NCI Study Examines Artificial Intelligence (AI) Versus Radiologists in Assessing Tumors
If your work hinges on accurate and rapid radiologist reports, you’ll be interested in this latest study from a team of NIH scientists, led by Dr. Baris Turkbey of NCI’s Center for Cancer Research.
When comparing AI results with those of radiologists, the researchers found that AI excelled in some areas but fell short in others. For example, in measuring the size (or volume) of cancer tumors using non-invasive imaging technologies, such as scans and whole slide images, the researchers found:
- AI was particularly useful in identifying and annotating localized cancers. Moreover, AI was able to perform these tasks very quickly, with reproducible results.
- radiologists could identify the extent of the tumor better than AI, especially on whole slide images, but this process remains tedious and time consuming.
What’s the best role for AI at this time? According to Dr. Turkbey, a senior author on the study, “Our findings show that this particular AI model is best suited as an adjunct to the radiologist rather than a standalone solution. This would allow radiologists to focus on complex cases that require a more critical assessment.” “The model was able to rapidly and consistently distinguish cases that needed further investigation, making it ideal for initial screenings, particularly in settings with high volumes of work and limited resources,” he added.
Read the full report, “Evaluating deep learning and radiologist performance in volumetric prostate cancer analysis with biparametric MRI and histopathologically mapped slides,” in Abdominal Radiology.