“Artificial intelligence will not replace radiologists. Yet, those radiologists who use AI will replace the ones who don’t.” – Curt Langlotz, MD
Machine Learning (ML) is about to change Radiology workflow. By simply escalating exams that might have critical findings, the TAT for positive cases will significantly drop. Longer term, more advanced ML systems will be able to generate a preliminary report, or perhaps even a final report. Whether quality goes up or down will depend upon how we integrate ML into the workflow.
ML does have some potential to improve quality. One big win will be a reduction of our reliance on the human visual system. The human visual system is fundamentally limited because it can only perceive 40 or 50 shades of gray. On top of that, humans have a much easier time perceiving moving objects. Color moving objects with sharp contrast are even better. Our eyes were not designed to look at flat, static gray scale images, but that is exactly what Radiologists do most of their day.
In addition to our visual limits, humans have a hard time consistently performing cognitive interpretation. The Medical Image Perception Society (MIPS) has been studying the limitations of using human beings to interpret radiological exams for many years. From the MIPS web site, “The medical image perception discipline seeks an improved understanding of the perceptual factors that underlie the creation and interpretation of medical images.” Image perception research has shown that Radiologist accuracy and efficiency can be linked to the time of day, human factors, ambient lighting, fatigue, search errors, decision errors, display problems, Satisfaction of Search (SOS) errors, and various perception and psychology related issues.
Given the limits of the human visual system and complexity of human cognition systems, it seems very likely that there are times when ML could improve radiologist performance. If done correctly, we can use this technology to improve efficiency while simultaneously improving quality.
Don’t torture yourself by resisting, submit to Machine Learning.