixed pattern are the regions highlighted in Figure 3A, i.e., regions that are strong indicators of the hypometabolic AD pattern.

[0074] Discussion This test is unprecedented. It not only sub-classifies people with AD based on FDG-PET images, but also uses FDG-PET images analyzed using machine learning algorithms trained with data from traditional visual analysis. Specifically, it also introduces an automated classification approach using advanced techniques of transfer learning for the classification using FDG-PET images. The inventors visually classified people with AD into typical AD and mixed subtypes based on the hypometabolic patterns seen in FDG-PET images and regarded this classification as the gold standard. The two subtypes of Alzheimer's disease based on the visual analysis of FDG-PET not only had significant differences in the visual decision patterns of glucose uptake, but also had different clinical and demographic characteristics in that people with the mixed subtype were younger and had greater cognitive impairment. To explore the findings of the visual analysis, semi-quantitative analysis was performed and SUVR was calculated in standard ROIs. The regions corresponding to metabolic deficits in the visual analysis had lower SUVR. Notably, in this test, glucose uptake was higher in people with typical AD compared to those with the mixed type in most ROIs, except for the posterior cortex where no significant difference was found. Since uptake was lower in people with the mixed type classification compared to those with typical AD, it is suggested that blood supply is less sufficient in most regions in people with the mixed pattern, consistent with widespread brain parenchymal hypometabolism. Since this is consistent with recent research, it is suggested that people with both AD and cerebrovascular lesions have a more severe disease than people with AD alone (De Reuck J. Neurol Res Int. 2019;2019:7247325). To establish the usefulness of automated FDG-PET classification, transfer learning of advanced machine learning techniques was introduced. The use of machine learning, specifically transfer learning, to classify people's FDG-PET images by AD subtype is novel.The ResNet-18 convolutional network by transfer learning was chosen because it provides accurate model construction in a very short time (Rawat and Wang, Neural Comput. 2017;29(9):2352-2449). Note that any deep neural network architecture suitable for image classification can be used, for example, squeezenet (Iandola et al., arxiv.org / abs / 1602.07360), googlenet, inceptionv3 (Szegedy et al., Proc of IEEE conf comp vis pat recog, pp.1-9, 2015), densenet201 (Huang et al., CVPR vol. 1, no. 2, p / 3, 2017), resnet-50 or -101 (He et al., Proc IEEE conf comp vis pat rec, pp.770-778, 2016), efficientnetb0 (Migxing Tan and Quoc, Arxiv:1905.1194, 2019), alexnet (Krizhevsky et al., Adv neur info proc sys, 2012), vgg16 (Simonyan and Ziserman, arxiv:1409.1556, 2014). The ResNet-18 convolutional network by transfer learning can be distinguished into two subtypes, and the regions that contribute most to typical AD are represented in red on the occlusion sensitivity map (reference number 30 on Figure 3A). The machine learning model was trained on full images and did not receive any information regarding the region of interest (ROI) as input, but was able to pick out informative regions regarding the difference between AD and mixed AD. The average accuracy when distinguishing the two subtypes was found to be 97.5% after 5-fold cross-validation.The prediction accuracy in this study was quite high in comparison with some previous studies based on deep learning techniques, despite previous attempts to distinguish people with AD or mild cognitive impairment (MCI) from healthy controls (Korolev et al., arXiv:170106643 [cs]. Published online January 23, 2017; Lu et al., Sci Rep. 2018;8) (Table 4). The accuracy of the classification demonstrated here to distinguish patients with mixed dementia from those with AD is similar to the best accuracy obtained to distinguish patients with AD or cognitive impairment from healthy controls. Identifying subtypes of AD is a considerably harder problem than AD vs. control, or MCI vs. control, or even AD vs. MCI classification. Therefore, it is surprising that this much more difficult classification task could reach such a high level of accuracy with this method.

[0075] The overall accuracy of the test in distinguishing AD from controls was higher compared to the test in distinguishing MCI from controls. This means that most of the models are not very accurate in distinguishing subtle differences between images of groups with some form of memory impairment, i.e., the subtle differences between AD and MCI and cognitively healthy individuals. This classifier based on ResNet-18 and transfer learning can distinguish the typical and mixed pattern subtypes of AD with 97.5% accuracy, despite classifying two subgroups of the same disease.

[0076]

Table 4

[0077] This study is unprecedented in introducing deep learning to classify FDG-PET images from people with AD into two clinically important subtypes, "typical AD" and "mixed". The overlapping mixed pathology in people with AD poses a challenge for clinicians in diagnosis and patient management. The introduction of automated transfer learning techniques to classify patients into typical AD and mixed subtypes not only facilitates the accurate separation of a pure subset of people with typical AD from the mixed type but also offers the possibility of more accurate, reproducible, and faster diagnosis.

[0078] Conclusion This study is unprecedented in differentiating two imaging subtypes of AD via visual analysis of FDG-PET images and then demonstrating that the two subtypes are distinguishable from each other by semi-quantitative analysis and clinically different. Furthermore, transfer learning, a type of machine learning, was used to predict the two subtypes with high accuracy, sensitivity, and specificity. In the clinical setting, most images are analyzed via visual analysis by experts, which is costly, time-consuming, and has well-known intra- and inter-observer biases. Machine learning techniques such as transfer learning can overcome these drawbacks. At a pragmatic level, it can also meet the growing needs considering the combination of the challenges of the aging population and the global shortage of radiologists. The novel application of transfer learning using the pre-trained network ResNet-18 in the study has the potential to greatly improve efficiency and accuracy in differentiating FDG-PET images of people with typical AD from those with co-existing cerebral small vessel disease. Such an application of AI would be beneficial not only for accurate diagnosis and prognosis prediction of individual patients but also, importantly, for identifying appropriate patients for future clinical trials.

[0079] The systems and methods of the above embodiments can be implemented in a computer system (specifically, in computer hardware or computer software) in addition to the described structural components and user interactions.

[0080] The term "computer system" includes hardware, software, and data storage devices for embodying the system according to the embodiments described above or for performing the method. For example, a computer system may include one or more processing units, such as a central processing unit (CPU) and / or a graphics processing unit (GPU), input means, output means, and data storage. Preferably, the computer system has a monitor for providing a visual output display. The data storage may include RAM, a disk drive, or other computer-readable media. The computer system may include a plurality of computing devices connected by a network and capable of communicating with each other over the network. It is explicitly contemplated that the computer system may consist of or include a cloud computer.

[0081] The method of the above embodiments may be provided as a computer program or as a computer-readable medium carrying a computer program arranged to perform the method described above when executed by a computer or a computer program product.

[0082] The term "computer-readable medium" includes, without limitation, any one or more non-transitory media that are directly readable and accessible by a computer or a computer system. The media may include, without limitation, magnetic storage media, such as floppy disks, hard disk storage media, and magnetic tape, optical storage media, such as optical disks or CD-ROMs, electrical storage media, such as memory (including RAM, ROM, and flash memory), and hybrids and combinations of the above, such as magnetic / optical storage media.

[0083] The features disclosed in the above description, or in the following claims, or in the accompanying drawings, in their specific forms or by means for carrying out the functions of the present disclosure, or the methods or processes for obtaining the results of the present disclosure, can be used, as appropriate, separately or in any combination of such features, to implement the present invention in its various forms.

[0084] Although the present invention has been described in combination with the exemplary embodiments described above, many equivalent modifications and variations will be apparent to those skilled in the art when the present disclosure is provided. Therefore, the exemplary embodiments of the present invention shown above are considered to be illustrative and not limiting. Various changes to the described embodiments can be made without departing from the spirit and scope of the present invention.

[0085] It should be stated to avoid any doubt that any theoretical explanations provided in this specification are provided for the purpose of improving the reader's understanding. The inventors do not wish to be bound by any of these theoretical explanations.

[0086] Any section headings used in this specification are for structural purposes only and should not be construed as limiting the subject matter described.

[0087] Throughout this specification (including the following claims), unless the context requires otherwise, the words "comprise", "include" and their variations, such as "comprises", "comprising", "including", etc., are to be understood to imply the inclusion of the stated integers or steps or groups of integers or steps, but not the exclusion of any other integers or steps or groups of integers or steps.

[0088] As used in this specification and the appended claims, it should be noted that the singular forms "a", "an", and "the" include plural referents unless the context clearly dictates otherwise. Ranges may be expressed herein as from one particular value with "about" attached thereto and / or to another particular value with "about" attached thereto. When expressed in such a range, other embodiments include from this particular value and / or to other particular values. Similarly, when a value is expressed as an approximation, it will be understood by use of the prefixed "about" that this particular value forms another embodiment. The term "about" associated with a numerical value is optional and means, for example, ±10%.

[0089] As used herein, "and / or" should be regarded as a specific disclosure of each of two specifically recited features or components, with or without the other. For example, "A and / or B" should be regarded as a specific disclosure of each of (i) A, (ii) B, and (iii) A and B, as if each were individually recited.

Claims

**Claim 1** A method for diagnosing a dementia subtype in one or more patients, the method comprising: obtaining brain imaging data related to the one or more patients; and classifying the one or more patients among a plurality of classes including a first patient class having a first dementia subtype and a second patient class having a second dementia subtype by providing the brain imaging data related to the one or more patients as an input to a deep learning model trained using brain imaging data from a plurality of patients, wherein the brain imaging data includes a first image set showing evidence of temporoparietal hypometabolism, and a second image set showing evidence of hypometabolism in regions of the brain other than the temporal and parietal regions, instead of or in addition to the temporal and parietal regions, the first image set being labeled as being associated with the first dementia subtype, and the second image set being labeled as being associated with the second dementia subtype; A method as described above. **Claim 2** The method according to claim 1, wherein the dementia subtype is an Alzheimer's disease (AD) subtype, the first dementia subtype is AD, the second dementia subtype is AD in combination with cerebrovascular disease or vascular dementia, the second dementia type is cerebrovascular disease or vascular dementia, the second dementia subtype is mixed AD, and / or the second dementia subtype is not AD. **Claim 3** The method according to claim 1 or 2, wherein the classifying is between the first patient class having the first dementia subtype and the second patient class having the second dementia subtype. **Claim 4** The deep learning model is a deep neural network classifier and / or the deep learning model includes a convolutional neural network (CNN), the deep learning model includes a model pre-trained with unrelated image data and / or the deep learning model includes a CNN pre-trained using a deep residual learning framework; The method according to any one of claims 1 to 3. **Claim 5** The deep learning model includes all layers of a CNN pre-trained for image recognition other than the classification layer, and a classification layer trained using the first and second image sets and associated labels, and optionally the classification layer includes a fully connected layer and a softmax layer. The method according to any one of claims 1 to 4.

6. The images of the first set and the second set show evidence of different metabolic activities in any one or more or all of the right prefrontal cortex, left prefrontal cortex, right temporal cortex, left temporal cortex, right parietal cortex, left parietal cortex, left cerebellum, and right cerebellum. The method according to any one of claims 1 to 5.

7. The images of the second image set show evidence of hypometabolism in regions of the brain that include one or more or all of the right prefrontal cortex, left prefrontal cortex, left cerebellum, and right cerebellum instead of or in addition to one or more of the right temporal cortex, left temporal cortex, right parietal cortex, and left parietal cortex. The method according to any one of claims 1 to 6.

8. Hypometabolism means a lower glucose uptake rate, and / or blood flow, and / or FDG-PET-derived standardized uptake value ratio (SUVr) than expected for a control. The method according to any one of claims 1 to 7.

9. The brain imaging data is imaging data obtained using any functional brain imaging modality that provides information regarding the metabolic activity of the imaged brain region, and optionally the information regarding the metabolic activity of the imaged brain region is obtained by detecting glucose uptake by the imaged brain region and / or blood flow to the imaged brain region. The method according to any one of claims 1 to 8.

10. The brain imaging data is FDG-PET data or ASL data, and / or the brain imaging data is baseline functional brain imaging data. The method according to any one of claims 1 to 9.

11. Analyzing the data using a deep learning model includes analyzing a single section of each patient's brain imaging data, and / or the method includes selecting a single section of each patient's brain imaging data set. The method according to any one of claims 1 to 10, wherein optionally, the single section is a single-axis section at the thalamic level and / or the single section is a section including at least a part of the hippocampus and olfactory cortex.

12. The brain imaging data used to train the deep learning model includes a single section of the brain imaging data set of each of the plurality of patients, and the method includes selecting a single section of the brain imaging data set of each of the plurality of patients. The method according to any one of claims 1 to 11, wherein optionally, the single section is a single-axis section at the thalamic level and / or the single section is a section including at least a part of the hippocampus and olfactory cortex.

13. The first image set includes one or more images of each of at least 30, at least 40, at least 50, at most 500, at most 200, or at most 100 patients, and / or the second image set includes one or more images of each of at least 30, at least 40, at least 50, at most 500, at most 200, or at most 100 patients, and / or the first image set includes one or more images within 10% or 20% of the patients or for each of the plurality of patients, and one or more images thereof are included in the second image set, and / or the first and second image sets include one or more images of each of the plurality of patients, where the plurality of patients in the first and second image sets have the same age and gender. The method according to any one of claims 1 to 12.

14. By expert review of the first and second image sets, the first image set is labeled as being associated with a first dementia subtype, and the second image set is labeled as being associated with a second dementia subtype. The method according to any one of claims 1 to 13, wherein optionally, each image of the first image set and the second image set is assigned the same label by at least two experts.

15. The brain imaging data used to train the deep learning model includes images from a plurality of patients and images obtained from the images from the plurality of patients by image augmentation, and optionally, the image augmentation includes creating an inverted version of one or more of the images and / or creating a randomly rotated version of one or more of the images. The method according to any one of claims 1 to 14.

16. The deep learning model classifies the images of the first and second classes with an accuracy of at least 90% or at least 95%, and / or the deep learning model classifies the images of the first and second classes with a sensitivity of at least 90% or at least 94%, and / or the deep learning model classifies the images of the first and second classes with a specificity of at least 90% or at least 95%. The method according to any one of claims 1 to 15.

17. The method is to train the deep learning model using brain imaging data from a plurality of patients, where the brain imaging data includes a first set of images showing evidence of temporoparietal hypometabolism, and in addition to or instead of the temporal and parietal regions, a second set of images showing evidence of hypometabolism in brain regions other than the temporal and parietal regions, where the first set of images is labeled as being associated with a first subtype of dementia and the second set of images is labeled as being associated with a second subtype of dementia. The method according to any one of claims 1 to 16, further including training.

18. A method of providing a tool for diagnosing a dementia subtype in one or more patients, the method comprising: Obtaining training image data including a first set of images showing evidence of temporoparietal hypometabolism, and in addition to or instead of the temporal and parietal regions, a second set of images showing evidence of hypometabolism in brain regions other than the temporal and parietal regions, where the first set of images is labeled as being associated with a first subtype of dementia and the second set of images is labeled as being associated with a second subtype of dementia. Obtaining. Training a deep learning model to classify patients among a plurality of classes including a first patient class having the first dementia subtype and a second patient class having the second dementia subtype using the training image data. A method comprising the above. **Claim 19** A method of selecting a subject diagnosed as having AD or likely to have AD for participation in a clinical trial, the method comprising: Providing brain imaging data related to the subject as an input to a deep learning model trained using brain imaging data from a plurality of patients to classify the subject among a plurality of classes including a first patient class having a first dementia subtype and a second patient class having a second dementia subtype, wherein the brain imaging data includes a first image set showing evidence of temporoparietal hypometabolism and a second image set showing evidence of hypometabolism in brain regions other than the temporal and parietal regions, either instead of or in addition to the temporal and parietal regions, where the first image set is labeled as being associated with the first dementia subtype and the second image set is labeled as being associated with the second dementia subtype. Selecting or excluding the subject for participation in the clinical trial based on whether the subject is classified as having the first dementia subtype or the second dementia subtype. A method comprising the above. **Claim 20** A method of providing a prognosis for a subject diagnosed as having AD or likely to have AD, the method comprising Providing brain imaging data related to the subject as an input to a deep learning model trained using brain imaging data from multiple patients, thereby classifying the subject among multiple classes including a first patient class having a first dementia subtype and a second patient class having a second dementia subtype, wherein the brain imaging data includes a first image set showing evidence of temporoparietal hypometabolism, and a second image set showing evidence of hypometabolism in brain regions other than the temporal and parietal regions, instead of or in addition to the temporal and parietal regions, wherein the first image set is labeled as being associated with the first dementia subtype and the second image set is labeled as being associated with the second dementia subtype, the classifying; Determining the prognosis of the subject based on whether the subject is classified as having a first dementia subtype or is classified as having a second dementia subtype, optionally, determining the prognosis includes determining that the subject is more likely to have a faster rate of cognitive decline when the subject is classified into the second class than when the subject is classified into the first class, the determining; A method comprising.

21. A system for diagnosing a dementia subtype and / or for providing a tool for diagnosing a dementia subtype, the system comprising: one or more processors and a computer-readable memory storing instructions for causing the processor to perform the method according to any one of claims 1 to 18, optionally, the system further comprising data acquisition means configured to obtain brain imaging data related to one or more patients.

22. A non-transitory computer-readable storage medium including machine-executable instructions that, when executed by a processor, cause the processor to perform the method according to any one of claims 1 to 18.

23. A computer program including executable code that, when executed by a computer, causes the computer to perform the method according to any one of claims 1 to 18.

Images