Hybrid Deep Learning Framework for Enhanced Melanoma Detection.

Area of Science:

• Dermatology and Medical Imaging
• Artificial Intelligence in Healthcare
• Computational Pathology

Background:

• Melanoma detection remains critical due to cancer's high mortality rates.
• Advancements in early detection and treatment technologies are urgently needed.
• Hybrid AI models offer potential for improved diagnostic accuracy.

Purpose of the Study:

• To develop and evaluate the SegFusion Framework, a novel hybrid approach for enhanced melanoma detection.
• To integrate U-Net for precise segmentation and EfficientNet for robust classification of skin lesions.
• To improve the accuracy and efficiency of automated melanoma diagnosis.

Main Methods:

• Utilized the HAM10000 dataset to train a U-Net model for accurate segmentation of cancerous regions.
• Employed the ISIC 2020 dataset to train an EfficientNet model for binary classification of skin cancer.
• Developed the SegFusion Framework by combining U-Net and EfficientNet for a hybrid detection approach.

Main Results:

• The SegFusion Framework achieved a 99.01% accuracy on the ISIC 2020 dataset.
• Achieved high performance metrics: 0.99 precision, 0.99 recall, 0.99 F1 score, and 0.97 MCC.
• Outperformed the SkinViT model (98.20% accuracy) and other recent hybrid methods.

Conclusions:

• The SegFusion Framework demonstrates superior accuracy and reliability in melanoma detection.
• The hybrid approach effectively leverages segmentation and classification for comprehensive analysis.
• This framework represents a significant advancement in automated skin cancer detection, aiding medical professionals in early diagnosis.