Advancing Dermoscopic Diagnosis: U-Net-Based Melanoma Cell Carcinoma (MCC) Detection with Embedded High-Quality Feature Selection Using Feature Visualisation Bag (FVB)

Dermoscopy is a skin microscopy technique used to measure the epidermis of the skin. Skin cancer, particularly Melanoma Cell Carcinoma (MCC), is a major public health concern that can be fatal if not detected and treated early. An innovative framework for early detection of skin melanoma cells using a sophisticated Deep Learning (DL) approach is proposed in this article. The handcrafted and deep features are combined to form a vocabulary of features using the concept of transfer learning (TL). The binary patterns of melanoma are detected and merged with deep features in the first stage using the U-Net architecture. Feature Visualisation Bag (FVB) is used to select a high-quality set of features after eliminating redundant features. Our experimental results demonstrate the efficacy of the proposed framework. This paper show significant improvements in accuracy and efficiency compared to the baseline U-Net approach in comprehensive evaluations on a curated dataset of dermoscopic images. Quantitative metrics with classification accuracy of 98.7%, and F1-score of 94.3% demonstrate the superiority of the U-Net architecture enhanced with feature selection. The proposed model is also computationally inexpensive and capable of detecting cancer cells at very early stages, claiming the model to be smarter than state-of-the-art models currently available.