Prospects for Artificial Intelligence-Based Pathological Diagnosis of Renal Transplant Biopsy

Artificial intelligence, initiated in the 1950s, has matured after two setbacks into deep neural networks (DNNs) and large language models (LLMs). Key drivers of success were the adoption of nonlinear models and autonomous learning. DNNs function as discriminative models that hierarchically interpret features to classify images, while LLMs, as generative models trained on vast datasets, generate text based on contextual meaning.

Digital pathology (DP) employing DNNs and LLMs is advancing globally, yet Japan lags behind. To standardize diagnostic interpretation and reduce the workload of pathologists, integration of DP into renal transplant pathology (RTP) is essential. The CAMELYON16 challenge demonstrated that AI can achieve diagnostic accuracy comparable to or surpassing expert pathologists. In the USA, over ten DP systems have been approved by the FDA as class II medical devices for primary diagnosis. Moreover, US law assigns liability for AI-related misdiagnosis jointly to pathologists and institutions, promoting both accuracy and legal protection for pathologists. In 2019, the Banff Digital Pathology Working Group was established to build a pathology repository, share AI algorithms, and foster model standardization through competitions. With numerous AI systems emerging, DP platforms should evolve in parallel with biennial Banff classification updates.

AI in RTP can enhance diagnostic objectivity and alleviate pathologists' workload. Linking Banff updates with AI retraining enables continuously updated, globally standardized DP. Advanced DP requires close collaboration between transplant pathologists, AI engineers, and cutting-edge graphics processing unit resources.