Image Analysis and Machine Learning Tools for Understanding Large Biomedical Data
This proposal will develop methods for applying, adapting, and extending major recent advances in the fields of image processing, computer vision, and machine learning to make progress in biomedical analysis. We propose to develop and validate algorithms for processing images and data from other modalities, acquired in other parts of the project. Then, through computer vision and machine learning techniques, we propose to integrate the results of the processing with background domain knowledge to obtain constructs for biomedical understanding valuable to those other parts. We thus propose to help extract new knowledge from given raw (multimodal, multivariate, multidimensional, and/or serial/bath) data. We propose to develop tools to first extract structural primitives or data segments, and their joint structure, as the basic representation of the data. We then propose to fuse the diverse, redundant information present in the representation to derive problem-specific high-level understanding, overcoming the challenges of volume and complexity using machine-learning algorithms.