Computer Sciences and Information Technology
Title : | Explainable artificial intelligence methods for studying structural variations in ageing brain |
Area of research : | Computer Sciences and Information Technology |
Focus area : | Explainable AI, Brain Ageing |
Principal Investigator : | Dr. Vaanathi Sundaresan, Indian Institute Of Science, Bangalore, Karnataka |
Timeline Start Year : | 2024 |
Timeline End Year : | 2026 |
Contact info : | vaanathi@iisc.ac.in |
Details
Executive Summary : | Neurodegenerative and cerebrovascular diseases are linked to brain atrophy, cognitive impairment, and high fatality rates, such as stroke. Common neuroimaging biomarkers include small vessel disease signs and stroke lesions. Accurate and reliable automated detection and quantification are crucial for understanding their role in neurological diseases and differential diagnosis. Deep learning methods like convolutional neural networks (CNN) have shown improvement in detecting neuroimaging biomarkers, but their lack of explainability is a major drawback. To develop fair, accountable, and trustworthy models for biomarker detection, there is a need for robust explainable AI methods that provide insights about features without compromising predictive performance across multiple datasets. This work aims to develop explainable AI methods that provide information regarding feature relevance for improving the predictive performance of deep learning models. The work will use various explainable methods to obtain visual feature saliency maps for tasks like biomarker detection and brain age prediction, and develop algorithms for model-based uncertainty quantification. Integrating uncertainty quantification techniques within domain-agnostic privacy protection frameworks, such as Federated learning, will ensure consistent high predictive accuracy across various datasets. The proposed method will be evaluated on multiple datasets, particularly with demographic characteristics specific to the Indian population. Various lesion-/image-level performance metrics will be used to compare the performance of the proposed AI method with existing state-of-the-art methods and evaluate their domain-agnostic predictive performance. The tools developed from this project will be open-source. |
Total Budget (INR): | 31,15,810 |
Organizations involved