Related Experiment Video
Updated: Nov 11, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.1K
Development of a deep learning-based image quality control system to detect and filter out ineligible slit-lamp
Zhongwen Li1, Jiewei Jiang2, Kuan Chen3
1Ningbo Eye Hospital, Wenzhou Medical University, Ningbo, 315000, China; School of Ophthalmology and Optometry and Eye Hospital, Wenzhou Medical University, Wenzhou, 325027, China.
Computer Methods and Programs in Biomedicine
|March 25, 2021
Summary
A new deep learning-based image quality control system (DLIQCS) automatically detects and filters ineligible slit-lamp images. This system enhances AI diagnostics by ensuring only high-quality images are analyzed, improving corneal disease detection accuracy.
Area of Science:
- Ophthalmology
- Artificial Intelligence
- Medical Imaging
Background:
- Existing AI diagnostic systems for corneal diseases rely solely on eligible slit-lamp images.
- Ineligible images (poor-field, defocused, poor-location) are common and can compromise AI analysis accuracy.
- Manual image quality assessment is time-consuming and labor-intensive for large datasets.
Purpose of the Study:
- To develop a deep learning-based image quality control system (DLIQCS).
- To automatically detect and filter out ineligible slit-lamp images.
- To improve the reliability of AI-based corneal disease diagnostic systems.
Main Methods:
- Developed and evaluated a DLIQCS using 48,530 slit-lamp images from 4 institutions.
- Trained and compared three deep learning algorithms: AlexNet, DenseNet121, and InceptionV3.
- Assessed algorithm performance using AUC, sensitivity, specificity, and accuracy for image classification.
Main Results:
- The DenseNet121 algorithm achieved high AUC values (0.997-1.000) for detecting various image quality issues.
- Internal and external validation demonstrated excellent performance across different datasets and camera types.
- The system accurately classified poor-field, defocused, poor-location, and eligible images.
Conclusions:
- The DLIQCS effectively automates the detection and filtering of ineligible slit-lamp images.
- This system can serve as a crucial prescreening tool for AI diagnostic pipelines.
- By ensuring image eligibility, DLIQCS enhances the accuracy and efficiency of AI-driven corneal disease diagnosis.
