Fig. 3
From: A YOLO-based AI system for classifying calcifications on spot magnification mammograms
YOLOv4-AMDF architecture. The backbone and feature aggregation parts of YOLOv4 are combined with the proposed AMDF module for fusion predetection representation dl of each scales, where pc denotes the objectness prediction of the corresponding levels, bx, by, bh, bw are the bounding box offsets, c1, c2 represent the class predictions, \(\alpha\), \(\beta\), \(\gamma\) are reliabilities of each scale and ⊕ denotes pixel-wise addition. ys is the tensor for final prediction. YOLO v4, You Only Look Once version 4; MC, microcalcification; ROI, region of interest