Artificial Intelligence Improves Prediction of Breast-Cancer Risk

By Megan Brooks

December 23, 2019

NEW YORK (Reuters Health) - A deep-neural-network model developed by Swedish researchers can more accurately predict a woman's risk for breast cancer than a standard risk model based on mammographic density.

Compared with the best mammographic density model, the deep neural network "showed a higher risk association for breast cancer and made fewer mistakes among women with more aggressive cancer," Dr. Karin Dembrower of Karolinska Institute in Stockholm told Reuters Health by email.

"The addition of mammographic density has improved traditional breast cancer risk models. In our study, we examined whether a deep learning (DL) approach could extract further risk-relevant information from the images," Dr. Dembrower and colleagues note in Radiology.

They developed and trained their neural network using mammogram data from more than 2,200 women, 278 of whom were later diagnosed with breast cancer.

Women who developed breast cancer were significantly older at mammography (55.7 years vs. 54.6 years), had higher breast dense area (38.2 cm2 vs. 34.2 cm2) and higher percentage breast density (25.6% vs. 24.0%).

The DL risk score was better able to predict which women were at risk for future breast cancer (odds ratio, 1.56; area under the receiver operating characteristic curve, 0.65) compared with age-adjusted dense area (OR, 1.31; AUC, 0.60).

In addition, the false-negative rate for the DL risk score was lower than that for age-adjusted dense area, especially for aggressive cancers; the false-negative rate for lymph node-positive breast cancer was 31% for the DL risk score versus 42% with age-adjusted dense area.

"The next step," Dr. Dembrower told Reuters Health, "is to test the algorithm in a clinical prospective study, which we are planning for in Stockholm and to combine it with algorithms for computerized tumor detection."

In the future, she is "convinced that artificial intelligence (AI) tools will be introduced more clinically, both for risk prediction and for tumor detection. AI tools will probably contribute with a deeper analysis of the images as well as reducing the work load for breast radiologists."

The study had no commercial funding and the authors have disclosed no relevant conflicts of interest.

SOURCE: http://bit.ly/2rNoF2Z Radiology, online December 17, 2019.

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....