Two-Stage Deep Neural Network via Ensemble Learning for Melanoma Classification.

Area of Science:

• Dermatology
• Medical Imaging
• Artificial Intelligence

Background:

• Melanoma, a high-fatality skin disease, requires early diagnosis for improved patient survival.
• Dermoscopy images are crucial for classifying melanoma, benign nevi, and seborrheic keratosis.
• Existing methods using low-level image information or single deep learning networks have limitations in melanoma classification accuracy.

Purpose of the Study:

• To develop an automatic melanoma classification method that leverages rich, deep image features.
• To propose an ensemble deep learning framework for enhanced melanoma classification accuracy.

Main Methods:

• Image segmentation using U-net to isolate lesion areas.
• Employing five distinct classification models enhanced with Squeeze-Excitation (SE) blocks.
• Integrating diverse classification results through a novel ensemble network.

Main Results:

• The proposed ensemble method achieved excellent performance on the ISIC 2017 dataset.
• Achieved a high accuracy of 0.909 on melanoma classification.
• Demonstrated the validity and effectiveness of the ensemble approach.

Conclusions:

• The developed classification framework offers an efficient and accurate method for melanoma diagnosis using dermoscopy images.
• This approach supports early melanoma detection and subsequent treatment planning.
• The ensemble method effectively utilizes deep feature information for improved classification outcomes.