Using Deep Learning to Improve Nonsystematic Viewing of Breast Cancer on MRI.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Breast MRI is crucial for cancer detection and staging.
• Manual review of stacked breast MRI slices can be time-consuming.
• AI offers potential solutions for optimizing image analysis workflows.

Purpose of the Study:

• To assess the feasibility of deep learning for identifying tumor-containing axial slices in breast MRI.
• To evaluate the accuracy of a convolutional neural network (CNN) for this task.
• To quantify potential time savings for radiologists.

Main Methods:

• Retrospective analysis of 273 breast MRI scans from patients with invasive breast cancer.
• Development and training of a CNN to classify subimages as 'cancer' or 'no cancer'.
• Validation of the algorithm's accuracy, sensitivity, and specificity against pathology results.

Main Results:

• The deep learning system achieved 92.8% accuracy in tumor detection on a held-out test set.
• Sensitivity was 89.5% and specificity was 94.3%.
• Manual scrolling without AI assistance took 3-45 seconds per case.

Conclusions:

• Deep learning effectively identifies tumor-containing slices in breast MRI.
• This technology can be integrated into PACS to streamline image review.
• Potential benefits include reduced reading time and improved efficiency, especially during multidisciplinary team meetings.