Design and Evaluation of a Prototype for a Platform for AI algorithms in Medical Imaging : A Human-Centered Approach

Artificial intelligence promises to facilitate radiological tasks, like segmentation. However, the quality of interaction experienced by humans is minimally investigated. In this study, a multiphase work consisting of three research phases is presented to 1. assess the quality of interaction between humans and AI in medical imaging, to 2. propose a new implementation based on human-centered design methodology and to 3. assess the quality of interaction between humans and the new prototype. In the first phase, a usability test (n=14) assessing the interaction between professionals and a segmentation AI model, embedded in open-source software, revealed that the implementation of the AI algorithm led to several usability problems (16 identified) and low satisfaction in use (50.2% (SD:10.88)). In the second phase, human-centered design methods were applied to develop a novel prototype for integrating medical imaging AI tools. In the third phase, a usability assessment of the new prototype with clinicians and other regular users of medical imaging software (n=13) showed that the new prototype seems to facilitate good usability with less usability problems (10 identified) and good satisfaction in use (89,6% (SD:4.55). Qualitative data were collected and analyzed for further feedback on the design of the prototype, and final suggestions for the further development of the new platform were provided.