Preprocessing and comparative analysis of dermatoscopic images of benign and malignant skin lesions using Python tools

Abstract

Introduction: Early diagnosis of skin cancer requires high-quality databases. Advances in AI improve accuracy, but challenges remain in the storage, representation, and objective evaluation of dermatological images. Objective: To explore the use of quantitative digital image processing and analysis methodologies to categorize skin lesions, evaluating the suitability of the database for use in training models. Methodology: Exploratory and descriptive research in which dermatoscopic images of benign and malignant lesions were collected and processed, using segmentation techniques, feature extraction, and statistical analysis through the development of executable code in Python and specialized libraries. Result: Differences between images of benign and malignant lesions were evident. Benign lesions had higher and more homogeneous values in the red, green, and blue color channels, as well as higher average brightness and lower contrast. Malignant lesions show greater color variability, lower brightness, higher contrast, and a more complex texture. Discussion: The results are consistent with findings previously described in the literature, where significant differences in contrast between malignant and benign lesions and greater color heterogeneity in malignant images were observed. Conclusion: The results obtained reinforce the importance of quantitative and automated analysis in the evaluation of medical images, providing objectivity and reproducibility to the analysis of complex data. The identification of differentiating patterns in color and texture allows the identification of lesion types.