Instance-Based Segmentation for Boundary Detection of Neuropathic Ulcers Through Mask-RCNN

Neuropathic ulcers form and proliferate because of peripheral neuropathy, usually in diabetic patients. The existing ulcer assessment process which relies on visual examination, potentially be imprecise and inefficient. Therefore this indicates the necessity of a more quantitative and cost-effective solution that enables ulcer diagnosing process much faster. In the current literature, different deep learning approaches are available for diagnosing illnesses through medical imagery. When diagnosing diabetic patients who are suffering from neuropathic ulcers through imagery, the locating and segmenting of ulcer boundaries is of great importance. In this study, we propose an approach to automate the process of locating and segmenting ulcers through Mask-RCNN model. We use a dataset of 400 ulcer imagery and corresponding annotations of ulcers for this task. This approach achieves an overall ulcer detection average precision (AP) at Intersection over union (IoU) threshold 0.5 of 0.8632 and mean average precision (mAP) at Intersection over union (IoU) threshold 0.5 to 0.95 by steps of size 0.05 of 0.5084 for ResNet-101 backbone.