Related Experiment Video
Updated: Apr 8, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.7K
Deep Learning for Content-Based Medical Image Retrieval in Picture Archiving and Communication Systems for Brain
Chin-Lin Lee1, Tzu-Hsuan Hsu1, Yu-Te Wu2
1Department of Information Management, National Taipei University of Nursing and Health Science, No. 365, Ming-te Rd, Beitou Dist, Taipei City, 112303, Taiwan, +886 2-2822-7101 ext 1230.
JMIR Medical Informatics
|April 6, 2026
Summary
This study introduces a content-based medical image retrieval (CBMIR) system for brain MRI, improving clinical workflow efficiency. The system successfully integrates with PACS, enhancing retrieval accuracy for radiologists.
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Healthcare
- Computer Science
Background:
- Traditional text-based search is inadequate for large medical imaging archives.
- Content-Based Medical Image Retrieval (CBMIR) offers visual search but faces integration challenges with Picture Archiving and Communication Systems (PACS).
- Deep learning advancements in feature extraction have not translated to widespread CBMIR integration in radiology information systems due to protocol barriers.
Purpose of the Study:
- Develop a CBMIR system for 7 types of brain tumors in brain MRI scans.
- Enhance clinical workflow and provide quantitative decision support for radiologists through efficient image retrieval.
- Facilitate evidence-based case comparison and improve retrieval efficiency, rather than directly improving diagnostic accuracy.
Main Methods:
- Utilized a deep learning-based feature extraction algorithm (GoogLeNet with generalized mean pooling and an embedding layer) for CBMIR.
- Developed a system tailored for retrieving 7 distinct types of brain tumors from brain MRI.
- Integrated the CBMIR system into a PACS environment by harmonizing two open-source projects, overcoming protocol barriers.
Main Results:
- The CBMIR system achieved a mean average precision of 89.16% and a Precision@10 score of 94.08% on a dataset of 15,873 brain MRI images from 658 participants.
- Demonstrated the performance and robustness of the deep learning-based feature extraction for medical image retrieval.
- Successfully integrated the CBMIR system into a PACS environment, validating its practical applicability.
Conclusions:
- Designed and implemented a PACS-integrated CBMIR system for brain MRI.
- The system enables efficient and accurate retrieval of medical images within a clinical workflow.
- The successful integration addresses a significant bottleneck in utilizing advanced CBMIR tools in clinical practice.