Robust dermatological wound image segmentation in clinical photos

Wound image segmentation of skin area is one of the key steps in automatic judgment, which can help clinicians in their diagnosis. This work is part of the telecare program dedicated to the design of a surgery site infection detection system We proposed a series of image processing steps composed of image processing algorithms and a novel segmentation method of complementary computation. As edge-based and color-based segmentation methods proved to be inefficient for complicated cases, we use these two methods in a joint fashion: to improve the segmentation results we apply an auxiliary algorithm to enforce edge detection and filtering of color pixels. The mechanism can derive the color value of wound parts directly from the processed photos. By analyzing the variation of color value, the system can evaluate the condition of a wound (e.g., the color of wound turns darker when being infected). As a result, of 122 clinical photos taken of patients in the National Taiwan University Hospital, one hundred six (86.9%) are judged as completely segmented by our proposed algorithm. The evaluation of completely segmentation is based on the judgement of experienced medical doctors. This ability to extract complete wound regions, make it possible to conduct further analysis such as infection detection and recovery evaluation.