Computer-aided pattern classification system for dermoscopy images.

Area of Science:

• Dermatology
• Medical Imaging
• Computer-Aided Diagnosis

Background:

• Computer-aided diagnosis is crucial for distinguishing melanoma and pigmented skin lesions.
• Accurate extraction of color, architectural order, symmetry, and homogeneity (CASH) is challenging for clinical diagnosis.

Purpose of the Study:

• To develop and evaluate a novel pattern classification system (PCS) for classifying six types of skin lesion patterns.
• To automate the diagnostic process based on the clinical CASH rule.

Main Methods:

• The PCS employs a five-step process: CIE L*a*b* color space transformation, image pre-processing (enhancement, hair removal), tumor segmentation, color and texture feature extraction, and multiclass support vector machine classification.
• The system is based on the clinical CASH rule.

Main Results:

• The PCS was evaluated on 180 dermoscopic images.
• The diagnostic classifier achieved high performance: 91.64% sensitivity, 94.14% specificity, and an area under the receiver operating characteristics curve (AUC) of 0.948.

Conclusions:

• The proposed pattern classifier demonstrates high accuracy in differentiating benign and malignant skin lesions.
• The fully automated PCS accurately detects patterns from dermoscopy images using color and texture properties.
• Further research can explore incorporating additional pattern features to enhance CASH rule-based classification.