Enhancing Microcalcification Detection in Mammography with YOLO-v8 Performance and Clinical Implications

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Breast microcalcifications are critical early indicators of breast cancer.
• Accurate detection is vital for timely diagnosis and treatment.
• Deep learning object detection models have advanced microcalcification identification.

Purpose of the Study:

• To evaluate the YOLO-v8 object detection algorithm for breast microcalcification detection.
• To assess the performance and clinical utility of YOLO-v8 compared to existing methods.

Main Methods:

• Utilized a dataset of 10,323 mammograms from 7615 participants with microcalcifications.
• Employed the YOLO-v8 model for detection, validated using five-fold cross-validation.
• Performance metrics included accuracy, recall, F1 score, mAP50, and mAP50-95.

Main Results:

• YOLO-v8 achieved high performance: mAP50 of 0.921, mAP50-95 of 0.709, F1 score of 0.82.
• Demonstrated superior detection accuracy (0.842) and recall (0.796) compared to prior techniques.
• Showed significant improvements in both detection speed and accuracy over methods like Mask R-CNN.

Conclusions:

• YOLO-v8 surpasses traditional methods for breast microcalcification detection.
• Its multi-scale detection enhances clinical practicality for large-scale breast cancer screenings.
• Future work should focus on classification of microcalcifications to aid radiologists.