Performance evaluation of E-VGG19 model: Enhancing real-time skin cancer detection and classification

Area of Science:

• Dermatology and Medical Imaging
• Artificial Intelligence in Healthcare
• Computational Pathology

Background:

• Skin cancer is a significant global health concern requiring early detection for better patient outcomes.
• Machine learning (ML) and deep learning models show potential for improving diagnostic accuracy in skin cancer.
• Automated detection systems can aid healthcare professionals in identifying skin lesions.

Purpose of the Study:

• To enhance the VGG19 pre-trained model for skin cancer prediction.
• To evaluate the performance of various pre-trained deep learning models (VGG19, ResNet152v2, InceptionResNetV2, DenseNet201, ResNet50, InceptionV3) for skin lesion classification.
• To compare the efficacy of traditional machine learning classifiers when combined with enhanced deep learning features.

Main Methods:

• Utilized the VGG19 pre-trained model, enhanced with max pooling and dense layers, for skin cancer detection.
• Explored and compared multiple pre-trained models including VGG19, ResNet152v2, InceptionResNetV2, DenseNet201, ResNet50, and InceptionV3.
• Trained models on a skin lesion dataset, categorizing lesions as malignant or benign, and extracted features for classification using Linear SVM, KNN, DT, LR, and SVM.

Main Results:

• The enhanced VGG19 (E-VGG19) model combined with traditional classifiers significantly improved overall classification accuracy for skin cancer detection.
• Performance comparison using metrics like recall, F1 score, precision, sensitivity, and accuracy revealed the effectiveness of the proposed approach.
• The study demonstrated the value of integrating deep learning feature extraction with classical ML classifiers for robust skin lesion analysis.

Conclusions:

• Combining the E-VGG19 model with traditional classifiers offers a powerful approach for accurate skin cancer detection and classification.
• The findings provide valuable insights into selecting optimal models and classifiers for automated skin cancer screening.
• This research contributes to developing advanced technologies for early skin cancer identification, potentially improving treatment outcomes.