Asymmetry Measures of Dermoscopic Images for Automated Melanoma Detection

Melanoma is the deadliest form of skin cancer and early detection is critical for successful treatment. Dermoscopy is an effective tool for assessing the likelihood of a suspicious lesion being malignant. In this study, we focus on improving the detection of lesion asymmetry, one of the key factors of automated melanoma recognition. We employ two approaches: irregularity of the lesion contour assessed with measures that compare lesion quadrants with respect to area, color, and melanin content, and size theory using one-dimensional measuring functions to determine asymmetry. Measuring functions were mapped into size functions and compared using bottleneck distances, which were then used as classification features. Annotated dermoscopic images were used to train and assess classifiers for both methods. Our results show that the combined methods exhibit 95% lesion classification accuracy and suggest that size functions may be suitable for detecting melanoma directly. Our findings confirm that smartphone-based systems can be valuable assistive devices with substantial benefits for both individual healthcare and public health outcomes. This technology has the potential to enhance patient accessibility, particularly in low-resource settings and areas with limited healthcare access.