• Open Access close
• Other research products close
• English close

A Neurodegenerative Disease (ND) is progressive damage to brain neurons, which the human body cannot repair or replace. The well-known examples of such conditions are Dementia and Alzheimer’s Disease (AD), which affect millions of lives each year. Although conducting numerous researches, there are no effective treatments for the mentioned diseases today. However, early diagnosis is crucial in disease management. Diagnosing NDs is challenging for neurologists and requires years of training and experience. So, there has been a trend to harness the power of deep learning, including state-of-the-art Convolutional Neural Network (CNN), to assist doctors in diagnosing such conditions using brain scans. The CNN models lead to promising results comparable to experienced neurologists in their diagnosis. But, the advent of transformers in the Natural Language Processing (NLP) domain and their outstanding performance persuaded Computer Vision (CV) researchers to adapt them to solve various CV tasks in multiple areas, including the medical field. This research aims to develop Vision Transformer (ViT) models using Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset to classify NDs. More specifically, the models can classify three categories (Cognitively Normal (CN), Mild Cognitive Impairment (MCI), Alzheimer’s Disease (AD)) using brain Fluorodeoxyglucose (18F-FDG) Positron Emission Tomography (PET) scans. Also, we take advantage of Automated Anatomical Labeling (AAL) brain atlas and attention maps to develop explainable models. We propose three ViTs, the best of which obtains an accuracy of 82% on the test dataset with the help of transfer learning. Also, we encode the AAL brain atlas information into the best performing ViT, so the model outputs the predicted label, the most critical region in its prediction, and overlaid attention map on the input scan with the crucial areas highlighted. Furthermore, we develop two CNN models with 2D and 3D convolutional kernels as baselines to classify NDs, which achieve accuracy of 77% and 73%, respectively, on the test dataset. We also conduct a study to find out the importance of brain regions and their combinations in classifying NDs using ViTs and the AAL brain atlas. This thesis was awarded a prize of 50,000 SEK by Getinge Sterilization for projects within Health Innovation.

Recent advances in multi-atlas based algorithms address many of the previous limitations in model-based and probabilistic segmentation methods. However, at the label fusion stage, a majority of algorithms focus primarily on optimizing weight-maps associated with the atlas library based on a theoretical objective function that approximates the segmentation error. In contrast, we propose a novel method—Autocorrecting Walks over Localized Markov Random Fields (AWoL-MRF)—that aims at mimicking the sequential process of manual segmentation, which is the gold-standard for virtually all the segmentation methods. AWoL-MRF begins with a set of candidate labels generated by a multi-atlas segmentation pipeline as an initial label distribution and refines low confidence regions based on a localized Markov random field (L-MRF) model using a novel sequential inference process (walks). We show that AWoL-MRF produces state-of-the-art results with superior accuracy and robustness with a small atlas library compared to existing methods. We validate the proposed approach by performing hippocampal segmentations on three independent datasets: (1) Alzheimer's Disease Neuroimaging Database (ADNI); (2) First Episode Psychosis patient cohort; and (3) A cohort of preterm neonates scanned early in life and at term-equivalent age. We assess the improvement in the performance qualitatively as well as quantitatively by comparing AWoL-MRF with majority vote, STAPLE, and Joint Label Fusion methods. AWoL-MRF reaches a maximum accuracy of 0.881 (dataset 1), 0.897 (dataset 2), and 0.807 (dataset 3) based on Dice similarity coefficient metric, offering significant performance improvements with a smaller atlas library (< 10) over compared methods. We also evaluate the diagnostic utility of AWoL-MRF by analyzing the volume differences per disease category in the ADNI1: Complete Screening dataset. We have made the source code for AWoL-MRF public at: https://github.com/CobraLab/AWoL-MRF.

In healthy individuals, behavioral outcomes are highly associated with the variability on brain regional structure or neurochemical phenotypes. Similarly, in the context of neurodegenerative conditions, neuroimaging reveals that cognitive decline is linked to the magnitude of atrophy, neurochemical declines, or concentrations of abnormal protein aggregates across brain regions. However, modeling the effects of multiple regional abnormalities as determinants of cognitive decline at the voxel level remains largely unexplored by multimodal imaging research, given the high computational cost of estimating regression models for every single voxel from various imaging modalities. VoxelStats is a voxel-wise computational framework to overcome these computational limitations and to perform statistical operations on multiple scalar variables and imaging modalities at the voxel level. VoxelStats package has been developed in Matlab® and supports imaging formats such as Nifti-1, ANALYZE, and MINC v2. Prebuilt functions in VoxelStats enable the user to perform voxel-wise general and generalized linear models and mixed effect models with multiple volumetric covariates. Importantly, VoxelStats can recognize scalar values or image volumes as response variables and can accommodate volumetric statistical covariates as well as their interaction effects with other variables. Furthermore, this package includes built-in functionality to perform voxel-wise receiver operating characteristic analysis and paired and unpaired group contrast analysis. Validation of VoxelStats was conducted by comparing the linear regression functionality with existing toolboxes such as glim_image and RMINC. The validation results were identical to existing methods and the additional functionality was demonstrated by generating feature case assessments (t-statistics, odds ratio, and true positive rate maps). In summary, VoxelStats expands the current methods for multimodal imaging analysis by allowing the estimation of advanced regional association metrics at the voxel level.

Abstract: APOE ε4, the most significant genetic risk factor for Alzheimer disease (AD), may mask effects of other loci. We re-analyzed genome-wide association study (GWAS) data from the International Genomics of Alzheimers Project (IGAP) Consortium in APOE ε4+ (10 352 cases and 9207 controls) and APOE ε4− (7184 cases and 26 968 controls) subgroups as well as in the total sample testing for interaction between a single-nucleotide polymorphism (SNP) and APOE ε4 status. Suggestive associations (P<1 × 10-4) in stage 1 were evaluated in an independent sample (stage 2) containing 4203 subjects (APOE ε4+: 1250 cases and 536 controls; APOE ε4−: 718 cases and 1699 controls). Among APOE ε4− subjects, novel genome-wide significant (GWS) association was observed with 17 SNPs (all between KANSL1 and LRRC37A on chromosome 17 near MAPT) in a meta-analysis of the stage 1 and stage 2 data sets (best SNP, rs2732703, P=5·8 × 10−9). Conditional analysis revealed that rs2732703 accounted for association signals in the entire 100-kilobase region that includes MAPT. Except for previously identified AD loci showing stronger association in APOE ε4+ subjects (CR1 and CLU) or APOE ε4− subjects (MS4A6A/MS4A4A/MS4A6E), no other SNPs were significantly associated with AD in a specific APOE genotype subgroup. In addition, the finding in the stage 1 sample that AD risk is significantly influenced by the interaction of APOE with rs1595014 in TMEM106B (P=1·6 × 10−7) is noteworthy, because TMEM106B variants have previously been associated with risk of frontotemporal dementia. Expression quantitative trait locus analysis revealed that rs113986870, one of the GWS SNPs near rs2732703, is significantly associated with four KANSL1 probes that target transcription of the first translated exon and an untranslated exon in hippocampus (Pless than or equal to1.3 × 10-8), frontal cortex (Pless than or equal to1.3 × 10-9) and temporal cortex (Pless than or equal to1.2 × 10−11). Rs113986870 is also strongly associated with a MAPT probe that targets transcription of alternatively spliced exon 3 in frontal cortex (P=9.2 × 10−6) and temporal cortex (P=2.6 × 10−6). Our APOE-stratified GWAS is the first to show GWS association for AD with SNPs in the chromosome 17q21.31 region. Replication of this finding in independent samples is needed to verify that SNPs in this region have significantly stronger effects on AD risk in persons lacking APOE ε4 compared with persons carrying this allele, and if this is found to hold, further examination of this region and studies aimed at deciphering the mechanism(s) are warranted.

L’objectif de cette thèse était la validation de l’existence ainsi que la découverte de nouveaux sous-types au sein de la maladie d’Alzheimer, première cause de démence au monde. Afin d’explorer son hétérogénéité, nous avons employé des méthodes d’apprentissage profond appliquées à une modalité de neuroimagerie, l’imagerie par résonance magnétique structurelle.Cependant, la découverte de biais méthodologiques importants dans de nombreuses études de notre domaine, ainsi que l’absence de consensus de la communauté sur la manière d’interpréter les résultats des méthodes d’apprentissage profond a fait en partie dévier la thèse de son objectif principal pour s’orienter d’avantage vers des problématiques de validation, de robustesse et d’interprétabilité de l’apprentissage profond. Ainsi, trois études expérimentales ont été menées pour s’assurer de la capacité des réseaux profonds de correctement détecter la maladie. La première est une étude expérimentale de méthodes d’apprentissage profond pour la classification de la maladie d’Alzheimer et a permis d’établir une juste comparaison des méthodes. La seconde étude a permis de constater un manque de robustesse de la classification avec l’apprentissage profond en termes de motifs d’atrophie découverts à l’aide de méthodes d’interprétabilité. Enfin, la dernière étude propose une méthode de découverte de sous-types aidée par l’augmentation de données. Bien que fonctionnant sur des données synthétiques, celle-ci ne généralise pas aux données réelles.Une contribution majeure de la thèse est la librairie ClinicaDL, grâce à laquelle les résultats expérimentaux de la thèse ont été produits de manière à être reproductibles. The goal of this PhD was the validation of the existence and the discovery of new subtypes of Alzheimer’s disease, the first cause of dementia worldwide. Indeed, despite its discovery more than a century ago, this disease is still not well defined and existing treatments are only weakly effective, possibly because several phenotypes exist within the disease. In order to explore its heterogeneity, we employed deep learning methods applied to a neuroimaging modality, structural magnetic resonance imaging.However, the discovery of important methodological biases in many studies in our field, as well as the lack of consensus regarding deep learning interpretability, partly changed the main objective of the PhD to focus more on issues of validation, robustness and interpretability of deep learning. Then, to correctly assess the ability of deep learning to detect Alzheimer’s disease, three experimental studies were conducted. The first one is a study of deep learning methods for Alzheimer’s classification and allowed a fair comparison of the methods. The second study found a lack of robustness of classification with deep learning in terms of atrophy patterns discovered using interpretability methods. Finally, the last study proposed a subtype discovery method aided by data augmentation. Although it works on synthetic data, it does not generalize to real data.Experimental results of this PhD were obtained thanks to ClinicaDL, one major contribution of this PhD. It is an open source Python library that was used to improve the reproducibility of deep learning experiments.

Containing the medical data of millions of patients, clinical data warehouses (CDWs) represent a great opportunity to develop computational tools. Magnetic resonance images (MRIs) are particularly sensitive to patient movements during image acquisition, which will result in artefacts (blurring, ghosting and ringing) in the reconstructed image. As a result, a significant number of MRIs in CDWs are corrupted by these artefacts and may be unusable. Since their manual detection is impossible due to the large number of scans, it is necessary to develop tools to automatically exclude (or at least identify) images with motion in order to fully exploit CDWs. In this paper, we propose a novel transfer learning method for the automatic detection of motion in 3D T1-weighted brain MRI. The method consists of two steps: a pre-training on research data using synthetic motion, followed by a fine-tuning step to generalise our pre-trained model to clinical data, relying on the labelling of 4045 images. The objectives were both (1) to be able to exclude images with severe motion, (2) to detect mild motion artefacts. Our approach achieved excellent accuracy for the first objective with a balanced accuracy nearly similar to that of the annotators (balanced accuracy>80 %). However, for the second objective, the performance was weaker and substantially lower than that of human raters. Overall, our framework will be useful to take advantage of CDWs in medical imaging and highlight the importance of a clinical validation of models trained on research data.

Indexing and classification tools for Content Based Visual Information Retrieval (CBVIR) have been penetrating the universe of medical image analysis. They have been recently investigated for Alzheimer's disease (AD) diagnosis. This is a normal "knowledge diffusion" process, when methodologies developed for multimedia mining penetrate a new application area. The latter brings its own specificities requiring an adjustment of methodologies on the basis of domain knowledge. In this paper, we develop an automatic classification framework for AD recognition in structural Magnetic Resonance Images (MRI). The main contribu- tion of this work consists in considering visual features from the most involved region in AD (hippocampal area) and in using a late fusion to increase precision results. Our approach has been first evaluated on the baseline MR images of 218 subjects from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database and then tested on a 3T weighted contrast MRI obtained from a subsample of a large French epidemiological study : "Bordeaux dataset". The experimental re- sults show that our classification of patients with AD versus NC (Normal Control) subjects achieves the accuracies of 87% and 85% for ADNI subset and "Bordeaux dataset" respectively. For the most challenging group of subjects with the Mild Cognitive Impairment (MCI), we reach accuracies of 78.22% and 72.23% for MCI versus NC and MCI versus AD respectively on ADNI. The late fusion scheme improves classification results by 9% in average for these three categories. Results demonstrate very promising classification performance and simplicity compared to the state-of-the-art volumetric AD diagnosis methods

A variety of algorithms have been proposed for computer-aided diagnosis of dementia from anatomical MRI. These approaches achieve high accuracy when applied to research data sets but their performance on real-life clinical routine data has not been evaluated yet.The aim of this work was to study the performance of such approaches on clinical routine data, based on a hospital data warehouse, and to compare the results to those obtained on a research data set. The clinical data set was extracted from the hospital data warehouse of the Greater Paris area, which includes 39 different hospitals. The research set was composed of data from the Alzheimer's Disease Neuroimaging Initiative data set.In the clinical set, the population of interest was identified by exploiting the diagnostic codes from the 10th revision of the International Classification of Diseases that are assigned to each patient.We studied how the imbalance of the training sets, in terms of contrast agent injection and image quality, may bias the results. We demonstrated that computer-aided diagnosis performance was strongly biased upwards (over 17 percent points of balanced accuracy) by the confounders of image quality and contrast agent injection, a phenomenon known as the Clever Hans effect. When these biases were removed, the performance was very poor. In any case, the performance was considerably lower than on the research data set. Our study highlights that there are still considerable challenges for translating dementia CAD systems to clinical routine.