Small Data Learning for Alzheimer's Disease: From Digital Biomarker to Personalized Intervention

Professor Yu, Sau Fung Doris   (Principal Investigator (PI))

Xing Guoliang   (Co-Investigator)

36

2022-06-30

832000

Small Data Learning for Alzheimer's Disease: From Digital Biomarker to Personalized Intervention

""biomarkers"", ""dementia"", ""expressed emotion"", ""similarity-based federated learning"", ""neuro-psychiatric symptoms""

Artificial Intelligence and Machine learning

C4034-21G

Collaborative Research Fund (CRF) - Group Research Project 2021/2022

2022

On-going

A new federated few-shot learning framework that trains a deep learning model from limited, unbalanced, and multi-modal data in a privacy-preserving and distributed manner, while achieving high accuracy in detecting a diverse set of digital biomarkers.2. A scalable federated learning approach that achieves high model accuracy and low communication overhead by exploiting data similarity of different subjects and layer sharing of different models dynamically.3. A contrastive learning approach that can generate positive and negative training samples for biomarker detection by exploiting the fusion of different sensor modalities.4. Beyond biomarker detection, a generative adversarial network (GAN) algorithm is proposed to generate a personalized caregiving plan for each patient, by integrating multidimensional digital biomarkers and the domain knowledge from caregiving practice, including nutrition recommendations, exercise training study protocol, and advice from doctors.5. An interpretable deep learning framework, which enables interpretation of the correlations between the generated intervention plan and multi- dimensional biomarkers, providing professionals and caregivers interpretable and trusted guidelines for AD early diagonosis and intervention.6. Two clinical trials with a cohort of 200 subjects will be conducted to validate the proposed technologies, by working closely with clinical AD practitioners and a large public hospital in Hong Kong.7. Leveraging on our current collaboration with global industrial/academic partners and the Alzheimer's Drug Discovery Foundation (https://www.alzdiscovery.org), our team aims to establish a world-class research center on AI for Alzheimer's to advance the state of the art in research and treatment.