Related Experiment Video
Updated: Dec 27, 2025
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
Calculating the target exposure index using a deep convolutional neural network and a rule base.
Takeshi Takaki1, Seiichi Murakami2, Ryo Watanabe3
1Department of Health Sciences, Graduate School of Medical Sciences, Kyushu University, 3-1-1 Maidashi, Higashi-ku, Fukuoka 812-8582, Japan; Department of Radiology, Hospital of University of Occupational and Environmental Health, Iseigaoka 1-1, Yahatanishi-ku, Kitakyushu-shi, Fukuoka 807-8555, Japan.
This study introduces a deep convolutional neural network (DCNN) and rule-based method to automatically assess chest X-ray image quality. The system achieved 81% accuracy, determining optimal exposure indices without visual review.
Area of Science:
- Medical Imaging Analysis
- Artificial Intelligence in Radiology
- Radiographic Quality Assessment
Background:
- Visual assessment of chest X-ray image quality is subjective and time-consuming.
- Automated methods are needed to objectively evaluate image quality and ensure diagnostic accuracy.
- Exposure Index (EI) is a critical parameter for image quality in digital radiography.
Purpose of the Study:
- To develop and validate a method for determining chest X-ray image quality using a deep convolutional neural network (DCNN) and a rule-based system.
- To establish a system for calculating the minimum diagnosable exposure index (EI) and the target exposure index (EIt) without human visual assessment.
- To automate the quality control process in digital chest radiography.
Main Methods:
- Utilized transfer learning with GoogLeNet (a DCNN) to analyze lung fields, mediastinum, and spine.
- Developed three detectors for local image region quality rating.
- Implemented a rule-based technique, informed by expert assessment, to determine overall image quality.
- Calculated minimum EI based on the distribution of suitable and non-suitable EI values to ascertain EIt.
Main Results:
- The combined DCNN and rule-based method achieved an accuracy rate of 81% in discriminating image quality.
- The minimum diagnosable exposure index (EI) was determined to be 230.
- The target exposure index (EIt) was established at 288.
Conclusions:
- The proposed automated method effectively discriminates varying chest X-ray image qualities without requiring visual inspection.
- The system successfully determined both the minimum EI for diagnosis and the target EIt.
- This approach offers a reliable, objective tool for enhancing radiographic quality control.
