New AI Model Improves Breast Cancer Detection on Mammography
According to a study published in Radiology: Artificial Intelligence, the use of AI technology can aid radiologists in reading breast cancer screening mammograms. Developers have trained the AI on existing images and made it possible to detect abnormalities associated with cancer and distinguish them from benign findings. Additionally, the model can be tested on a variety of images. It will help not only in better cancer detection but also in improving efficiency for radiologists.
Earlier, breast cancer with mammography has shown to improve prognosis by detecting the disease at an earlier stage. However, many cancers can be missed out by the process, and suspicious findings often turn out to be benign. Hence, AI-based detection will assist in improving the accuracy of digital mammography. Earlier research from Radiology showed that, on average, about 10 percent of women recalled from screening for further medical workup based on unusual results are eventually found to have cancer.
For the research, scientists used MammoScreen, a mammography AI tool, to improve cancer diagnosis. The AI method recognizes areas that are suspicious and their risk of malignancy in 2D digital mammograms. The method uses the full collection of four views that make up a mammogram to generate a set of image positions that provide a corresponding suspicion rating.
A total of 240 2D digital mammography photographs from 2013 to 2016, containing multiple forms of anomalies, were examined by 14 radiologists. Half of the dataset was conducted in a first and second session without AI and with AI respectively.
When using AI assistance, the average cancer sensitivity slightly improved. AI managed to minimize the rate of false-negative or normalized outcomes despite the prevalence of cancer. Moreover, this new development in the detection of breast cancer worked without prolonging radiologists’ workflow.
In the second reading portion, reading time decreased in cases of a low risk of malignancy. This reduced read time could improve the productivity of the radiologist so that they can concentrate on the more suspicious assessments, the researchers said. They are planning to investigate the AI’s ability to detect breast cancer in the early stages and its behaviors and actions on a large screening-based population.
Serena Pacilè, Ph.D., clinical research manager at Therapixel, where the model was developed said that the findings show that the AI model increases the efficiency of radiologists to diagnose breast cancer. Dr. Pacilè also said that MammoScreen was cleared by the U.S. Food and Drug Administration for use in the clinic in March where it could help to minimize workloads of radiologists.