Vitiligo is a chronic and progressive autoimmune skin disorder that depigments the skin due to the absence of functional melanocytes in any part of the body. Approximately 0.5 to 2% of the worldwide population has the condition, and for many patients, it bears considerable skin complaints, ranging from irritation to cancer as well as psychosocial and emotional consequences because of its cosmetic appearance. Precise diagnosis and monitoring of progression are of critical importance for effective management. However, current examination methods depend on subjective visual assessment and there is a lack of sensitive and standardized direct procedures in the clinical routine. To address these limitations, we present a deep learning-based approach that uses a custom-designed convolutional neural network trained with a Mexican dataset. The architecture is composed of an encoder-decoder structure with skip connections that allow the model to learn rich spatial features and perform pixel-level segmentation of vitiligo lesions. Results demonstrated a robust model’s performance achieving a sensitivity of 0.897, specificity of 0.999, precision of 0.982, and F1-score of 0.938. Notably, it performs well on entirely unseen patients, demonstrating its potential for real-world clinical application. Findings establish that convolutional models, even in data-limited scenarios, can provide objective and reproducible segmentation of vitiligo lesions, supporting improved monitoring of treatment progress in clinical practice.