Optimizing melanoma diagnosis: A hybrid deep learning and quantum computing approach for enhanced lesion classification

Melanoma is one of the most aggressive forms of skin cancer, necessitating advanced diagnostic tools to improve early detection. This study presents a novel AI-driven approach that combines deep neural networks with quantum computing techniques for enhanced lesion classification. Specifically, we employ a U-Net model for segmentation and a hybrid Convolutional Neural Network - Quantum Neural Network (CNN-QNN) for classification. Our approach achieves a precision of 99.67%, recall of 99.67%, and an overall accuracy of 99.35% on the HAM10000 dataset. Additionally, we report a sensitivity of 99.4%, a specificity of 99.2%, and a macro F1-score of 99.5%, significantly surpassing traditional CNN-based classifiers. This hybrid model outperforms conventional deep learning approaches, demonstrating its potential for aiding dermatologists in clinical decision-making. A comparative analysis with state-of-the-art models further validates the effectiveness of our method. Melanoma is one of the most aggressive forms of skin cancer and a leading cause of mortality related to skin cancer worldwide. Its incidence is steadily increasing, particularly among fair-skinned populations. In recent decades, the rising number of melanoma cases has highlighted the need for more accurate and timely diagnostic tools to improve survival rates. The use of artificial intelligence, especially deep learning models such as convolutional neural networks, has proven to be a promising technology to support dermatologists in the early diagnosis of melanoma. The application of these advanced models allows for more precise analysis of dermatological images, facilitating the identification of suspicious lesions and improving the ability to distinguish between benign and malignant ones. This study demonstrates that the use of models such as U-Net, combined with preprocessing techniques like Autoencoder, can significantly enhance the accuracy of skin lesion segmentation and classification, thus improving melanoma diagnosis and treatment. After applying the Autoencoder and segmenting with the U-Net network, we used a hybrid CNN-quantum neural network model that achieved an accuracy of 99.67%, a precision of 99.35%, and a recall of 99.67%. • Implementation of an autoencoder to effectively remove hair from dermoscopic images, improving visual quality for more accurate skin lesion diagnosis. • Utilization of a hybrid CNN-quantum model for lesion classification, achieving an accuracy of 99.67%. • The combination of autoencoder and quantum neural networks offers a substantial advancement in automated skin lesion and melanoma diagnosis.