AI improves diagnostic performance in mammography


A team of researchers led by Ayelet Akselrod-Ballin, PhD, and Michal Chorev, PhD, of IBM Research trained a combined model of machine learning and deep learning that resulted in an area under the curve (AUC) of 0.91 to predict breast cancer. In addition, he identified cancer in nearly half of women who had a first negative mammogram but who were later diagnosed with biopsy cancer in a year.

Interestingly, the algorithm also determined that white blood cell profiles and thyroid function tests – results that are currently not considered in published risk models – have been associated with breast cancer.

"The algorithm, which combines approaches to machine learning and deep learning, can be applied to assess breast cancer at a level comparable to that of radiologists and has the potential to substantially reduce missed diagnoses of breast cancer," the authors wrote. .

The researchers retrospectively collected 52,936 images of 13,234 women who had received at least one mammogram between 2013 and 2017 at Assuta Medical Centers in Israel and had health records for at least one year prior to their mammogram examination. The team then used mammography and EHR data from 9,611 women to train the deep learning model to predict malignancy and differentiate normal screening exams from abnormal.

For the purposes of the study, women were considered carriers of cancer if they had a diagnosis after biopsy within 12 months after mammography. After validation in 1,055 women, the algorithm was then tested in the images and clinical information of 2,548 women with mean age of 55 years.

With specificity of 77.3% at a specimen 87% specificity operating point to predict a positive biopsy for cancer, the model results were well within the acceptable range of radiologists for screening mammography as defined by the sensitivity index of 75% of the Breast Cancer Surveillance Consortium. 88% to 95% specificity, according to the authors.

The researchers also emphasized that the model is performed differently – not better – than radiologists.

"In a scenario where dual reading on screening mammography is not available, as is the case at Assuta Medical Centers, we believe using this model as a second reader could be beneficial," the authors write.

The model does not use the same tools as radiologists to interpret mammograms, for example, incorporating comparisons with previous mammograms or using ultrasound imaging.

"We believe that, in the future, incorporating such data and additional clinical data, such as genetic information, can further improve the performance of the machine learning / deep learning model," the authors write. "Machine learning technology emphasizes the need to link multi-modal data sets to improve the accuracy of breast cancer detection and to save the valuable time of specialists in healthy high-probability individuals. model of reducing false negative results by half is of immediate clinical relevance. "

Copyright © 2019


Source link