Two-step Deep Learning Framework for Chronic Wounds Detection and Segmentation: A Case Study in Colombia

Chronic wounds segmentation using images emerges as an essential tool in ulcers treatment, mainly in patients living far from medical centers, as is the case of some villages in Colombia. There is a significant difference between the severity of wounds of Colombian patients compared with those from developed countries; Therefore, the direct use of available datasets or models yields to significantly reduced performance when tested with images of Colombian patients. This work proposes two-step detection, and segmentation, to minimize the unnecessary information of the scene, center the wound, standardize the process, and take advantage of the available datasets reducing the loss of performance. Centering the wound improves accuracy results in up to 4%. Furthermore, the proposed two-step framework is evaluated in chronic foot wounds from Colombian patients in Contratación town, Santander validating the robustness of the proposed framework.