Clinical Utility of AINAFHIC: AI-Guided Navigation for Hunner's Lesion and Interstitial Cystitis

Chun-Hsien WuTaiwan Moderator

Nobutaka ShimizuJapan Speaker Clinical Utility of AINAFHIC: AI-Guided Navigation for Hunner's Lesion and Interstitial CystitisBackground: Hunner lesion (HL)-type interstitial cystitis (IC) is a distinct subtype of IC/BPS characterized by epithelial denudation and submucosal inflammation. However, endoscopic detection is highly operator-dependent, with reported detection rates ranging from 5% to 57%. To enhance diagnostic consistency, we developed AINAFHIC (AI Navigation for Hunner and IC), a deep-learning–based system designed to assist in HL detection using cystoscopic images under white light imaging (WLI) and narrow band imaging (NBI). Methods: A total of 6,230 cystoscopic images (WLI, 2,238; NBI, 3,992) were retrospectively extracted from the video recordings of 103 patients with IC/BPS. The images were annotated by an expert urologist based on the definition of ESSIC-HL. The AINAFHIC was developed using a Cascade Mask R-CNN framework to detect HL, non-HL mucosal changes, and artifacts such as air bubbles. The models were trained separately for WLI and NBI images. Results: The AINAFHIC demonstrated an HL detection accuracy of over 90% for WLI and 67% for NBI. Clinical case analysis revealed improved identification of subtle HLs missed during visual inspection. Conclusions: AINAFHIC facilitates objective, high-accuracy detection of Hunner’s lesions from standard cystoscopic videos. This tool holds promise for standardizing HL diagnosis and supporting tailored treatment decisions in patients with IC/BPS. Future directions include multi-institutional validation and development of real-time AI-guided cystoscopy.