Anatomical prognosis after idiopathic macular hole surgery: machine learning based-predection

Area of Science:

• Ophthalmology
• Medical Imaging
• Machine Learning

Background:

• Idiopathic macular holes (MH) pose a surgical challenge.
• Vitrectomy with inverted internal limiting membrane (ILM) peeling is a common surgical approach.
• Predicting surgical outcomes is crucial for patient management.

Purpose of the Study:

• To develop a machine learning (ML) model for predicting macular hole (MH) status 9 months post-vitrectomy and inverted ILM peeling.
• To evaluate the efficacy of preoperative OCT parameters in predicting MH closure.
• To assess the performance of a medical decision support system (MDSS) in forecasting surgical success.

Main Methods:

• Retrospective analysis of 120 eyes from 114 patients undergoing MH surgery.
• Acquisition and analysis of 510 B-scan macular OCT images 9 months post-surgery.
• Calculation of various MH indices (e.g., diameter, DHI, area index) from preoperative OCT scans.
• Development and validation of an ML model using ROC curve analysis and calculating AUC and kappa values.

Main Results:

• MH indices like MH diameter, diameter hole index (DHI), MH index, and hole formation factor predicted successful MH closure.
• Basal diameter, DHI, and MH area index were indicative of non-closure.
• The developed MDSS achieved a high Area Under the Curve (AUC) of 0.984 and a kappa value of 0.934.

Conclusions:

• Preoperative OCT parameters, analyzed by an ML model, can accurately predict MH outcomes after pars plana vitrectomy with inverted ILM peeling.
• The MDSS demonstrates remarkable accuracy in predicting MH closure.
• This ML-based MDSS has the potential to optimize surgical planning for full-thickness macular hole patients.