What are the intrinsic gaps between Artificial Intelligence (AI) researchers and Healthcare workers and how can we overcome them? Knowing these challenges and solutions is essential to develop medical AI within a clinical environment in 5 years; towards this, the workshop gathers 2 AI researchers in Medical Imaging, 2 physicians, and 3 generalists between Healthcare and Informatics, conducting 2 activities: a) Learning: Know the overview and status quo of Medical Imaging, along with its well-known challenges and solutions; b) Thinking: Find the intrinsic gaps and solutions between AI researchers and Healthcare workers, sharing the common and different thinking/working styles in each field.
Through this workshop, we identify that the main gaps consist in i) the clinical meaningfulness and interpretability of medical AI, especially in diagnosis, to persuade both physicians and patients; ii) the commerciality of medical AI to introduce it into hospitals; iii) the robustness of AI regardless of variance among scanners/individuals to trust computer-assisted diagnosis; iv) the acquisition of medical data and their annotation despite severe ethical reviews and changing diagnostic criteria in Japan.
For i), we can use Medical Imaging for alerting the risk of overlooking the diagnosis in a clinical environment, rather than replacing physicians, especially during medical examination. Regarding ii), medical machine manufacturers (e.g., X-ray) can introduce AI into their machines or hospitals can exploit AI to reduce staff. Concerning iii), we can generally prioritize sensitivity over specificity to alleviate overlooking the diagnosis and combine medical images with other medical data, such as white blood cell count. iv) can be alleviated by data augmentation and domain adaptation techniques, along with compiling a database of electronic medical records.