Evaluation of AI for prostate cancer detection in biparametric-MRI screening population data

Comprehensive Summary

A comparison between a deep-learning-based artificial intelligence (AI) system and the performance of radiologists has revealed that AI is capable of producing relatively significant screening information for prostate cancer detection. The AI system was trained using MRI data from prostate cancer screening populations with the goal of detecting significant prostate cancer indicated by a grade 2 or higher. Results from the study revealed an area under the receiver operating characteristic curve (AUROC) of 0.83 with a significantly lower specificity at matched sensitivity levels compared to that of radiologists.

Outcomes and Implications

Artificial intelligence (AI) has been advancing in its use for patient screening, and this study illustrates the potential for AI to be used to improve screening efficiency. Of note, the lower specificity of the deep-learning AI model in comparison to the performance of radiologists points towards the need for continued advancements of the technology; however, the use of AI in prostate cancer screening can be instrumental in supporting and strengthening radiologist performance as a diagnostic tool.