Microscopic melanoma detection and classification: A framework of pixel-based fusion and multilevel features...

Area of Science:

• Dermatology
• Medical Imaging
• Computer Vision

Background:

• Melanoma diagnosis is critical due to rising cases and mortality, especially in young populations.
• Automated systems are essential for efficient and accurate skin cancer diagnosis.
• Existing diagnostic methods require improvement in speed and precision.

Purpose of the Study:

• To develop an automated method for skin lesion detection and recognition.
• To improve the accuracy and efficiency of melanoma diagnosis through advanced image processing techniques.
• To present a novel approach combining image fusion, feature extraction, and machine learning for skin lesion analysis.

Main Methods:

• Image preprocessing using mean-based function, top-hat, and bottom-hat filters for contrast stretching.
• Lesion segmentation via seed region growing and graph-cut methods, followed by pixel-based fusion.
• Multilevel feature extraction including Histogram of Oriented Gradients (HOG), Speeded Up Robust Features (SURF), and color features.
• Feature reduction using variance precise entropy and classification with a Support Vector Machine (SVM) using a cubic kernel function.

Main Results:

• Segmentation accuracy achieved 95.86% on PH2, 94.79% on ISBI2016, and 94.92% on ISIC2017 datasets.
• Classification accuracy reached 98.20% on PH2 and 95.42% on ISBI2016 datasets.
• The proposed automated system demonstrated outstanding performance compared to current state-of-the-art techniques.

Conclusions:

• The developed automated method is highly effective for skin lesion detection and recognition.
• The system shows significant potential for improving early melanoma diagnosis and patient outcomes.
• The proposed approach validates the use of advanced image fusion and feature reduction techniques in dermatological applications.