Deep Learning and Handcrafted Method Fusion: Higher Diagnostic Accuracy for Melanoma Dermoscopy Images.

Area of Science:

• Dermatology
• Medical Imaging
• Artificial Intelligence

Background:

• Melanoma classification relies on accurate image analysis.
• Conventional image processing and deep learning offer complementary strengths.
• Integrating these methods may enhance diagnostic performance.

Purpose of the Study:

• To develop and evaluate a hybrid approach combining conventional image processing and deep learning for melanoma classification.
• To assess the synergistic effect of fusing features from distinct analytical techniques.

Main Methods:

• A conventional image processing arm with biologically inspired modules and clinical data.
• A deep learning arm utilizing a ResNet-50 network for feature extraction and classification.
• Ensemble logistic regression to fuse scores from both arms for final melanoma probability prediction.

Main Results:

• The fused technique achieved a cross-validated area under the receiver operator characteristic curve (AUC) of 0.94.
• The deep learning classifier alone yielded an AUC of 0.87.
• The conventional image processing classifier alone yielded an AUC of 0.90.

Conclusions:

• Fusion of conventional image processing and deep learning demonstrates superior performance in melanoma classification.
• The synergistic combination of diverse analytical methods holds promise for improved diagnostic accuracy.
• Further research into hybrid approaches is warranted for clinical application.