A recent study published in Radiology reports that researchers have created a novel, interpretable model to predict five-year breast cancer risk using mammograms.
Architecture comparison of AsymMirai (left) and Mirai (right). Both models feed the four screening views into the same convolutional neural network (CNN) layers, but reasoning diverges thereafter. AsymMirai has fewer computational layers and instead calculates differences in the latent features, as shown by heat maps in the craniocaudal (CC) asymmetry and mediolateral oblique (MLO) asymmetry steps. AsymMirai then finds the prediction window containing the highest differences for each view, represented by red boxes in the Get Prediction Window step. The maximum feature differences within these windows are averaged to create a risk score. AHL = additive hazard layer. Image Credit: RSNA 2024
According to the American Cancer Society, one in eight women, or around 13% of all women in the United States, will get invasive breast cancer at some point in their lives, and one in 39 women (3%) will pass away from the disease.
Regularly scheduled mammograms can significantly decrease the chance of breast cancer-related death. It is still unknown, though, how to precisely predict which women will have breast cancer based solely on screening.
Modern deep learning-based algorithms like Mirai have shown promise as a tool for predicting breast cancer, but since little is known about how the algorithm makes its decisions, radiologists could misuse them and make inaccurate diagnoses.
Mirai is a black box—a very large and complex neural network, similar in construction to ChatGPT—and no one knew how it made its decisions. We developed an interpretable AI method that allows us to predict breast cancer from mammograms 1 to 5 years in advance. AsymMirai is much simpler and much easier to understand than Mirai.
Jon Donnelly, BS, Study Lead Author and PhD Student, Department of Computer Science, Duke University
In the study, Donnelly and associates compared Mirai’s one- to five-year breast cancer risk predictions with their recently created AsymMirai deep learning model, which is based on mammography. AsymMirai was developed using the “front end” deep learning part of Mirai, substituting an interpretable module—local bilateral dissimilarity—for the remainder of the complex technique.
Donnelly added, “Previously, differences between the left and right breast tissue were used only to help detect cancer, not to predict it in advance. We discovered that Mirai uses comparisons between the left and right sides, which is how we were able to design a substantially simpler network that also performs comparisons between the sides.”
Findings May Affect Mammogram Frequency
The 210,067 mammograms from 81,824 patients in the EMory BrEast imaging Dataset (EMBED) from January 2013 to December 2020 were compared using AsymMirai and Mirai models by the researchers for this study. The researchers discovered that for predicting breast cancer risk over a one to five-year period, their simplified deep learning model outperformed the cutting-edge Mirai.
The findings further emphasized the clinical significance of breast asymmetry and, thus, drew attention to the possibility of using bilateral dissimilarity as an imaging marker for breast cancer risk in the future.
According to Donnelly, AsymMirai’s predictions have an understandable logic, making it a potentially useful tool to assist human radiologists in breast cancer diagnosis and risk assessment.
Donnelly concluded, “We can, with surprisingly high accuracy, predict whether a woman will develop cancer in the next 1 to 5 years based solely on localized differences between her left and right breast tissue. This could have a public impact because it could, in the not-too-distant future, affect how often women receive mammograms.”
Journal Reference:
Donnelly, J., et. al. (2024) AsymMirai: Interpretable Mammography-based Deep Learning Model for 1–5-year Breast Cancer Risk Prediction. Radiology. doi:10.1148/radiol.232780