Lessons learned from COVID-19 medical imagining deep learning-based solutions: Towards clinically applicable models
The breakthroughs of artificial intelligence on image analysis in the last years provide equal or even higher accuracy than specialists in the area, opening a new world of possibilities. The automatization of tasks conducted by experts enables not only time-saving analysis but also tasks that would not be feasible otherwise. During the current pandemic, science and technology were used as powerful tools against the virus, including deep learning-based solutions for the prediction and prognosis of COVID-19 using chest medical imaging like X-ray and CT. Unfortunately, none of these models was clinically useful. This work explores the reasons behind such issues from both theoretical and experimental approaches. While these models excel in research papers, the recent transition of this technology into real clinical settings brings new challenges. Such problems derive from several factors, including their dataset origin, composition, and description, hampering their fairness and secure application. Considering the potential impact of incorrect predictions in applied-ML healthcare research is urgent. Undetected bias induced by inappropriate dataset usage and improper consideration of confounders prevents the translation of prediction models into clinical practice.

Leave a Reply

Your email address will not be published. Required fields are marked *