A commercially available artificial intelligence (AI) used for mammography performed with sufficient diagnostic performance to act as an independent reader in prospective clinical studies.
These are the conclusions from a new study by a team from the Department of Oncology-Pathology, Karolinska Institute, Sweden, which sought to evaluate three different AI computer-aided detection (CAD) algorithms for mammography assessment without radiographer intervention.
The study included a sample of 8805 women of whom 739 had a diagnosis of breast cancer and a random sample of 8066 healthy controls. Each of the three CAD systems processed the mammogram images and yielded a prediction score for each breast based on the suspicion of cancer, ranging from 0 (no cancer) to 1 (high suspicion). The authors then compared the predictive accuracy of each CAD with those of a radiologist. The overall results across the three CAD algorithms showed a sensitivity of 86.7% and a specificity of 92.5%. However, the performance of one CAD algorithm surpassed the performance of radiologists and was better at detecting cancer than the other two. The highest performing CAD algorithm had an area under the curve of 0.956 for detecting cancer at screening or within 12 months thereafter.
In their conclusion, the authors noted that the best performing CAD was on a par with a radiologist and called for further evaluations of AI CAD systems as independent readers in mammography screening programs.
Reference
Salim M et al. External evaluation of 3 commercial artificial intelligence algorithms for independent assessment of screening mammograms. JAMA Oncol doi:10.1001/jamaoncol.2020.3321
