Code-Free Machine Learning Approach for EVO-ICL Vault Prediction: A Retrospective Two-Center Study.

Area of Science:

• Ophthalmology
• Medical Machine Learning
• Data Science

Background:

• Machine learning development is challenging for medical researchers due to coding barriers.
• Accurate prediction of postoperative vault is crucial for successful implantable collamer lens (ICL) sizing.

Purpose of the Study:

• To enhance the accuracy of a postoperative vault prediction model for ICL sizing.
• To develop a machine learning model without coding experience for medical researchers.

Main Methods:

• Utilized Orange data mining, an open-source, code-free machine learning tool.
• Trained a random forest model using eye-pair data from 294 patients (B&VIIT Eye Center) and 26 patients (Kim's Eye Hospital).
• Validated the model through 10-fold cross-validation (internal) and external validation using data from Kim's Eye Hospital.

Main Results:

• The no-code random forest model achieved mean absolute errors of 124.8 µm (internal) and 152.4 µm (external).
• Achieved areas under the curve of 0.725 (internal) and 0.760 (external) for high vault prediction (>750 µm).
• The developed model demonstrated superior performance compared to classic statistical regression and Google no-code platforms.

Conclusions:

• A no-code machine learning tool accurately predicts postoperative vault for ICL implantation.
• The developed model offers improved accuracy over traditional regression and other no-code approaches.
• Customized no-code machine learning nomograms can enhance ICL implantation accuracy, addressing inter-clinic measurement and surgical biases.