Skin lesion segmentation by using object detection networks, DeepLab3+, and active contours

Developing an automatic system for detection, segmentation, and classification of skin lesions is very useful to aid well-timed diagnosis diseases. Lesion segmentation a crucial task automated cancers, as it affects significantly the accuracy subsequent steps. Varieties in sizes locations lesions, with low-contrast boundaries make this challenging. In paper, three-stage CNN-based method presented accurate from dermoscopic images. At first step, normalization, approximate are estimated. Due importance normalization stage, three networks (Mask R-CNN, RetinaNet, YOLOv3) used lesion detection. A convolutional network combine results object detection novel approach. The output stage normalized cropped image containing detected center. second CNN DeepLab3+ structure, extract image. Finally, active contour postprocessing enhance boundary segmented lesion. proposed evaluated on well-known datasets. Experiments show that outperforms all previous state-of-the-art methods.