| Popis: |
This article exposes a semi-automatic method with the potential to aid the doctor while supervising the progression of skin lesions. The proposed methodology pre-segments skin lesions using the SLIC0 algorithm for the generation of superpixels. Following this, each superpixel is represented using a descriptor constructed of a mix from GLCM and Tamura texture features. The feature's gain ratios were utilized to choose the data applied in the semi-supervised clustering algorithm Seeded Fuzzy Cmeans. This algorithm uses certain specialist-marked regions to group the superpixels into lesion or background regions. Finally, the segmented image undergoes a post-processing step to eliminate sharp edges. The experiments were performed on a total of 3974 images. We used the 2995 images from PH2, DermIS and ISIC 2018 datasets to establish our method's specifications and the 979 images from ISIC 2016 and ISIC 2017 datasets for performance analysis. Our experiments demonstrate that by manually identifying a few percentages of the generated superpixels, the proposed approach reaches an average accuracy of 95.97%, thus giving a superior performance to the techniques presented in the literature. Even though the proposed method requires physicians' intervention, they can obtain segmentation results similar to manual segmentation from a significantly less time-consuming task. |
