An innovative AI model developed by Microsoft’s AI for Good Lab is enhancing breast cancer detection on MRI scans, outperforming traditional methods and potentially transforming diagnostic procedures.
A new AI model has been developed to revolutionize breast cancer detection on MRI scans, significantly enhancing the accuracy in identifying tumor locations compared to existing benchmark models. This innovative tool, trained on a large dataset of nearly 10,000 breast MRI exams, was revealed in a study published today in the journal Radiology.
“AI-assisted MRI could potentially detect cancers that humans wouldn’t find otherwise,” lead investigator Felipe Oviedo, a senior research analyst at Microsoft’s AI for Good Lab, said in a news release.
Current Challenges in Breast Cancer Screening
While screening mammography remains the standard of care in breast cancer detection, its effectiveness diminishes among patients with dense breast tissue. Dense breasts not only increase the risk of cancer but also complicate tumor detection through mammography.
To address this, physicians often recommend breast MRI as a supplementary screening tool, especially for women with dense breasts or those at high risk for cancer.
“MRI is more sensitive than mammography,” added Oviedo. “But it’s also more expensive and has a higher false-positive rate.”
Enhancing Accuracy With AI
In light of these challenges, Oviedo’s team collaborated with clinical investigators from the University of Washington’s Department of Radiology to develop an AI model capable of anomaly detection.
This model excels in distinguishing between normal and abnormal data, helping to flag potential malignancies for further investigation. Traditional models have struggled in this area, often built on unrealistic data distributions that fail to perform well in low-prevalence cancer populations.
“Previously developed models were trained on data of which 50% were cancer cases and 50% were normal cases, which is a very unrealistic distribution,” Oviedo added. “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.”
AI Model Development and Testing
The innovative model was trained using data from nearly 10,000 contrast-enhanced breast MRI exams spanning 17 years. Approximately 80% of the patients were white, with 42.9% having heterogeneously dense breasts and 11.6% having extremely dense breasts.
Unlike binary classification models, this anomaly detection tool learns from a robust representation of benign cases, enhancing its ability to flag abnormal malignancies even when underrepresented during training.
“Our model provides an understandable, pixel-level explanation of what’s abnormal in a breast,” added Oviedo. “These anomaly heatmaps could highlight areas of potential concern, allowing radiologists to focus on those exams that are more likely to be cancer.”
The AI model underwent rigorous testing on internal and external datasets, including pre-treatment breast MRI exams from multiple centers. The anomaly detection model accurately pinpointed tumor locations and outperformed existing benchmark models across different validation groups and detection tasks.
Implications for Clinical Practice
If integrated into radiology workflows, this AI tool could significantly enhance the efficiency of breast MRI screenings by excluding normal scans for triage purposes and focusing on potentially abnormal ones. This could streamline the diagnostic process, allowing radiologists to prioritize more likely cancer cases.
A Path Forward
Before clinical adoption, the model requires evaluation on larger datasets and prospective studies to fully ascertain its potential in enhancing radiologists’ workflow and potentially saving lives through more accurate early detection of breast cancer.