An Investigation of Distance Measures for Development of Effective Content Based Tumor Image Retrieval System

Abstract

Background/Objectives: The MRI has proven to be extremely effective in detecting tumors, with millions of images created each day throughout the world. To find similar images from a vast collection, Content-Based Tumor Image Retrieval (CBTIR) technology has been used to analysis the medical image. In the traditional retrieval methods, retrieving a similar image from the large database is crucial task. To overcome this issue we developed deep learning based retrieval method. Methods: This research offers a retrieval approach based on predefined ResNet models for quick and accurate image retrieval. We tested various prominent ResNet models with different distance similarity metrics, and the best option was determined by this work. Findings: After the various evaluation of ResNet models with varied distance measures on the CE-MRI data set, ResNet50 model applied with Hamming distance yields 99.33% of retrieval precision. Novelty: This work used predefined ResNet models with the combination of Distance similarity metrics to achieve more accurate results on medical image retrieval compared to the other conventional methods. Keywords: Content Based Image Retrieval, Tumor Retrieval, Hamming Distance, Euclidean Distance, Minkowski