The collections of various digital image databases have significantly grown and many users have recognized that finding and recovering important images from large collections is a difficult task. Where the success of any image retrieval system is heavily dependent on the feature extraction capacity of the feature descriptor, therefore successful and effective retrieval method has been developed to provide an effective and rapid search and retrieval process. We present a unique deep learning-based approach for extracting high-level and compact features from medical images in this paper. To capture the discriminative features of medical images, we use Residual Networks (ResNets), a popular multi-layered deep neural network. The query is then broadened by reformulating the query image using the mean values for deep features from each database class's top-ranking images. Two publicly available databases in various forms were used to evaluate the performance of our technique. These studies demonstrated the benefits of our proposed strategy, with retrieval accuracy greatly improved.