Scientists from the Radiological Society of North America have taught artificial intelligence to effectively detect breast cancer in representatives of different races. The results are published on the society's website (RSNA).
Traditional models of breast cancer risk assessment involve analyzing information obtained from patients' medical records. To improve diagnostic efficiency, scientists have developed new AI algorithms based on scanning images generated during a mammogram, a non-invasive breast examination.
The scientists noted that some of the widely used AI models used to diagnose this type of cancer have been developed predominantly for Caucasian women. Because of this, it is impossible to correctly assess the risk of developing breast cancer, with dark-skinned women showing the lowest relative survival rates. The new technology will eliminate this disparity.
The machine learning involved 129340 mammogram images, 106839 of which were obtained by examining patients with white skin color, 6154 - dark-skinned, 6435 - women of Asian origin, and 6257 - representatives of other nationalities.
The researchers emphasized that the new model outperformed traditional cancer prediction techniques in terms of performance. The tests showed that the algorithms can detect ductal and invasive carcinoma (a type of malignant tumor that develops from epithelial tissue cells) with high accuracy.