BME PhD Proposal Seminar: Mahesh B. Nagarajan
Improving the Performance of Computer Aided Diagnosis (CADx) in Automated Pattern Classification on Medical Images
Supervised by Professor Axel Wismueller
Computer Aided Diagnosis (CADx) can assist radiologists by extracting clinically relevant findings from suspicious patterns on medical images in a consistent manner. Most CADx systems are implemented as image pattern processing pipelines with three components - feature extraction (pattern description), feature reduction (selection of relevant features) and pattern classification. This proposal presents two specific aims to improve the performance of CADx at classifying healthy and pathological patterns. The first aim is to improve the representation of patterns with topological and geometrical features using Minkowski Functionals and the Scaling Index Method. Such methods are expected to better represent complex patterns annotated on medical images as opposed to currently used statistical approaches. The second aim is to evaluate the impact of feature reduction through dimensionality reduction and feature selection on the CADx classification performance. Such methods can identify and exclude features that are either correlated or irrelevant to the classification task. These proposed enhancements will be investigated using two medical datasets:
- The classification of small diagnostically challenging lesions on dynamic contrast-enhanced breast MRI.
- The classification of healthy and osteoarthritic cartilage on phase contrast CT images of knee patella.