Skin Lesion Detection Using Handcrafted and DL-Based Features Fusion

Skin cancer is a common cancer, with melanoma being the most dangerous type. Early diagnosis is crucial for effective treatment and better outcomes. Dermatologists have widely used digital dermoscopy for cancer diagnosis, although it continues to rely substantially on the clinician's experience and knowledge. Automated skin lesion categorization systems are a promising approach to helping dermatologists make accurate diagnoses. This study is focused on creating a novel approach for distinguishing between benign and malignant tumors in dermoscopic skin images. The procedure starts with image preprocessing techniques, including enhancement, normalization, segmentation, and zero-padding. Then, the extracted features were implemented by fusing convolutional neural networks and gray-level co-occurrence matrix features. Finally, a support vector machine model classifies the extracted features as benign or malignant. A dataset of 900 dermoscopic images of the skin was classified with 97.1% accuracy. The area under the curve was 98.1%. Combining gray-level co-occurrence matrix and convolutional neural network features with a large publicly available dataset improves detection accuracy for skin cancer.