Advanced Deep Learning Approaches for Content-Based Image Retrieval System in Skin Cancer Diagnostics

The objective of this paper is to develop and assess the performance of a deep learning model-based content skin cancer detection image retrieval system. In this regard, CNN, VGGNet, ResNet, DenseNet, InceptionNet, and MobileNet were tested on recall, accuracy, mean average precision, F1, and precision metrics. VGGNet performed best among them, with an accuracy of 92.3%, outperforming all other models in precision, F1-score, and recall. CNN also gave the best performance with 91.5% accuracy. The CBIR system retrieved relevant images for the given lesions, especially melanoma lesions, which were relevant and high in precision within the top 5 retrieved images. These results also confirm the efficiency of CNN and VGGNet in enhancing accuracy during the retrieval of skin cancer diagnosis images.