Computer Assisted Diagnosis: does it make sense for Lungs Tumor Detection? A Qualitative study with Pulmonologists.

This study presents the evaluation by 10 pulmonologists of a diagnostic support system for lung cancer detection through optical endoscopy based on deep learning techniques. The evaluation relies on semi-structured interviews and System Usability Scale (SUS) tests to analyze the utility, reliability, and acceptability of this device during patient procedures. The results reveal a good SUS score, which is better than the vast majority of evaluated systems (88th percentile). Furthermore, the assessment highlights the absence of cognitive biases commonly associated with medical artificial intelligence (AI) systems. Our contribution is structured around three axes: the validation of an assistance system by clinical experts, the evaluation of potential biases during the use of AI tools in interventional pulmonology, and concrete recommendations for the design of medical diagnostic support interfaces. These findings emphasize the potential of assistance systems to improve early detection of lung cancer while preserving the decision-making autonomy of practitioners.