Studying the Manifold Structure of Alzheimer’s Disease: A Deep Learning Approach
Many classical machine learning techniques have been used to explore Alzheimer’s disease (AD), evolving from image decomposition techniques such as principal component analysis toward higher complexity, non-linear decomposition algorithms. With the arrival of the deep learning paradigm, it has become possible to extract high-level abstract features directly from MRI images that internally describe the distribution of data in low-dimensional manifolds. In this work, we try a new ex- ploratory data analysis of AD based on deep convolutional autoencoders. We aim at finding links between cognitive symptoms and the underlying neurodegeneration process by fusing the information of neuropsychological test outcomes, diagnoses, and other clinical data with the imaging features extracted solely via a data-driven decomposition of MRI. The distribution of the extracted features in different combinations is then analyzed and visualized using regression and classification analysis, and the influence of each coordinate of the autoencoder manifold over the brain is estimated. The imaging-derived markers could then predict clinical variables with correlations above 0.6 in the case of neuropsychological evaluation variables such as the MMSE or the ADAS11 scores, achieving a classification accuracy over 80% for the diagnosis of AD.
INTRO:
ALZHEIMER'S Disease (AD) is the most common type of neurodegenerative disease in the world, affecting more than 5% of the population in Europe and with an incidence of 11.08 per 1000 person-years. With a yet unknown etiology, a current diagnosis of AD often depends on the clinical history and the outcomes of widely extended neuropsychological tests such as the Mini-Mental State Exam (MMSE) that, according to re- cent studies may add confounding information to the procedure of diagnosis. Therefore, understanding the disease progression as well as studying and standardizing new disease markers is paramount.
Substantial advances in the technology used for neuroimaging can now track neurodegeneration even before the development of full-blown dementia [5]. However, the prodromal stages of AD are often confused with the cognitive decline associated with age or other diseases, in what is commonly known as Mild Cognitive Impairment (MCI). Longitudinal studies, such as the Alzheimer’s Disease Neuroimaging Initiative (ADNI) [6], the Open Access Series of Imaging Studies (OASIS) [7] or the Dominantly Inherited Alzheimer Network (DIAN) [8] combine Magnetic Resonance Imaging (MRI), biological and cognitive markers to study these stages and observe which of the MCI affected subjects to convert to AD.
These initiatives have largely contributed to the validation of computational analysis methodologies. Processing pipelines that combine spatial and intensity normalization and feature extraction, such as FreeSurfer [9] or Statistical Parametric Mapping (SPM) [10] are now commonplace. Structural and functional features are frequently used in differential diagnosis studies with great success and, together with the rise of Machine Learning (ML) methodologies, there are now complete systems that can recognize patterns associated with the diseases using automated feature extraction.
In addition to blood-based protein markers, blood-based microRNAs (miRNAs) have emerged as promising AD diagnosis markers (23–25). miRNAs are small (approximately 22 nucleotides), non-coding RNAs that primarily regulate gene ex- pression posttranscriptionally. Circulating miRNAs are remarkably stable, a characteristic that supports their use in the clinic. Advancements in modern genomics technologies such as multiplex quantitative reverse transcription PCR (RT-qPCR), microarray, and RNAseq have enabled miRNA- based biomarker discovery. Previous reports suggested that miRNAs are causally linked to AD by directly affecting the underlying pathogenic pathways. Owing to sample heterogeneity and pre-analytical and analytical variabilities, blood-based miRNA AD markers remain inconclusive.
MATERIALS AND METHODS:
Study design-
A cohort of 96 serum samples was obtained from the Oxford Project to Investigate Memory and Ageing (OPTIMA) study. Ethical approval was given to the Research Tissue Bank (OPTIMA) by the Research Ethics Committee (Oxfordshire RECC), reference number 09/H0606/70. Individuals who met the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer’s Disease and Related Disorders Association criteria for the diagnosis of probable AD were classified as AD patients. Individuals were diagnosed with mild cognitive impairment (MCI) according to the practice parameter described by Petersen et al. Controls were defined as elderly participants whose cognitive test scores were above the cutoffs for impairment and had remained stable over at least the previous 2 years. Clinical diagnosis of a subset in this cohort was confirmed by postmortem (PM) histologic examination of brain tissue with established Consortium to Establish a Registry for Alzheimer Disease criteria.
Sample collection:
Serum samples were collected at the University of Oxford clinical site. Fasting was not requested be- fore sampling. For serum preparation, venous blood was drawn into BD red top vacutainers by venipuncture from each volunteer. The blood tubes were incubated at room temperature for 1 h and subsequently centrifuged at 3000g for 5 min at 4 °C to separate serum and clot. Serum was transferred to a 5 mL polypropylene tube (Falcon, cat. no. 352063), spun at 3000g for 10 min at 4 °C, and then divided into 500 μL aliquots into 1 mL polypropylene cryotube (NUNC, cat. no. 366656). Serum aliquots were stored at −70 °C.
RESULTS:
Study setup and multiplex RT-qPCR profiling analysis of human serum miRNAs:
The goals of this study were to discover a blood miRNA signature that can discriminate AD from non-dementia controls by identifying and cross-validating a multimarker signature and to validate the prediction performance of the signature in a separate and blinded cohort.
Serum samples of 96 participants in the OPTIMA study were used in the biomarker signature training and test study. This OPTIMA cohort had 51 control participants, 32 AD patients, and 13 MCI participants. In the machine learning approach, AD and control participants were split 70:30 as the training and test cohorts. The training cohort consisted of 36 control participants and 22 AD patients, and the blinded cohort for prospective validation included 15 controls and 10 AD patients. To assess whether the constructed AD signature can discriminate MCI samples from the controls, all MCI participants were included in the test cohort but were assessed separately from the AD patients in the test cohort. Table 1 illustrates demographics for the training and test cohorts. The criteria used for choosing participants in the training cohort were age- and sex-matched AD and control participants as illustrated by the P values. Owing to the sample size limitation of the OPTIMA cohort, it was impossible to match age between AD and control in the test cohort, in which control participants were older than AD patients. Among the 96 participants, for those who went through PM histologic examination of brain tissue, 17 were confirmed controls, and 28 were confirmed, AD patients.
Circulating MicroRNA Signature for Alzheimer Disease:
ROC curves of the 12-miRNA signature in discrimination of AD patients from control participants in the training (black) and test (blue) cohorts. , Classification of AD patients and control participants by the 12-miRNA signature. ©, ROC curves of the 12-miRNA signature in discrimination of MCI from control participants in the training (black) and test (purple) cohorts. Classification of MCI and control participants by the 12-miRNA signature.
CONCLUSION:
In this article, we presented an integrated method for finding subregions of the hippocampus that were significant for dis- criminating between patients with AD and healthy control subjects and building effective classifiers based on these regional changes. The major advantage of the machine learning methods compared with the univariate method was that it could detect subtle and spatially complex deformation patterns of the hippocampus in patients with AD compared with healthy control subjects. The results were objective and reliable because the methods were validated by a permutation test, and the findings were consistent with previous studies.
In summary, the shape analysis methods presented in the article provided a useful tool for detecting regional differences of the subcortical structures. These methods can also be applied to other neuropsychiatric diseases.
