Artificial Intelligence for Thyroid Histopathology Image Analysis: A Systematic Review

Gui Tao, Razali Yaakob and Sina Abdipoor

Digital pathology and whole-slide imaging have created new opportunities for computer-assisted thyroid histopathology; however, the high resolution of slides, sparsity of diagnostic regions and limited public datasets continue to make this task challenging. This systematic review highlights recent advances in artificial intelligence for thyroid histopathology within the broader context of whole-slide image analysis. The field shows a clear methodological shift from convolutional neural network-based patch approaches to weakly supervised multiple instance learning, transformer-based slide modeling, hybrid architectures and emerging paradigms such as self-supervised learning, foundation models and multimodal learning. Despite these developments, limitations in dataset availability, evaluation consistency and computational efficiency remain significant barriers. Issues of cross-institution generalization, interpretability and clinical trust are still unresolved. Overall, the evidence suggests a transition from local patch-level analysis toward more scalable, context-aware slide-level reasoning. Future progress is likely to depend not only on improved model architectures but also on richer datasets, standardized evaluation protocols and more rigorous validation frameworks.

View Fulltext Back