Application of Swin Transformer Model to Retrieve and Classify Endoscopic Images

The machine learning community is very interested in image classification and retrieval, especially in the area of computer vision and with an emphasis on medical image retrieval. Numerous machine learning approaches have been used for image retrieval problems and have made as a result of the ongoing developments in techniques like Convolutional Neral Networks (CNN) and Vision Transformers with quite good performances. The Swin Transformer model is used to create a specialized medical image retrieval system in this paper that is well suited to gastric endoscopic pictures. The suggested technique takes advantage of the Swin Transformer model's classification process to create feature vectors by combining fragmented image segments collected from local windows, making it easier to calculate similarity on the Kvasir dataset that we have added some additional images. Empirical results show that the Swin Transformer model retrieves endoscopic images with a remarkable classification accuracy of 90.5% and an 85% mean average precision at top 20 (mAP@20).