Prediction of postoperative visual acuity after vitrectomy for macular hole using deep learning-based artificial...

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Macular holes (MHs) are a common cause of visual impairment.
• Vitrectomy is a standard surgical treatment for MHs.
• Predicting postoperative visual acuity (VA) is crucial for patient management.

Purpose of the Study:

• To develop a deep learning (DL) model for predicting postoperative VA after MH surgery.
• To utilize preoperative optical coherence tomography (OCT) images as input for the DL model.
• To compare the predictive accuracy of the DL model against a multivariate linear regression model.

Main Methods:

• Retrospective analysis of 259 eyes undergoing vitrectomy for MHs.
• Eyes were categorized into four groups based on 6-month postoperative VA.
• A DL network was trained on preoperative OCT images and corresponding VA data.
• A multivariate linear regression model used preoperative VA, MH size, and age.

Main Results:

• The DL model achieved a precision of 46%, outperforming the multivariate model's 40% precision.
• Both models showed a significant correlation between predicted and actual postoperative VA (P < .0001).
• The DL model demonstrated a stronger correlation (r = .62) compared to the multivariate model (r = .55).

Conclusions:

• Deep learning models can accurately predict postoperative VA after MH surgery using preoperative OCT images.
• The DL approach offers superior predictive accuracy compared to traditional multivariate regression.
• This AI-driven prediction holds potential for improving patient counseling and surgical planning.