Junnan Zhang and Hanyi Nie, National University of Defense Technology, China

Chronic wound have a long recovery time, occur extensively, and are difficult to treat. They cause not only great suffering to many patients but also bring enormous work burden to hospitals and doctors. Therefore, an automated chronic wound detection method can efficiently assist doctors in diagnosis, or help patients with initial diagnosis, reduce the workload of doctors and the treatment costs of patients. In recent years, due to the rise of big data, machine learning methods have been applied to Image Identification, and the accuracy of the result has surpassed that of traditional methods. With the fully convolutional neural network proposed, image segmentation and target detection have also achieved excellent results. However, the accuracy of chronic wound image segmentation and identification is low due to the limitation of the deep convolution neural network. To solve the above problem, we propose a post-processing method based on fully connected CRFs with multi-layer score maps. The experiment results show that our method can be used to improve the accuracy of chronic wound image segmentation and identification.

Fully Connected CRFs, Chronic Wound Segmentation, Post-processing Method