AI-derived MRI indices aid early differential diagnosis of three cognitive disorders

Recent research carried out at the Chinese University of Hong Kong (CUHK) has demonstrated high sensitivity and specificity of artificial intelligence (AI)–derived MRI indices for detection of three early-stage cognitive disorders, namely, Alzheimer’s disease (AD), dementia with Lewy bodies (DLB) and frontotemporal dementia (FTD).

Since 2019, several CUHK teams in collaboration with other Chinese universities have been assessing MRI scans of older adults to validate various AI-derived MRI indices. A total of 2,212 MRIs of seniors with AD (n=635), DLB (n=61), FTD (n=97), or normal cognition (n=1,328) were evaluated by 2021, which produced three distinct pieces of research.

“Slight early-stage brain atrophy is hard to detect by visual inspection, while computerized volumetric technology [CVT] detects the exact volume of different brain regions to the closest 0.01 mL,” said Professor Winnie Chu of the Department of Imaging and Interventional Radiology, CUHK. “CVT is an automatic process that takes about 10 minutes.”

“It is sometimes challenging for clinicians to differentiate DLB from normal ageing and AD, particularly at the early stage,” wrote the authors of the most recently presented MRI research. “Since structural differences in multiple regions exist between DLB, normal elderly and AD brains, we aimed to derive a single index [DLB-resemblance atrophy index (DLB-RAI)] by combining these multiple MRI features with machine learning and evaluate its performance in differentiating mild DLB from normal brain ageing and AD.”

The whole-brain structural difference analysis of DLB, AD and normal ageing MRIs validated DLB-RAI, demonstrating an optimal sensitivity of 0.83 and specificity of 0.86, with the authors noting its potential in the clinical practice. [Liu et al, Alzheimer’s Association International Conference 2021]

“We investigated the performance of AD-RAI in detecting preclinical and prodromal AD among mildly cognitively impaired [MCI] and cognitively unimpaired [CU] subjects and compared its diagnostic performance with that of hippocampal measure,” wrote the authors of one of the other papers. [Aging (Albany NY) 2021;13:13496-13514] “AD-RAI achieved the best metrics among all subjects [sensitivity, 0.74; specificity, 0.91; accuracy, 85.94 percent] and among MCI subjects [sensitivity, 0.92; specificity, 0.81; accuracy, 0.86] in detecting prodromal AD over other measures. However, hippocampal volume achieved a higher sensitivity [0.73] than AD-RAI [0.47] in detecting preclinical AD.”

“Subjects with AD-RAI ≥0.5 should receive either PET brain or cerebrospinal fluid analysis for final confirmation, while subjects with AD-RAI <0.5 are 91 percent unlikely to have AD,” explained one of the authors, Dr Lisa Au of the Division of Neurology, CUHK.

A third piece of research evaluated a novel MRI index, the frontotemporal dementia index (FTDI), derived from the ratio of the weighted sum of the volumetric indices in FTD- against AD-dominant structures. “FTDI showed excellent accuracy in differentiating FTD from AD, with a sensitivity of 0.96 and specificity of 0.70,” reported the researchers. [Alzheimers Res Ther 2021;13:23]

“With the advent of disease-specific treatments such as anti-amyloid therapy, multi-domain lifestyle-based intervention, and nutritional therapy, progression of AD can be potentially slowed down when detected at an early stage. Thus, a more convenient and precise screening method for detection of early AD will be very important,” stated co-author of all three studies, Professor Vincent Mok of the Division of Neurology, CUHK.