Related Experiment Video
Updated: Jun 24, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
Hybrid deep learning and optimal graph search method for optical coherence tomography layer segmentation in diseases
Zhi Chen1,2, Honghai Zhang1,2, Edward F Linton3
1Iowa Institute for Biomedical Imaging, University of Iowa, Iowa City, IA 52242, USA.
Biomedical Optics Express
|June 13, 2024
Summary
Deep LOGISMOS accurately segments retinal layers in OCT images, outperforming existing methods for optic nerve disease assessment. This advanced deep learning approach improves robustness and generalizability for precise disease diagnosis and management.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Accurate segmentation of retinal layers in optical coherence tomography (OCT) is vital for diagnosing optic nerve diseases.
- Existing automated methods struggle with pathological variations, like extreme thinning of the ganglion cell-inner plexiform layer (GCIPL).
Purpose of the Study:
- To develop and evaluate Deep LOGISMOS, a hybrid deep learning and 3D graph search algorithm, for improved retinal layer segmentation.
- To enhance accuracy, robustness, and generalizability in segmenting OCT images, particularly in cases of irregular layer topology.
Main Methods:
- Deep LOGISMOS was trained on 124 OCT volumes from non-arteritic anterior ischemic optic neuropathy (NAION) patients.
- The algorithm was tested on cross-sectional datasets (NAION, glaucoma, multiple sclerosis/control) and a longitudinal glaucoma dataset.
- Performance was evaluated using Dice similarity coefficients and compared against established algorithms (Iowa, nnU-Net).
Main Results:
- Deep LOGISMOS achieved high Dice scores for GCIPL segmentation (e.g., 94.06% in Test-JHU).
- It significantly outperformed Iowa reference algorithms (1.0–17.5% improvement) and nnU-Net (1.0–4.4% improvement).
- Reliable longitudinal GCIPL thickness measurements were demonstrated in severe glaucoma cases over five years.
Conclusions:
- Deep LOGISMOS offers a robust and accurate solution for retinal layer segmentation in OCT images.
- The method shows significant potential for precise quantification of retinal structures.
- This aids in the diagnosis and management of optic nerve diseases, including glaucoma and NAION.
