Hierarchical Clustering for Image Classification in Dermatology: Towards Mobile Deploying

Area of Science:

• Dermatology
• Computer Science
• Medical Imaging

Background:

• Growing interest in computer-assisted diagnosis for skin lesions.
• Need for accessible, real-world mobile applications in dermatology.

Purpose of the Study:

• To propose a combined desktop-smartphone solution for dermatological image classification.
• To evaluate the effectiveness of clustering methods and RGB color histograms for skin lesion analysis.

Main Methods:

• Implementation of hierarchical agglomerative and divisive clustering for cluster analysis.
• Utilizing RGB color histogram as a descriptor for global image analysis.
• Employing cosine similarity for classifying query images based on cluster centroids.

Main Results:

• Tested the solution using a public database of dermoscopic images.
• Achieved an overall accuracy of 0.73 (95% Confidence Interval: 0.58; 0.85).

Conclusions:

• The combined desktop-smartphone approach demonstrates a viable method for dermatological image classification.
• The system shows potential for computer-assisted diagnosis in real-life settings.