Classification of skin diseases with deep learning based approaches

Area of Science:

• Dermatology
• Computer Science
• Medical Imaging

Background:

• Skin diseases like eczema, seborrheic dermatitis, and skin cancer significantly impact quality of life.
• Accurate and prompt diagnosis is crucial for effective management and treatment of these conditions.
• Current diagnostic methods may require improvement in speed and accuracy for large-scale classification.

Purpose of the Study:

• To develop and evaluate a machine learning model for accurate classification of three common skin diseases: eczema, seborrheic dermatitis, and skin cancer.
• To compare the performance of different classification algorithms, including AlexNet and Support Vector Machine (SVM) with the Relief algorithm.
• To assess the effectiveness of feature selection using the Relief algorithm in enhancing classification accuracy.

Main Methods:

• Utilized a dataset comprising 693 individuals with eczema, 750 with skin cancer, and 770 with seborrheic dermatitis.
• Employed the Relief algorithm for feature selection to improve classification performance.
• Implemented and compared AlexNet and SVM with the Relief algorithm for skin disease classification using cross-validation.
• Evaluated model performance on the ISIC 2017 dataset with varying training and testing data splits (80/20 and 70/30).

Main Results:

• SVM classification with the Relief algorithm achieved a higher accuracy rate of 92.10% compared to AlexNet (89.39%) with an 80% training and 20% test split.
• On the ISIC 2017 dataset, the SVM with Relief algorithm achieved 89.16% accuracy (80/20 split) and 91.11% accuracy (70/30 split).
• The proposed model integrating feature selection and a simplified architecture demonstrated superior performance, highlighting the efficacy of deep learning and transfer learning in skin disease classification.

Conclusions:

• SVM classification combined with the Relief algorithm is a highly accurate and effective method for classifying eczema, seborrheic dermatitis, and skin cancer.
• The study confirms the potential of deep learning and transfer learning techniques for early and precise diagnosis of skin diseases.
• Improved classification accuracy can lead to reduced mortality rates from skin cancer through timely and effective treatment.