Abstract
This paper proposes an Incremental Learning (IL) approach to enhance theaccuracy and efficiency of deep learning models in analyzing T2-weighted (T2w)MRI medical images prostate cancer detection using the PI-CAI dataset. We usedmultiple health centers' artificial intelligence and radiology data, focused ondifferent tasks that looked at prostate cancer detection using MRI (PI-CAI). Weutilized Knowledge Distillation (KD), as it employs generated images from pasttasks to guide the training of models for subsequent tasks. The approachyielded improved performance and faster convergence of the models. Todemonstrate the versatility and robustness of our approach, we evaluated it onthe PI-CAI dataset, a diverse set of medical imaging modalities including OCTand PathMNIST, and the benchmark continual learning dataset CIFAR-10. Ourresults indicate that KD can be a promising technique for IL in medical imageanalysis in which data is sourced from individual health centers and thestorage of large datasets is not feasible. By using generated images from priortasks, our method enables the model to retain and apply previously acquiredknowledge without direct access to the original data.