| Abstrakt: |
Today's advanced medical facilities have created a space for a better understanding of many health-related issues such as fractures, tumors, infections, etc. through augmented digital imaging. High-quality medical imaging technologies, which includes Magnetic Resonance Imaging (MRI), Positron Emission Tomography (PET), Computerized Tomography (CT) scans, and X-rays, has enriched our understanding of various health ailments such as fractures, tumors, and infections. These imaging techniques provide a considerable advantage in early disease detection by analysing visual cues. The objective of this work is to find the Region-of Attention (RoA) or the saliency region of an image using deep learning i.e., U-Net for precise RoA segmentation. The saliency region is enhanced and fused with the original image to generate a new enhanced image. The CNN features of this enhanced image are used for the similarity finding and image indexing. In this work a novel framework, termed Saliency-Region Enhanced Content-Based Image Retrieval (SRECBIR) is proposed. The efficacy of the SRE-CBIR model was tested on two biomedical image datasets, specifically, Brain tumor and COVID-19 datasets. The classification results show there are enhancements in the No-tumor and Covid-negative cases. The average precision value of original Brain tumor dataset is 94.3% where as it is 95.0% for the enhanced Brain tumor dataset, likewise the average precision value of original COVID19 dataset is 97.0%, which is 98.0% when tested for the enhanced COVID19 dataset. Retrieval results using CNN features of the enhanced images outperform the retrieval of the CNN features of the original images in terms of class-wise as well as the average retrieval rate. [ABSTRACT FROM AUTHOR] |
