In health care, diagnostic imaging of different kinds/modalities contains complementary information about the disease. Multi-modality image data fusion through machine learning provides the opportunity for early detection of the disease. Early detection and early intervention provide the best hope for treating or managing the disease. In Alzheimer’s Disease (AD), structural and functional neuroimaging of various types such as MRI and PET have been found to have complementary value for early detection of AD. This project develops machine learning algorithms to fuse multi-modality images together with other clinical assessment datasets, specifically addressing the challenging issues such as modality-wise missing data, high-dimensionality, and heterogeneity, to facilitate early detection and prognosis of AD.
Methodologies:
- Incomplete-modality transfer learning
- Multitask/reverse-multitask machine learning & deep learning