IC‐P‐170: A SIGNATURE OF BRAIN ATROPHY HIGHLY PREDICTIVE OF PROGRESSION TO ALZHEIMER'S DEMENTIA AND COGNITIVE DECLINE

Many works have aimed at finding biomarkers in order to predict future progression to Alzheimer's dementia (AD) in individuals with mild cognitive impairment (MCI). However, this problem is challenging given that AD is a heterogeneous disorder. We present a brain signature of grey matter atrophy shared by a subset of patients with MCI that is highly predictive of progression to AD within 3 years. We used baseline T1 scans from 1136 subjects (controls, MCI, AD) from the ADNI1 and ADNI2 cohorts to generate individual voxel-based morphometry maps with SPM. Hierarchical clustering was used to reduce the number of features by identifying 7 subgroups of subjects, within ADNI1 AD and controls, based on similarity of atrophy. We derived spatial correlations between each individual's map to the mean of each subgroup, termed subtype weights, and used these subtype weights, along with age, gender, mean grey matter volume, and total intracranial volume, as features in our support vector machine to classify controls and AD from ADNI1. Then we trained a logistic regression classifier to identify AD patients with high-confidence predictions and obtain a highly predictive signature of AD (HPS+) in ADNI1 (Figure 1). After we optimized our hyperparameters on stable and progressor MCI (sMCI and pMCI respectively) from ADNI1, we applied our model to the ADNI2 dataset to classify 1) AD vs controls and 2) sMCI vs pMCI. Seven subtypes of grey matter atrophy and their coefficients, along with sex, age, mean grey matter volume (GMV) and total intracranial volume (TIV), for the resulting model to predict AD patients with high specificity. We achieved at least 90% specificity in identifying AD and pMCI patients in each of the training and replication samples. HPS+ subjects were more likely to be progressors. HPS+ subjects also had significantly worse baseline cognition and declined at a faster rate, compared to low-confidence subjects (Non-HPS+) and subjects who were flagged as negative (Figure 2). However, HPS+ subjects did not differ in amyloid or tau burden or APOE4 positivity. Characteristics of MCI subjects with a highly predictive brain signature of progression to Alzheimer's dementia (HPS+), MCI subjects with a low-confidence prediction (Non-HPS+), and MCI subjects who were not flagged as hits (Negative) in a) ADNI1 and b) ADNI2. ADAS13: Alzheimer Disease Assessment Scale - Cognitive. Significant differences are denoted by * for p <0.05 and ** for p< 0.001. With a new machine learning algorithm, we found an anatomical MRI brain signature that is highly predictive of future conversion to AD dementia in a subset of MCI patients. The results of this work may be developed into methods for earlier diagnosis and enrichment in clinical trials.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.