Abstract
Anal incontinence is common after childbirth, often due to structural anal sphincter injury, affecting up to 42% of postpartum women, with 6% experiencing stool loss. Exo-anal ultrasound aids in detecting anal sphincter injury through tomographic ultrasound imaging (TUI), but manual TUI sequence extraction is expert-dependent and time-intensive. In this international multi-centre study, we developed and independently evaluated the performance of an interactive AI-assisted pipeline for automated TUI positioning and alignment with real-time adjustability, from exo-anal ultrasound images acquired in postpartum, pelvic floor clinic and antenatal clinical settings from a single ultrasound vendor ecosystem. The pipeline was developed using 245 cases (117 in-house, 128 external) and independently validated on 125 cases from pelvic floor dysfunction and postpartum clinics across multiple geographical locations, including UZ Leuven, an industrial dataset, and The Institute for the Care of Mother and Child, Prague. The anal canal region was localised using nnUnet. The pipeline achieved angular rotation accuracy within the range of expert manual extraction, reduced workflow time of exo-anal image analysis by threefold, and maintained visual clinical acceptability, enhancing standardisation for clinical users. Its interactive elements align with Food and Drug Administration and the European Union AI act guidelines for human oversight by fostering clinical trust in AI adoption in pelvic floor imaging and mitigating clinically relevant risks, such as false external anal sphincter defect detection due to cranial misplacement. The pipeline is designed to support workflow optimisation and does not perform automated diagnosis. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1038/s41598-026-39665-3.