Artificial intelligence-enhanced interpretation of kidney transplant biopsy: focus on rejection.

Abstract

Purpose of review: The objective of this review is to provide an update on the application of artificial intelligence (AI) for the histological interpretation of kidney transplant biopsies.

Recent findings: AI, particularly convolutional neural networks (CNNs), has demonstrated great potential in accurately identifying kidney structures, detecting abnormalities, and diagnosing rejection with improved objectivity and reproducibility. Key advancements include the segmentation of kidney compartments for accurate assessment and the detection of inflammatory cells to aid in rejection classification. Development of decision support tools like the Banff Automation System and iBox for predicting long-term allograft failure have also been made possible through AI techniques. Challenges in AI implementation include the need for rigorous evaluation and validation studies, computational resource requirements and energy consumption concerns, and regulatory hurdles. Data protection regulations and Food and Drug Administration (FDA) approval represent such entry barriers. Future directions involve the integration of AI of histopathology with other modalities, such as clinical laboratory and molecular data. Development of more efficient CNN architectures could be possible through the exploration of self-supervised and graph neural network approaches.

Summary: The field is progressing towards an automated Banff Classification system, with potential for significant improvements in diagnostic processes and patient care.