Outcome prediction in tumour therapy based on Dempster-Shafer theory

Outcome prediction plays a vital role in cancer treatment. It can help to update and optimize the treatment planning. In this paper, we aim to find discriminant features from both PET images and clinical characteristics, so as to predict the outcome of a treatment to adapt the therapy. As both information sources are imprecise, we propose a novel feature selection method based on Dempster-Shafer theory to tackle this problem. Then, a specific objective function with spar-sity constraint is developed to search for a feature subset that leads to increasing prediction performance and decreasing data imprecision simultaneously. Our approach was applied to two real data sets concerning to lung tumour et esophageal tumour, showing good performance.