Enhancing Mammogram Accuracy with AI Hybrid Strategy

A recent study has demonstrated that combining artificial intelligence (AI) with radiologist assessments in a hybrid reading approach significantly improves the efficiency of breast cancer screening while maintaining accuracy. Developed by Dutch researchers and tested on over 40,000 mammogram exams, this strategy leverages AI's ability to detect malignancies with high confidence, thereby reducing radiologist workload by approximately 38%. The core of this approach involves AI not only providing a probability of malignancy (PoM) for each case but also quantifying its certainty through an uncertainty score. When AI predictions are confident, radiologists can opt to forego further review, streamlining the process. Conversely, for uncertain cases, radiologists undertake double reading to ensure accuracy. The study focused on the entropy of the mean PoM score of the most suspicious region, which yielded comparable cancer detection and recall rates to traditional double reading. Results showed that when the AI model was certain, its performance was comparable or superior to double radiologist readings, with an area under the curve (AUC) of 0.96 versus 0.87, and a sensitivity close to 89%. This indicates that AI can be trusted for a substantial portion of cases, especially when its confidence exceeds a predetermined threshold. The researchers emphasize that incorporating uncertainty metrics into AI models is crucial for safe and effective deployment in clinical settings, especially as it can facilitate healthier workflows and reduce the burden on radiologists. The findings suggest that in the near future, AI could handle a significant portion of normal assessments independently, with radiologists focusing on more complex cases, ultimately advancing breast cancer screening protocols while conserving medical resources. Further prospective studies are recommended to evaluate how such strategies could impact overall screening efficiency and outcomes.