Diabetic foot ulcer segmentation using logistic regression,DBSCAN clustering and morphological operators
-
Armando Heras Tang [1] ; Damián Valdés Santiago ; Ángela León Mecías ; Marta Lourdes Baguer Diaz Romañach ; Alejandro Mesejo Chiong ; Carlos Cabal Mirabal
-
[1] Universidad de La Habana
Universidad de La Habana
Cuba
-
- Localización: ELCVIA. Electronic letters on computer vision and image analysis, ISSN-e 1577-5097, Vol. 21, Nº. 2, 2022 (Ejemplar dedicado a: Current Issue), págs. 23-39
- Idioma: inglés
- Enlaces
-
Resumen
Digital images are used for the evaluation and diagnosis of diabetic foot ulcers. Selecting the region ofinterest in an image is a preliminary step for subsequent analysis. Most of the time, manual segmentationis not very reliable because specialists may have different opinions regarding the ulcer border; which hasencouraged researchers to find and test automated segmentation techniques. This paper presents a computer-aided ulcer region segmentation algorithm for diabetic foot images. The proposed algorithm has two stages:ulcer region segmentation and post-processing of segmentation results. The first stage focused on the se-lection of a trained machine learning model to classify pixels inside the ulcer’s region, after comparing fivelearning models. We performed exhaustive experiments with our own annotated dataset of images of Cubanpatients. A second stage was needed due to the existence of some misclassified pixels. In order to solvethis, we applied the DBSCAN clustering algorithm, together with dilation and closing morphological oper-ators. The best-trained model after the post-processing stage was the logistic regressor (Jaccard Index0.81,accuracy0.94, recall0.86, precision0.91, and F1 score0.88). This trained model was sensitive to irrelevantobjects in the scene, but successfully detected the patient’s foot. Physicians found these results promisingfor measuring the lesion area and to monitoring the ulcer healing process during treatments, which wouldsignificantly reduce errors
