Classification of Nevi, Seborrheic Keratosis and Melanomas Based on Image Fractal Dimension

Inhalt:
Skin disease is prevalent in human beings. Due to delayed detection and inadequate early intervention, many skin diseases ultimately progress to skin cancer. Fractal dimension serves as a key parameter for characterizing surface roughness in images. In dermoscopy analysis, this metric is extensively employed to investigate internal structural variations within biological tissues. Malignant melanoma is one kind of most deadly skin cancers, which leads to the death of many people annually. Nevus and seborrheic keratosis are common benign skin tumors. The seborrheic keratosis is prone to be misdiagnosed as other skin diseases. While the studies of seborrheic keratosis based on the image fractal dimension are still limited. For this study, we utilized data from the ISIC 2017 (the publicly available International Skin Imaging Collaboration (ISIC) 2017 Challenge skin lesion segmentation dataset) to obtain a relatively large amount of data. Excluding excessive hair follicles, skin ulcers et al., the final data set comprised 2,371 images: 1,615 benign nevus cases, 311 seborrheic keratosis cases, and 445 melanoma cases. Notably, melanoma demonstrated higher fractal dimensions than seborrheic keratosis or benign nevus, while seborrheic keratosis exhibited greater fractal dimensions than benign nevus, despite variations in equipment specifications, magnification levels, exposure duration, and other factors. The intermediate fractal dimension of seborrheic keratosis ranging between those of melanoma and benign nevus aligns with its tendency to be misdiagnosed as other skin diseases —including malignant melanoma and benign nevus. Also the alteration of fractal dimension may reflect pathological metabolic changes in melanoma.