Recent research indicates that using artificial intelligence (AI) to reevaluate breast MRIs deemed high-risk could markedly enhance early breast cancer detection. A retrospective study published in Academic Radiology examined the effectiveness of a convolutional neural network trained on breast MRI data to predict breast cancer development within a year following initially negative MRI results. The study involved 3,029 MRI scans from 910 patients (average age 52), including 115 cases where breast cancer was later diagnosed on MRI. Initially, all patients received a BI-RADS assessment of less than 3.
The study found that the AI model predicted breast cancer within one year with a 72 percent accuracy as measured by the area under the receiver operating characteristic curve (AUC). Of the initial MRI scans, 83 out of 115 cases showed visual correlations with biopsy-proven breast cancer, with the majority of these correlations measuring less than 0.5 cm.
The researchers emphasized that using the AI model to reassess MRIs identified as high-risk could improve early breast cancer detection rates by up to 30 percent. "It's crucial to recognize that the reported 30% sensitivity refers to cancers that would likely have remained undetected until the subsequent exam. These are additional detections, supplementing those already identified with high sensitivity by radiologists," explained lead study author Dr. Lukas Hirsch of the City College of New York.
The study also highlighted that with the AI model’s reassessment, the positive predictive value (PPV) could reach 6 percent for high-risk MRI scans. If radiologists were to recall only half of these cases, they would achieve the PPV standard for tissue diagnosis and detect at least an additional 15% of tumors, representing a clinically significant improvement.
Furthermore, the AI model successfully pinpointed the future location of breast cancer in 57 percent of the cases, with a 71 percent accuracy rate in the 35 true positive cases.
Despite the promising results, the study’s authors acknowledged several limitations, including the single-center, retrospective nature of the study, the assessment limited to sagittal scans, and the small number of cancers detected through screening.
_________
RadParts, a TTG Imaging Solutions Company, is the world’s largest independent distributor of OEM replacement parts. We specialize in low-cost parts for repairing linear accelerators and radiation equipment. Our mission is to provide high-quality, user-friendly, low-cost components and support for linear accelerators and radiation equipment. Contact RadParts at 877-704-3838 to learn more.
Written by the Digital Marketing Team at Creative Programs & Systems: https://www.cpsmi.com/.