The potential of AI to reduce interval cancer in a middle income country breast cancer screening program

AIM AND OBJECTIVE
To evaluate the performance of an artificial intelligence (AI) system in detecting interval cancers among women with negative screening exams attending a national screening program of a middle-income country.

MATERIALS AND METHOD
This retrospective study was performed with a data cohort between 2016 and 2019 derived from women in the age of 40 to 69 who attended the national breast cancer screening program. During this period, the average recall rate was 5.3%. We collected the negative screening exams of 406 women who developed interval cancers before the next screening round and the negative screening exams of 446 women with normal follow-up. The pathological outcome and time-to-diagnosis were retrieved. All mammograms were processed by an AI cancer detection system (Transpara, ScreenPoint Medical), assigning a score between 1 and 10 to the exam, which represented an increasing likelihood of malignancy. The performance of Transpara for detection of interval cancers on negative screening exams was estimated in terms of the area under the receiver operating characteristic curve (AUC), and sensitivity at 90% and 95.0% specificity, with 95% confidence for interval cancers.

RESULTS
More than half of all interval cancers (53%) were flagged with the highest Transpara score 10. The AUC of Transpara to detect signs of interval cancer on negative screening exams was 0.80 (95% CI = 0.77-0.83). The sensitivity was 53.7% and 38.5% at specificity of 90% and 95.0%, respectively. The highest performance of Transpara was found for cases that were diagnosed within 6 months after screening (AUC: 0.85 (95% CI = 0.78-0.92)), compared to cases diagnosed within 24 months after screening (AUC: 0.74 (95% CI = 0.67-0.81)).

CONCLUSION
Transpara has the potential to reduce the rate of interval cancers, when applied as a second or third independent reader within a national breast cancer screening program.