82 Researchers have unveiled an innovative deep learning model designed to enhance the early detection of skin cancer with unprecedented accuracy. The study introduces a hybrid framework that combines ConvNeXtV2 and focal self-attention mechanisms, offering a more precise and scalable diagnostic tool. The findings were published in the journal Results in Engineering. Skin cancer remains one of the most common malignancies worldwide, with early detection being critical for successful treatment. But distinguishing between benign and malignant lesions is challenging due to the subtle nature of their visual characteristics. Traditional diagnostic methods, including manual evaluations and classical machine learning approaches, have struggled with accuracy, often leading to delayed diagnoses. This new deep learning model aims to address these issues by leveraging advanced artificial intelligence techniques. The research team behind the study designed a model that integrates ConvNeXtV2 blocks with focal self-attention layers. ConvNeXtV2 is responsible for capturing fine-grained local features from dermatological images, ensuring that critical visual details are preserved. The focal self-attention mechanism enhances the model’s ability to focus on diagnostically significant regions, improving the overall classification of different skin cancer types. The model was tested using the ISIC 2019 dataset, one of the most comprehensive and widely recognised databases for skin cancer classification. The dataset includes images of various skin lesion types, such as melanoma, basal cell carcinoma, and squamous cell carcinoma. Due to significant class imbalances, where some categories contain far fewer images than others, many existing machine learning models struggle to provide reliable performance across all classes. The newly developed model overcomes these issues by employing advanced data augmentation techniques, improving its ability to generalise across diverse cases. Results from the study indicate that the model achieves an accuracy of 93.60%, outperforming existing deep learning frameworks. Compared to ResNet50, a commonly used convolutional neural network for image classification, the new model demonstrates a 10.8% increase in accuracy. It also surpasses the best-performing vision transformer model, Swinv2-Base, by 3.3%. Such improvements mark a significant advancement in the field of dermatological diagnostics, as higher accuracy rates translate to reduced false positives and negatives, ultimately improving patient outcomes. The researchers highlight that one of the most critical aspects of their model is its efficiency. With a reduced number of parameters compared to conventional deep learning models, it remains lightweight and mobile-friendly. This aspect makes it particularly well-suited for integration into real-world clinical settings, where rapid and reliable diagnosis is essential. Mobile accessibility also opens the possibility of remote diagnostics, allowing patients in underserved areas to benefit from advanced AI-driven screening tools. Despite its success, the model faces some challenges in classifying more complex skin cancer types, particularly those with overlapping characteristics. Future research aims to refine the model’s sensitivity to these cases by further optimising the attention mechanisms and incorporating additional training data. Researchers are also exploring potential collaborations with healthcare institutions to implement the model in clinical practice. This development represents a step forward in AI-assisted medical diagnostics, offering a promising solution for improving early skin cancer detection. As technology continues to evolve, deep learning models like this one could become an integral part of routine dermatological assessments, helping medical professionals provide faster and more accurate diagnoses.