A Comprehensive Study on Melanoma Detection: A Hybrid Framework Using Machine Learning, Deep Learning, and Quantum Computing

Melanoma remains the most lethal form of skin cancer, and early automated detection has the potential to save thousands of lives annually by flagging suspicious lesions before metas-tasis occurs. The review of seventeen recent research papers across deep learning, U-Net segmentation, class imbalance handling, quantum machine learning, and clinical epidemiol-ogy reveals a consistent gap: no existing system integrates all three computational paradigms — classical ML, deep learning, and quantum computing — into a single unified pipeline de-signed specifically for dermoscopy-based melanoma detection. QEMDS addresses that gap with a clearly motivated, five-stage software architecture: (1) standardised preprocessing with CLAHE and augmentation; (2) U-Net lesion segmen-tation to isolate the diagnostic region; (3) EfficientNet-B3 feature extraction producing 1536-dimensional vectors; (4) feature-level SMOTE oversampling and PCA reduction to 16 dimensions; and (5) a four-layer, 8-qubit Variational Quan-tum Circuit classifier trained end-to-end via PennyLane’s parameter-shift gradients. Every component is implemented in software and requires no physical quantum hardware. The system is planned for evaluation on the ISIC 2020 benchmark using AUC-ROC, Sensitivity, Specificity, Balanced Accuracy, and F1-Score, with three classifier configurations compared to isolate the quantum layer’s contribution. Future work will extend QEMDS along several directions: replacing EfficientNet with a contrastively pretrained backbone following Kim et al. [15]; exploring Vision Transformer-based feature extractors as suggested by Zhao et al. [11]; increasing qubit count as simulator efficiency improves; adding quantum error mitigation techniques to assess robustness; expanding to multi-class skin lesion classification beyond binary melanoma detection; and evaluating on external datasets from diverse geographic populations and imaging devices to test generali-sation beyond ISIC.

[1]     A. Esteva et al., “Dermatologist-level classification of skin cancer with deep neural networks,” Nature, vol. 542, pp. 115–118, 2017. [2]     N. Codella et al., “Skin lesion analysis toward melanoma detection 2018: A challenge hosted by the International Skin Imaging Collaboration,” arXiv preprint arXiv:1902.03368, 2019. [3]     O. Ronneberger, P. Fischer, and T. Brox, “U-Net: Convolutional net-works for biomedical image segmentation,” in Proc. MICCAI 2015, Springer, pp. 234–241, 2015. [4]     M. Tan and Q. V. Le, “EfficientNet: Rethinking model scaling for convolutional neural networks,” in Proc. ICML 2019, pp. 6105–6114, 2019. [5]     N. V. Chawla, K. W. Bowyer, L. O. Hall, and W. P. Kegelmeyer, “SMOTE: Synthetic minority over-sampling technique,” Journal of Artificial Intelligence Research, vol. 16, pp. 321–357, 2002. [6]     J. Biamonte et al., “Quantum machine learning,” Nature, vol. 549, pp. 195–202, 2017. [7]     M. Cerezo et al., “Variational quantum algorithms,” Nature Reviews Physics, vol. 3, pp. 625–644, 2021. [8]     J. Li et al., “A hybrid quantum-classical framework for brain tumour classification using MRI,” IEEE Journal of Biomedical and Health Informatics, vol. 26, no. 8, pp. 4013–4022, 2022. [9]     V. Bergholm et al., “PennyLane: Automatic differentiation of hybrid quantum-classical computations,” arXiv preprint arXiv:1811.04968 (v4, 2022). [10]     V. Rotemberg et al., “A patient-centric dataset of images and metadata for identifying melanomas using clinical context,” Scientific Data, vol. 8, no. 34, 2021.