Deep Learning for Skin Cancer: Hybrid Feature Extraction with GoogleNet and MobileNet

Detecting skin cancer is a crucial task in dermatology, where the use of deep learning models has showed promising results. This study explores various hybrid convolutional neural network (CNN) architectures for skin cancer detection, leveraging the complementary strengths of different models. Experimentation includes combinations such as ResNet with MobileNet, MobileNet with VGG, VGG with ResNet, etc.. Additionally, a novel hybrid model is proposed, consisting of GoogleNet (InceptionV3) with MobileNet as the feature extraction backbone. Extracted features are then fed into a feedforward neural network (FFNN) for classification. The effectiveness of each hybrid model was assessed based on the accuracy, sensitivity, and specificity through comprehensive testing of a skin cancer dataset. The final proposed model, GoogleNet with MobileNet, achieves an accuracy of 87.9%, demonstrating its effectiveness in the accurate identification of malignant and benign skin lesions. This study highlights the potential of hybrid CNN architectures in enhancing skin cancer detection capabilities and underscores the importance of leveraging diverse model combinations for improved diagnostic accuracy in clinical practice.