Enhancing Usability and Efficiency in AI-Assisted Breast Cancer Diagnosis: The Role of Thematic Analysis and User Testing

Artificial intelligence has shown strong potential to support breast cancer diagnosis, yet many AI-assisted systems struggle to achieve a seamless adoption in clinical practice due to poor usability, misaligned workflows, and limited clinician trust. In high-stakes diagnostic contexts, even technically accurate systems may fail if their interfaces do not reflect how clinicians think, work, and make decisions. This study investigates how Thematic Analysis (TA) can be systematically integrated with quantitative usability metrics to uncover latent usability issues in AI-assisted diagnostic interfaces that are often overlooked by conventional evaluation methods. <div> We reanalyzed qualitative and behavioral data from prior user testing studies involving radiologists and clinicians interacting with a multimodality breast cancer diagnosis interface, interview transcripts, observed interaction patterns, and task performance were coded into recurring usability themes using TA, these themes were then cross-referenced with quantitative measures of diagnostic accuracy (DOTS), usability (SUS), and cognitive workload (NASA-TLX) through a weighted prioritization method that links user-reported issues to measurable performance deviations, enabling this innovative approach to identify and prioritize usability problems based not only on frequency, but also on their actual impact on workflow efficiency and user performance. </div> <div> Derived from these findings, a high-fidelity prototype was developed to ad dress the most critical usability issues, with particular emphasis on interface adaptability. Comparative evaluation under equivalent testing conditions revealed substantial improvements in workflow efficiency, including a 65.7% reduction in lost time and a 76.2% decrease in image access time, together with lower cognitive load, increased perceived usability and trust. Addition ally, the redesigned system achieved near-ceiling usability scores while significantly reducing performance variability, with Mean Absolute Deviation and Inter Quartil Range decreasing by up to 85%. </div> <div> This work demonstrates that Thematic Analysis, when tightly coupled with quantitative usability data, can function as a robust, evidence-driven tool for guiding interface redesign in AI-assisted healthcare systems. Beyond breast cancer diagnosis, the proposed methodology offers a replicable framework for refining human-centered AI interfaces in medical imaging and other safety-critical domains, ensuring that AI systems adapt to clinical workflows rather than forcing clinicians to adapt to technology. </div>