AI System Outperforms Radiologists in Breast Cancer Prediction
Reduction seen in false positives, false negatives in datasets from the United Kingdom, United States
THURSDAY, Jan. 2, 2020 (HealthDay News) -- An artificial intelligence (AI) system can reduce false positives and false negatives in prediction of breast cancer and outperforms human readers, according to a study published online Jan. 1 in Nature.
Scott Mayer McKinney, from Google Health in Palo Alto, California, and colleagues examined the performance of an AI system for breast cancer prediction in a clinical setting. Data were curated from a large representative dataset from the United Kingdom and a large enriched dataset from the United States.
The researchers observed an absolute reduction of 5.7 and 1.2 percent in false positives in the U.S. and U.K. datasets, respectively, and 9.4 and 2.7 percent, respectively, in false negatives. The system was also able to generalize from the United Kingdom to the United States. The AI system outperformed six human readers in an independent study involving radiologists; the area under the receiver operating characteristic curve was greater for the AI system than the average radiologist (absolute margin, 11.5 percent). The AI system maintained noninferior performance in a simulation in which the AI system participated in the double-reading process that is used in the United Kingdom and reduced the workload of the second reader by 88 percent.
"These analyses highlight the potential of this technology to deliver screening results in a sustainable manner despite workforce shortages in countries such as the United Kingdom," the authors write.
Several authors disclosed financial ties to technology companies, including Google, which funded the study.