VCCINet: A Multi-model Deep Learning Approach for Melanoma Skin Cancer Detection

Recent years have witnessed an increase in skin cancer cases globally, with 2–3 million non-melanoma skin cancers and 132,000 melanoma skin cancer cases occurring annually. Early detection is crucial for effective treatment, and deep learning techniques offer a promising avenue for diagnosis. However, several existing individual deep learning models used for skin cancer detection, such as Convolutional Neural Networks (CNN), Visual Geometry Group-based Neural Networks (VGG16), Capsule Networks (CapsNet), and InceptionV3 can often deliver suboptimal results on multiple testing parameters, affecting the model's overall performance. Hence, this paper proposes a multi-model approach called VCCINet, which combines four deep-learning techniques (CNN, Feature Extracted VGG16, CapsNet, and InceptionV3) by ensemble learning, to help improve the performance of the individual models. It also introduces a further extension on the VCCINet model, called FT-VCCINet, by implementing feature extraction (on VGG16) and fine-tuning (on CapsNet and Inceptionv3) before using ensemble learning on the models and achieving a better result than all the individual models and VCCINet. The accuracy of VCCINet and FTVCCINet came out to be 93.18% and 93.02% respectively, which is significantly higher compared to individual model accuracies. Furthermore, the multi-model approach showed exponential improvements in other parameters such as Precision, Recall, F1Score, and Specificity.