AI may help radiologists reduce missed breast cancer cases
A combined deep and machine learning approach trained to read mammograms and electronic health records (EHRs) may be used to greatly reduce missed breast cancer cases, according to a June 18 study published in Radiology.
The algorithm, trained on more than 9,000 mammograms and EHRs, predicted breast malignancy within one year of the index examination with 87% sensitivity, said Ayelet Akselrod-Ballin, PhD, with IBM Research’s Department of Healthcare Informatics, University of Haifa Campus in Israel, and colleagues.
“In a scenario where double reading at screening mammography is not available…we believe that the use of this model as a second reader could be beneficial,” Akselrod-Ballin et al. added. Such a model may even be used to recommend an individual breast cancer screening plan, the group noted.
The researchers set out to evaluate the performance of their machine learning, deep learning (ML-DL) model for early breast cancer prediction. They hypothesized such a model could perform similarly to radiologists and be accepted as a second reader in clinical practice.
A total of 52,936 images were gathered from 13,234 women who underwent at least one mammogram between 2013 and 2017. All had health records for at least one year prior to mammography. The algorithm was trained to predict biopsy malignancy and to differentiate normal from abnormal findings. It was validated and tested in 1,055 and 2,548 women, respectively.
When analyzing the test set, the ML-DL model identified 34 of 71 (48%) false-negative findings. For predicting malignancies it achieved an area under the receiver operating characteristic curve (AUC) of 0.91, specificity of 77.3% and sensitivity of 87%. The ML-DL notched an AUC of 0.85 for identifying normal exams.
Akselrod-Ballin pointed out that their model performed in the acceptable range of radiologists for breast cancer screening, but “the model did not perform better than radiologists, it performed differently.”
The study was not without limitations. For one, the ML-DL model was trained on images from a single healthcare system which uses only one vendor. The results, according to the authors, must be validated across various vendors, facilities and populations.
Down the line the model should incorporate some of the same tools radiologists use, such as previous mammography and ultrasound images to further improve its performance. But its early results are promising, the authors wrote.
“ML technology emphasizes the need for linking data sets from multiple modalities to improve the accuracy of breast cancer detection and save experts’ valuable time on high-probability healthy individuals,” the researchers concluded. “In particular, this model’s ability to lower false-negative results by half is of immediate clinical relevance.”
Matt joined Chicago’s TriMed team in 2018 covering all areas of health imaging after two years reporting on the hospital field. He holds a bachelor’s in English from UIC, and enjoys a good cup of coffee and an interesting documentary.