Economical hybrid novelty detection leveraging global aleatoric semantic uncertainty for enhanced MRI-based ACL tear diagnosis

Athanasios Siouras, Serafeim P. Moustakidis, George Chalatsis, Tuan Aqeel Bohoran, Michael Hantes, Marianna Vlychou, Sotiris K. Tasoulis, Archontis Giannakidis, Dimitrios Tsaopoulos

Abstract: Highlights•The formulation of the ACL tear diagnosis task as a novelty detection problem for the first time to address class imbalance.•The integration of rapidly-obtained aleatoric semantic uncertainty-based scores into the novelty detection framework.•The inclusion of global tissue information into the decision-making process.•Superior accuracy, resource efficiency and eco-friendly performance, when compared with two state-of-the-art methods.•Cross-database experiments verified the robustness and excellent generalization capabilities of the proposed framework.