A recent study published in Radiology has found that a deep learning model based on mammograms can identify more breast cancers during screening MRI exams compared to traditional risk assessment models. The study, led by Dr. Leslie Lamb from Massachusetts General Hospital, showed that the deep learning model had higher cancer detection rates and positive predictive values (PPVs) in high-risk patients compared to the Tyrer-Cuzick and National Cancer Institute Breast Cancer Risk Assessment Tool (BCRAT) models.
Risk assessment models like Tyrer-Cuzick and BCRAT are commonly used to determine if women undergoing breast cancer screening should have supplemental MRI exams. However, these models have limited accuracy in predicting breast cancer. To assess the diagnostic accuracy of the deep learning model, the researchers compared it to the traditional risk models in patients who underwent supplemental screening with MRI.
The deep learning model predicts a patient’s five-year risk of developing breast cancer using standard 2D mammographic views. These images are processed through an image encoder and an aggregation module to combine information across the views. Based on this, the model predicts a patient’s traditional risk factors. The model was trained on a large dataset and showed promising results in predicting breast cancer risk.
The study included 2,168 women with a median age of 54 who underwent 4,247 high-risk screening MRI exams. The researchers found that the deep-learning model had higher cancer detection rates and PPVs compared to the traditional risk models. For example, the deep learning model had a significantly higher cancer detection rate of 20.6 per 1,000 exams, compared to 6 and 6.8 per 1,000 exams for Tyrer-Cuzick and BCRAT, respectively.
The findings of this study support the use of a deep learning model in clinical practice for breast cancer screening. The researchers believe that with further validation and prospective studies, this deep learning model could be widely adopted. Guidelines supporting its use in clinical practice are also anticipated.
In an accompanying editorial, Dr. Min Sun Bae from Korea University Ansan Hospital emphasized the need for more prospective studies to support personalized breast cancer screening and to address questions about the clinical use of deep learning models. The information contained in mammographic images itself can be valuable in improving screening performance, and screening frequency and supplemental imaging can be personalized based on individual breast cancer risk.
Source: Radiology (no URL was provided)
