CNN-SVM hybrid approaches for melanoma detection in dermatological images

Melanoma is an aggressive form of skin cancer that if left undiagnosed promptly will lead to death. Diagnosing melanoma in its early stages is essential to increase the survival rate among patients in recent years, machine learning methods such as Convolutional Neural Networks and Support Vector Machines have proven effective in automatic melanoma detection. We present a hybrid approach that takes advantage of the CNN and SVM methods for accurate and efficient melanoma detection in dermoscopic images. Initially, the images are pre-processed and features can be extracted using CNN which is further forwarded to SVM for classification. In hybrid approach CNN features with high classification accuracy combine into SVM for solving these problems. Data augmentation is applied to fully take advantage of the available bandwidth sources and improve model performance (add more diversity into training dataset). We demonstrate the efficacy of this approach on a publicly available dermatology image dataset and compare its performance to that of existing methods. The results demonstrate the advantage of our formula over individual CNN and SVM models, giving rise to a better accuracy and specificity in detecting melanomas. The hybrid method introduced for detecting and delineating melanoma in early stages can be a significant support to assist dermatologists in effective treatment outcomes.