Related Experiment Video
Updated: Jun 16, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
NCME-Net: Nuclear cataract mask encoder network for intelligent grading using self-supervised learning from anterior
Jiani Zhao1, Cheng Wan1, Jiajun Li2
1College of Electronic and Information Engineering /College of Integrated Circuits, Nanjing University of Aeronautics and Astronautics, Nanjing, Jiangsu, 211106, China.
Heliyon
|August 16, 2024
Summary
A new deep learning model, the nuclear cataract mask encoder network (NCME-Net), effectively grades nuclear cataract severity using self-supervised pretraining. This approach improves diagnostic accuracy, especially with limited medical data.
Area of Science:
- Ophthalmology
- Artificial Intelligence
- Medical Imaging
Background:
- Cataracts are a primary cause of global blindness, necessitating precise diagnosis and surgical planning.
- Deep learning models struggle with medical images due to limited labeled data and high interclass similarity.
- Self-supervised pretraining addresses these limitations by reducing annotation costs and domain disparity.
Purpose of the Study:
- To develop an intelligent grading system for nuclear cataract severity.
- To propose a novel hybrid model, the nuclear cataract mask encoder network (NCME-Net), for four-class nuclear cataract analysis.
- To evaluate the effectiveness of self-supervised pretraining in medical image analysis.
Main Methods:
- A hybrid model, NCME-Net, was developed utilizing self-supervised pretraining.
- A dataset of 792 nuclear cataract images was used for training, validation, and testing.
- The impact of various self-supervised tasks on semantic information extraction was investigated.
Main Results:
- NCME-Net achieved a diagnostic accuracy of 91.0% on the test set.
- This represents a 5.0% improvement over the ResNet50 model.
- Image restoration tasks within self-supervised learning significantly enhanced semantic information extraction.
Conclusions:
- NCME-Net demonstrates strong performance in distinguishing cataract severities, particularly with limited sample sizes.
- The model offers a valuable tool for intelligent cataract diagnosis.
- Self-supervised pretraining is crucial for enhancing deep learning model performance in medical imaging tasks.
