CARE: Clinical AI Predictor for Posterior Urethal

Valves - Design, Explainability and Evaluation

Autori: Roberta De Fazio, Stefano Marrone, Paola Tirelli, Raffaele Chianese, Clelia Di Nardo, Pierluigi Marzuillo, Laura Verde

Anno: 2025

In the last decades, the remarkable impact achieved by Artificial Intelligence (AI) in business and industry has not been mirrored in critical real-world applications. The industrial diffusion of AI in healthcare is facing some resistance due to the lack of uniform legal frameworks and general scepticism among society and medical personnel. This paper proposes a multidisciplinary approach to fill the gap between the theoretical AI-based framework and real clinical practice, tailored to the problem of Posterior Urethral Valves (PUVs) diagnosis in paediatric patients. The multidisciplinary core of the work allows tackling the problem not only under the technical lens, but also from a clinical and industrial perspective: through the adoption of classifier composition mechanisms, this study presents the lessons learned in developing a reliable PUV classifier, as well as in its empirical assessment against real-world data and within a structured diagnostic process. The main contributions are threefold:  the support provided to a medical experts, which clinically evaluate the behaviour of the model, validating the rules extracted via explainability technique; the adoption of an extensive dataset that captures the complexity of real-world clinical pathways; and the design of a composition strategy enabling the development of complex AI classifiers that leverage the vertical training of specialised models, achieving an overall 70% accuracy.

S​copri di più