Pelvic neRves autOmatic Segmentation using hybrId Trustworthy AI – PROSITAI

Personalised medicine in surgery is based on the development of 3D, precise and patient-specific models of the anatomical structures and pathologies. These individual models are of uttermost importance for pre-operative planning and per-operative guidance. They are highly developed and perform well for dense structures (e.g., bones, organs, or tumors), but very few works have been dedicated to the peripheral nervous system, and in particular to the pelvic nerves. Nerve imaging from diffusion MRI and reconstruction using tractography are mainly applied to the brain and the central nervous system. However, there is a need to plan complex surgical interventions in the pelvic region, where nerve damage can cause significant complications, like genito-urinary dysfunctions. Currently, no licensed software nor research works propose an automatic solution for modelling and segmenting the pelvic nerves. The goal of this interdisciplinary project is to provide an ensemble of trustworthy algorithms based on hybrid AI to automatically segment, model and statistically analyse the most important pelvic nerves in children and adults for a nerve-sparing pelvic surgery. The proposed methods will be: 1) robust to tumors and malformations and to different imaging protocols and scanners; 2) frugal/efficient, to be trained with small datasets and tested on low computational power computers; 3) explainable, to be accepted by clinicians and gain their confidence. We will focus on three applications: 1) pathological pediatric surgery in the pelvic region (e.g., pelvic tumors and congenital malformations), 2) complex gynaecologic cancer surgery (ovarian and uterine cancers) and 3) endometriosis, a chronic gynaecological disease. All methods will be gathered into the first, and only software, able to automatically model and visualise in 3D the entire pelvic anatomy, including all pelvic nerves. It will be user-friendly and conceived with and for clinicians.