Statistical analysis to identify MRI markers for Alzheimer’s Disease: why moving towards AI?

Abstract

AbstractBackgroundAlzheimer’s disease (AD) is characterized by progressive neurodegeneration of some brain structures and cognitive function decline. To evaluate the anatomical changes, structural Magnetic Resonance Imaging (sMRI) is a non‐invasive examination and allows the quantification of brain atrophy. Some of the affected structures could be common with other dementia types such as frontotemporal dementia (FTD) and dementia with Lewy bodies (DLB). The objective of this work is to identify specific affected structures in patients with AD to healthy individuals, and to patients suffering from FTD and DLB. This is a part of a larger effort to develop methods to better diagnose AD and to find specific markers of this dementia type.MethodThe ADDIA clinical study (EU‐H2020 project led by Firalis company) enrolled 800 participants and was designed to obtain high‐quality samples for the discovery of novel biomarkers for early and differential diagnosis of AD. The sMRI of 623 participants with the above‐mentioned pathologies and healthy individuals were analyzed and the volume of different areas measured with the software FreeSurfer. Statistics were performed to define significant differences in brain volumes between groups.ResultOne hundred and twelve brain areas were analyzed for volume by study participant. Compared with a healthy population, the atrophy of structures observed in AD patients almost completely overlaps with the one observed in DLB and FTD patients. When the DLB and FTD groups are compared to a healthy population, one brain area is specific in DLB patients and fifty‐eight in FTD patients. However, on these fifty‐eight affected brain areas, forty‐seven overlap with the mild AD patient group.ConclusionThe measurement from imaging modality is not enough precise to report biomarkers as affected areas specific to the type of dementia. However, a promising framework lies in the application of machine and deep learning algorithms on a set of modalities: imaging, clinical and omics. These new tools allow for a finer understanding of data and a personalized approach to diagnosis. In preparation for this multimodal integration framework, the above categories of algorithms will be applied on MRI images. This will permit to discriminate AD from other dementia types at the individual level.