Semi-supervised Outcome Prediction for a Type of Human Brain Tumour Using Partially Labeled MRS Information

摘要

The diagnosis and prognosis of human brain tumours, especially when they are aggresive, are sensitive clinical tasks that usually require non-invasive measurement techniques. Outcome information for aggressive tumours, in particular, is usually scarce. In this paper, we aim to gauge the capability of a novel semi-supervised model, SS-Geo-GTM, to infer outcome stages from a very limited amount of available stage labels and Magnetic Resonance Spectroscopy (MRS) data corresponding to Glioblastoma, which is an aggressive tumor type. This model stems from a geodesic distance-based extension of Generative Topographic Mapping (Geo-GTM) that prioritizes neighbourhood relationships along a generated manifold embedded in the observed data space.