Exploring the color feature power for psoriasis risk stratification and classification: A data mining paradigm

Area of Science:

• Dermatology
• Medical Imaging
• Machine Learning

Background:

• Dermatologist's assessment of psoriasis heavily relies on skin color.
• Current computer-aided diagnosis systems often neglect the significance of color features.
• There is a need for advanced systems that incorporate color analysis for psoriasis classification.

Purpose of the Study:

• To develop an automated psoriasis computer-aided diagnosis (pCAD) system.
• To effectively classify psoriasis skin images using dominant color features.
• To improve the accuracy and reliability of psoriasis diagnosis through color analysis.

Main Methods:

• Exploration of fourteen color spaces yielding 86 distinct color features.
• Implementation of a support vector machine (SVM) learning framework.
• Utilization of principal component analysis (PCA) for dominant color feature selection.

Main Results:

• The pCAD system achieved 99.94% accuracy with 99.93% sensitivity and 99.96% specificity using a 10-fold cross-validation on 540 images.
• Color features alone demonstrated comparable performance to grayscale or combined features in varying data size protocols.
• The system showed high reliability across different feature sets, with color space alone achieving 94.42% reliability.

Conclusions:

• The developed pCAD system effectively utilizes color features for psoriasis classification.
• Color space analysis is a viable and powerful tool for automated psoriasis diagnosis.
• The pCAD system's performance is consistent and reliable, even when validated against facial color databases.