Multimodal Classification

With increasing availabity of medical data to researchers, the potential for combining different data modalities for a patient has greatly increased. Preliminary investigations have been undertaken for using partial Least Squares (PLS) to classify Alzheimer's disease patients. Datasets used were T1-images, FA-images, and SNP genotype data that each contains different information for the same patient. The results of such multi-modal classification suggested that PLS is suitable means of handling the problem, though the incorporation of SNPs and FA data did not increase classification accuracy greatly, suggesting the SNPs used may not correspond to the image features for Alzheimer's differentiation. The main outcome of the research was that PLS is successful in identifying medically relevant areas of the brain for Alzheimer's patients - the figure below shows the areas of the brain selected by PLS as significant for classification.

significance maps of partial least squares for T1 brain images significance maps of partial least squares for FA brain images
Last updated 2014 | School of Computer Science