Abstract
Accurate classification of computed tomography (CT) images is essential fordiagnosis and treatment planning, but existing methods often struggle with thesubtle and spatially diverse nature of pathological features. Currentapproaches typically process images uniformly, limiting their ability to detectlocalized abnormalities that require focused analysis. We introduce UGPL, anuncertainty-guided progressive learning framework that performs aglobal-to-local analysis by first identifying regions of diagnostic ambiguityand then conducting detailed examination of these critical areas. Our approachemploys evidential deep learning to quantify predictive uncertainty, guidingthe extraction of informative patches through a non-maximum suppressionmechanism that maintains spatial diversity. This progressive refinementstrategy, combined with an adaptive fusion mechanism, enables UGPL to integrateboth contextual information and fine-grained details. Experiments across threeCT datasets demonstrate that UGPL consistently outperforms state-of-the-artmethods, achieving improvements of 3.29%, 2.46%, and 8.08% in accuracy forkidney abnormality, lung cancer, and COVID-19 detection, respectively. Ouranalysis shows that the uncertainty-guided component provides substantialbenefits, with performance dramatically increasing when the full progressivelearning pipeline is implemented. Our code is available at:https://github.com/shravan-18/UGPL