Evidence continues to pile up supporting the promise of artificial intelligence in diagnosing medical issues. We’ve written about numerous studies in which machine learning proved equal to or better than physicians diagnosticians. Researchers at Heidelberg University’s German Cancer Research Center found that a convolutional neural network out-performed certified histopathologists evaluating the malignancy of tissue images. IBM scientists trained a machine learning system that could diagnose diabetic retinopathy from retinal images in 20 seconds which otherwise can be a time-consuming, subjective process that requires expert clinicians with specialized training. We could go on and on with more examples of superior diagnostic performance with trained algorithms. AI specialists around the world are finding success training machines to analyze vast amounts of highly-specific diagnostic data with accuracy levels that match or beat expert clinicians.
Now we can add prostate cancer to the list. Researchers at Radboud University Medical Center in the Netherlands published a study in Lancet Oncology in which a deep learning system demonstrated superiority in determining the aggressiveness of prostate cancer. After training the system with images of 5,759 biopsies from more than 1,200 patients, the Radboud researchers compared the system’s accuracy with that of 15 pathologists. The clinicians represented various levels of experience from multiple countries. The AI system was more accurate in assessing prostate cancer aggressiveness than 10 of the pathologists and comparable to the most experienced physicians. The Radboud researchers pointed to the advantages of AI’s consistent performance independent of the presence of a pathologist.
Going forward, the Radboud researchers are working with peers from Sweden’s Karolinska Institute and Google subsidiary Kaggle. The group is organizing an international competition to further improve AI diagnostic performance. The rapidly expanding body of successful AI diagnostic studies and the deepening expertise of AI scientists suggest that the application of AI diagnostics in the real world is not far off.
