Ravi Shankar, PhD, MTech, BTech | Alexandra Hospital (NUHS); Amartya Mukhopadhyay, FRCP, NUHS; Effie Chew, FAMS, NUHS;
Anjali Bundele, MPH, NUHS
Competition Sponsor: Ministry of Health and National Research Foundation of Singapore
Awardee year: 2024
This project aims to develop an automated system for early detection and longitudinal monitoring of cognitive impairment in stroke survivors using natural language processing (NLP) and machine learning techniques. The system will analyze spontaneous speech samples collected through an intelligent virtual agent to predict Montreal Cognitive Assessment (MoCA) scores. Key components include advanced automatic speech recognition models using transfer learning and fine-tuned foundational models, multimodal feature extraction incorporating linguistic, acoustic, and clinical metadata, and machine learning models (random forests, SVMs, deep neural networks) for MoCA score prediction and cognitive impairment detection. The system will also incorporate longitudinal tracking capabilities to monitor cognitive decline over time. A pilot study with 20-30 stroke survivors will validate the system against standard cognitive assessments, with performance evaluated using correlation coefficients, ROC curves, and classification metrics. The project aims to achieve a minimum correlation of 0.7 between predicted and actual MoCA scores and an AUC of 0.8 for detecting cognitive impairment. If successful, this non-invasive, objective approach could enable earlier intervention and more efficient utilization of healthcare resources for managing cognitive decline in stroke survivors.