Related Experiment Video
Updated: Mar 14, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.6K
Dictionary Pruning with Visual Word Significance for Medical Image Retrieval
Fan Zhang1, Yang Song2, Weidong Cai2
1School of Information Technologies, University of Sydney, Australia; Dept of Radiology, Brigham & Womens Hospital, Harvard Medical School, United States.
Neurocomputing
|October 1, 2016
Summary
This study introduces a novel retrieval method for medical images, enhancing disease diagnosis by identifying subtle visual differences. The approach improves accuracy and efficiency in content-based medical image retrieval (CBMIR).
Area of Science:
- Medical Imaging
- Computer Vision
- Biomedical Informatics
Background:
- Content-based medical image retrieval (CBMIR) faces challenges due to subtle visual variations in anatomical structures.
- Accurate disease diagnosis and treatment planning rely on effective medical image analysis.
Purpose of the Study:
- To propose a novel retrieval method, Pruned Dictionary based on Latent Semantic Topic (PD-LST) description, for improved CBMIR.
- To enhance the identification of discriminative characteristics between medical images.
Main Methods:
- Utilized a bag-of-visual-words (BoVW) model combined with PD-LST.
- Calculated topic-word significance to link low-level visual features with high-level semantics.
- Developed an iterative ranking method to determine overall-word significance for discriminative power.
Main Results:
- The PD-LST method demonstrated improved retrieval accuracy on public medical imaging datasets.
- The proposed approach showed enhanced efficiency in content-based medical image retrieval.
- Identified visual words with significant discriminative power for differentiating medical images.
Conclusions:
- The PD-LST retrieval method effectively addresses challenges in CBMIR caused by minor visual variations.
- This approach offers a more meaningful comparison between medical images by bridging visual features and semantic understanding.
- The PD-LST method represents a significant advancement in medical image retrieval for clinical applications.
