Background: Melanoma is the most aggressive form of skin cancer and is associated with high mortality when not diagnosed at an early stage. Recent advances in dermatoscopic image analysis combined with artificial intelligence have demonstrated considerable potential for improving diagnostic accuracy. ConvMixer, a hybrid deep learning architecture that integrates convolutional neural networks with a mixer-style design, has recently emerged as a powerful model for image classification tasks. This study aimed to evaluate the effectiveness of the ConvMixer model for automated melanoma detection using dermatoscopic images. 
Methods: Dermatoscopic images were collected from two publicly available datasets: The International Skin Imaging Collaboration (ISIC) and the Public Health (PH2) database. The ISIC dataset comprised 31,696 benign lesions and 7,319 malignant melanoma images, which were divided into training (80%), validation (10%), and test (10%) sets. The PH2 dataset, consisting of 40 melanoma and 160 melanocytic nevi images, was used exclusively for external testing. Image preprocessing, normalization, and data augmentation were performed prior to model training. Model performance was assessed using sensitivity, specificity, accuracy, F1 score, and the area under the receiver operating characteristic curve (AUC).
Results: The ConvMixer model demonstrated strong discriminative ability between malignant and benign skin lesions across both datasets. On the ISIC dataset, the model achieved a sensitivity of 0.9126, specificity of 0.6683, and accuracy of 0.7142. On the PH2 dataset, higher specificity (0.95) and accuracy (0.905) were observed, along with a sensitivity of 0.725. High AUC values further confirmed robust classification performance and generalizability across datasets with differing characteristics.
Conclusion: The ConvMixer model shows strong potential as an effective AI-assisted tool for melanoma detection from dermatoscopic images. Its consistent performance on both large-scale and controlled datasets supports its applicability in diverse clinical settings, highlighting its value for early melanoma screening and decision support in dermatology.