Related Experiment Video
Updated: Dec 26, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.3K
A generative adversarial network approach to predicting postoperative appearance after orbital decompression surgery
Tae Keun Yoo1, Joon Yul Choi2, Hong Kyu Kim3
1Department of Ophthalmology, Aerospace Medical Center, Republic of Korea Air Force, Cheongju, South Korea.
Computers in Biology and Medicine
|March 17, 2020
Summary
Deep learning using generative adversarial networks (GANs) can create realistic postoperative images for thyroid-associated ophthalmopathy (TAO) patients undergoing orbital decompression surgery. This AI tool aids patient decision-making, though image quality requires further enhancement.
Area of Science:
- Ophthalmic plastic surgery
- Artificial intelligence in medicine
- Medical image synthesis
Background:
- Thyroid-associated ophthalmopathy (TAO) often requires orbital decompression surgery to prevent vision loss and reduce proptosis.
- Postoperative appearance changes significantly, complicating surgical decision-making for patients.
- Realistic visualization of surgical outcomes is crucial for patient consultation.
Purpose of the Study:
- To develop a deep learning technique for synthesizing realistic postoperative facial images following orbital decompression surgery for TAO.
- To create a tool that aids patients in visualizing potential outcomes and making informed decisions about surgery.
- To explore the utility of generative adversarial networks (GANs) in medical image synthesis for plastic and cosmetic surgery.
Main Methods:
- A conditional generative adversarial network (GAN) was employed to transform preoperative facial images into predicted postoperative images.
- The GAN model was trained on 109 pairs of pre- and postoperative facial images, utilizing data augmentation techniques.
- The model synthesized images by altering conditional variables, simulating the transition from preoperative to postoperative appearance.
Main Results:
- Synthesized postoperative images closely resembled ground truth postoperative images.
- The GAN-based synthesized images improved the performance of deep learning models in classifying pre- and postoperative status, even with limited training data.
- Clinicians noted that the quality of the synthesized images was relatively low, indicating room for improvement.
Conclusions:
- The developed deep learning framework successfully synthesized TAO facial images based on orbital decompression status.
- Synthesized images can potentially assist patients in understanding the impact of decompression surgery.
- While image quality needs enhancement, GANs show promise as decision support tools for plastic and cosmetic surgery by rapidly generating realistic postoperative appearance simulations.
