Detection of Skin Cancer Through Dermoscopy Images Using Hybrid Deep Feature Extraction

A common and dangerous type of cancer that affects millions of people each year is known as Skin Cancer. Skin cancer, like melanoma, is difficult to identify because of its resemblance with benign moles. Melanoma early detection is very necessary as it increases the chances of survival. With technological advancement, AI has emerged as a powerful technique in the medical field. Machine- deep learning and neural networks have shown promising results in automating skin cancer detection using dermoscopy images. This paper proposes an efficient skin cancer detection technique through dermoscopy images using Hybrid Deep Feature Extraction. A publicly available dataset is used in this paper. Data preprocessing technique: Morphological closing and gamma correction are applied to it. Then, feature extraction with handcraft methods HOG, LBP, and automated methods DenseNet201 and ResNet50 will be performed on the preprocessed dataset. All the extracted features from these algorithms are concatenated into a single array. Different feature extraction combinations are tested with SVM, Random Forest, Artificial Neural Network, XGBoost, and LightGBM classifiers. After classification, the results are measured using the performance evaluation metrics: accuracy, precision, recall, and F1 score. All models give efficient results, but the classifier ANN with the feature extraction combination HOG and DenseNet201 are giving the best result with the highest accuracy, recall, precision, and F1-score as 0.988, 0.988, 0.988, and 0.988, respectively.