A Hybrid Explainable Continual Learning Framework for Trustworthy Medical Image Classification Using Self-Supervised Representation Learning
IB Ranitha,
O Srinivas,
Gundala Swarnalatha
Artificial Intelligence (AI) has significantly transformed medical image analysis by enabling
automated diagnosis, disease classification, lesion detection, and clinical decision support across
diverse healthcare applications. Deep learning models have demonstrated remarkable performance in
analyzing radiological, pathological, ophthalmological, dermatological, and histopathological images.
However, conventional deep neural networks are generally trained under static learning assumptions
and frequently suffer from catastrophic forgetting when new medical datasets or disease categories
become available. Moreover, their dependence on large annotated datasets, limited explainability, and
lack of transparency reduce clinical trust and hinder widespread deployment in real-world healthcare
environments. Recent advances in Self-Supervised Learning (SSL), Continual Learning (CL), and
Explainable Artificial Intelligence (XAI) provide promising solutions for developing adaptive, data-
efficient, and trustworthy medical image analysis systems. This paper proposes a Hybrid Explainable
Continual Learning Framework for Trustworthy Medical Image Classification Using Self-Supervised
Representation Learning, integrating self-supervised feature learning, Vision Transformers, continual
learning with memory replay, contrastive representation learning, explainable artificial intelligence,
uncertainty estimation, and adaptive knowledge distillation into a unified intelligent medical imaging
framework. Initially, large-scale unlabeled medical images are utilized to learn robust feature
representations through self-supervised contrastive learning. Subsequently, continual learning enables
incremental acquisition of new disease categories while preserving previously acquired diagnostic
knowledge through adaptive memory replay and knowledge distillation. The proposed framework
establishes a scalable, trustworthy, and adaptive medical imaging platform suitable for intelligent
diagnosis across radiology, pathology, ophthalmology, dermatology, oncology, neurology, cardiology,
and future AI-assisted healthcare systems.
