Deep learning-based postoperative visual acuity prediction in idiopathic epiretinal membrane.
Dejia Wen1, Zihao Yu1, Zhengwei Yang1
1Tianjin Key Laboratory of Retinal Functions and Diseases, Tianjin Branch of National Clinical Research Center for Ocular Disease, Eye Institute and School of Optometry, Tianjin Medical University Eye Hospital, 251 Fukang Road, Tianjin, 300384, China.
BMC Ophthalmology
|August 20, 2023
Summary
A new deep learning model using optical coherence tomography (OCT) accurately predicts visual outcomes after surgery for idiopathic epiretinal membrane (iERM). This OCT-based deep learning (DL) model shows promise for surgical planning.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Idiopathic epiretinal membrane (iERM) affects visual acuity.
- Predicting postoperative visual outcomes is crucial for surgical planning.
- Current prediction methods may lack precision.
Purpose of the Study:
- To develop a deep learning (DL) model for predicting 6-month postoperative visual acuity.
- To utilize preoperative optical coherence tomography (OCT) imaging for model training.
- To assess the model's accuracy and potential clinical utility in iERM patients.
Main Methods:
- A retrospective cohort study included 442 eyes with iERM.
- Deep learning models (Inception-Resnet-v2) were trained using macular OCT images and clinical data.
- A multimodal deep fusion network (MDFN) approach was employed.
- Model performance was evaluated against a regression model using a dedicated testing dataset.
Main Results:
- The DL model achieved a mean absolute error of 0.070 logMAR and RMSE of 0.11 logMAR.
- The DL model's predictive performance (R²=0.80) significantly outperformed the regression model (R²=0.49).
- Over 94% of predictions were within ±0.20 logMAR error.
Conclusions:
- OCT-based DL models can accurately predict postoperative visual acuity in iERM patients.
- The developed DL model demonstrates high sensitivity and accuracy.
- This technology holds potential for integration into surgical planning workflows.
