Nested segmentation and multi-level classification of diabetic foot ulcer based on mask R-CNN

A diabetic foot ulcer(DFU) is a common chronic complication of diabetes because of the dysfunction of islets or receptors of insulin, and it has a high disability and mortality rate. Measuring diabetic foot ulcers is also one of the popular application areas where computer vision combines with deep learning techniques. However, some remaining defects in these studies prevent them from accurately visualizing the wound of different severity. Based on this, we used a multi-classification model to mark the wounds into five grades according to the Wagner diabetic foot grading method. It segmented the different grades in each different level wound using colorfully nested ring shapes to reflect the gradual change of wound grades. We collected 1426 DFU images, of which 967 had nested labels and 459 were single-level labels, with images marked with colored rings to show different degrees of wounds. And then, we constructed a deep learning model of diabetes foot ulcer wounds for semantic segmentation based on Mask Region-based convolutional neural networks (Mask R-CNN), and obtain different levels of diabetes nested segmentation results to reflect the different severity in one wound. Finally, we test and evaluate the performance data of the model. Compared with the state-of- the-art results concerning segmentation and classification and diagnosis of diabetic foot wounds, our model has achieved better performance data (specificity = 99.50%, sensitivity = 70.62%, precision = 84.56%, Mean Average Precision = 85.70%). It shows the effectiveness of our nested segmentation and multi-level classification method. It provides some suggestions and directions for the subsequent evaluation and diagnosis and treatment of diabetic foot ulcers.