Hybrid Deep Learning Framework for Enhanced Melanoma Detection

Cancer remains a leading cause of death worldwide, underscoring the urgent need for advancements in early detection and treatment technologies. In this paper, we propose the SegFusion Framework, a novel and highly efficient hybrid approach for melanoma detection, combining the segmentation power of U-Net with the classification strength of EfficientNet. The primary objective of our study is to improve the accuracy and efficiency of melanoma detection through this innovative integration. To achieve this, we utilized the HAM10000 dataset to train the U-Net model for accurate segmentation of cancerous regions. Simultaneously, the ISIC 2020 dataset was employed to train the EfficientNet model, optimizing it for binary classification of skin cancer. The proposed SegFusion Framework significantly outperforms existing models, achieving an accuracy of 99.01% on the ISIC 2020 dataset, along with a precision of 0.99, recall of 0.99, F1 score of 0.99, and a MCC of 0.97. Additionally, we conducted experiments comparing our approach with the advanced SkinViT model and other recent hybrid methods. The SkinViT model, trained on the same dataset, achieved an accuracy of 98.20%, which is notably lower than that of the SegFusion Framework, further highlighting the superiority of our method. By leveraging U-Net's precise segmentation capabilities and EfficientNet's robust classification performance, the SegFusion Framework offers a comprehensive solution for melanoma detection. Our extensive experiments demonstrate the high accuracy and reliability of our method in both segmentation and classification tasks, marking a significant step forward in automated skin cancer detection. We believe that our framework sets a new benchmark in the field, providing a reliable tool for medical professionals in the early diagnosis and treatment of melanoma and fostering further research in medical imaging.