Towards automated lymphoma prognosis based on PET images

This paper proposes a simple method to identify candidate tumors in a set of Positron Emission Tomography (PET) images obtained from patients suffering from lymphoma, and then extract statistics from the image most active tumor. These statistics are used as input for three machine learning algorithms, which generate models for overall survival and event-free survival. The results obtained by these methods are better than the ones obtained by visual analysis, and competitive or better than the ones obtained by a quantitative measure of prognosis. Besides, the results indicate that there is a lot of redundant information coming from the images, and only 2 out of 10 attributes might be enough to predict prognosis.