New AI Model Enhances Breast Cancer Detection on MRI with Clear, Interpretable Results
A new artificial intelligence model has been developed by experts to enhance the accuracy of breast cancer detection in MRI screenings.
What sets this model apart from earlier efforts is its foundation: it was trained on a large, diverse dataset and is designed to offer evidence supporting its decisions—an essential requirement for use in real clinical environments.
The research team behind this project believes their explainable AI approach could help lower the high rate of false positives that commonly occur in breast MRI scans.
AI-assisted MRI could potentially detect cancers that humans wouldn’t find otherwise. Previously developed models were trained on data of which 50% were cancer cases and 50% were normal cases, which is a very unrealistic distribution,” explained Felipe Oviedo, PhD, lead investigator and senior research analyst at Microsoft’s AI for Good Lab. “Those models haven’t been rigorously evaluated in low-prevalence cancer or screening populations (where 2% of all cases or less are cancer), and they also lack interpretability, both of which are essential for clinical adoption.
To overcome those limitations, Oviedo and his collaborators at the University of Washington’s Department of Radiology created a model designed to perform in both high- and low-prevalence cancer populations. They trained it using over 10,000 contrast-enhanced breast MRI scans collected at the university between 2005 and 2022. The dataset included a wide range of breast densities, allowing the AI to learn how to identify irregularities even in some of the most challenging imaging scenarios.
Unlike traditional binary classification models, our anomaly detection model learned a robust representation of benign cases to better identify abnormal malignancies, even if they are underrepresented in the training data,” said Oviedo. “Since malignancies can occur in multiple ways and are scarce in similar datasets, the type of anomaly detection model proposed in the study is a promising solution.
The AI system was tested against standard benchmark models and delivered superior results across both internal and external datasets. It performed well in identifying abnormalities in both common and rare cancer scenarios, producing heatmaps that pointed out suspicious areas. These visual explanations closely matched lesions confirmed by biopsy and substantially outperformed existing models in accurately detecting abnormal findings.
The research team envisions future clinical applications where the model acts as a triage tool for radiologists managing high exam volumes. By prioritizing potentially abnormal scans, it could help streamline workflows and improve diagnostic efficiency.
Our model provides an understandable, pixel-level explanation of what’s abnormal in a breast,” Oviedo said. “These anomaly heatmaps could highlight areas of potential concern, allowing radiologists to focus on those exams that are more likely to be cancer.