Predicting post-operative vault and optimal implantable collamer lens size using machine learning based on various ophthalmic device combinations

A Deep Learning Model for Automatically Quantifying the Anterior Segment in Ultrasound Biomicroscopy Images of Implantable Collamer Lens Candidates

OBJECTIVE

This study aimed to develop and evaluate a deep learning-based model that could automatically measure anterior segment (AS) parameters on preoperative ultrasound biomicroscopy (UBM) images of implantable Collamer lens (ICL) surgery candidates.

METHODS

A total of 1164 panoramic UBM images were preoperatively obtained from 321 patients who received ICL surgery in the Eye Center of Renmin Hospital of Wuhan University (Wuhan, China) to develop an imaging database. First, the UNet++ network was utilized to segment AS tissues automatically, such as corneal lens and iris. In addition, image processing techniques and geometric localization algorithms were developed to automatically identify the anatomical landmarks (ALs) of pupil diameter (PD), anterior chamber depth (ACD), angle-to-angle distance (ATA), and sulcus-to-sulcus distance (STS). Based on the results of the latter two processes, PD, ACD, ATA, and STS can be measured. Meanwhile, an external dataset of 294 images from Huangshi Aier Eye Hospital was employed to further assess the model's performance in other center. Lastly, a subset of 100 random images from the external test set was chosen to compare the performance of the model with senior experts.

RESULTS

Whether in the internal test dataset or external test dataset, using manual labeling as the reference standard, the models achieved a mean Dice coefficient exceeding 0.880. Additionally, the intra-class correlation coefficients (ICCs) of ALs' coordinates were all greater than 0.947, and the percentage of Euclidean distance distribution of ALs within 250 μm was over 95.24%.While the ICCs for PD, ACD, ATA, and STS were greater than 0.957, furthermore, the average relative error (ARE) of PD, ACD, ATA, and STS were below 2.41%. In terms of human versus machine performance, the ICCs between the measurements performed by the model and those by senior experts were all greater than 0.931.

CONCLUSION

A deep learning-based model could measure AS parameters using UBM images of ICL candidates, and exhibited a performance similar to that of a senior ophthalmologist.

VAULT: vault accuracy using deep learning technology: new image-based artificial intelligence model for predicting implantable collamer lens postoperative vault

PURPOSE

To develop an accurate deep learning model to predict postoperative vault of phakic implantable collamer lenses (ICLs).

SETTING

Parkhurst NuVision LASIK Eye Surgery, San Antonio, Texas.

DESIGN

Retrospective machine learning study.

METHODS

437 eyes of 221 consecutive patients who underwent ICL implantation were included. A neural network was trained on preoperative very high-frequency digital ultrasound images, patient demographics, and postoperative vault.

RESULTS

3059 images from 437 eyes of 221 patients were used to train the algorithm on individual ICL sizes. The 13.7 mm size was excluded because of insufficient data. A mean absolute error of 66.3 μm, 103 μm, and 91.8 μm were achieved with 100%, 99.0%, and 96.6% of predictions within 500 μm for the 12.1 mm, 12.6 mm, and 13.2 mm sizes, respectively.

CONCLUSIONS

This deep learning model achieved a high level of accuracy of predicting postoperative ICL vault with the overwhelming majority of predictions successfully within a clinically acceptable margin of vault.

Enhancing Vault Prediction and ICL Sizing Through Advanced Machine Learning Models

PURPOSE

To use artificial intelligence (AI) technology to accurately predict vault and Implantable Collamer Lens (ICL) size.

METHODS

The methodology focused on enhancing predictive capabilities through the fusion of machine-learning algorithms. Specifically, AdaBoost, Random Forest, Decision Tree, Support Vector Regression, LightGBM, and XGBoost were integrated into a majority-vote model. The performance of each model was evaluated using appropriate metrics such as accuracy, precision, F1-score, and area under the curve (AUC).

RESULTS

The majority-vote model exhibited the highest performance among the classification models, with an accuracy of 81.9% area under the curve (AUC) of 0.807. Notably, LightGBM (accuracy = 0.788, AUC = 0.803) and XGBoost (ACC = 0.790, AUC = 0.801) demonstrated competitive results. For the ICL size prediction, the Random Forest model achieved an impressive accuracy of 85.3% (AUC = 0.973), whereas XG-Boost (accuracy = 0.834, AUC = 0.961) and LightGBM (accuracy = 0.816, AUC = 0.961) maintained their compatibility.

CONCLUSIONS

This study highlights the potential of diverse machine learning algorithms to enhance postoperative vault and ICL size prediction, ultimately contributing to the safety of ICL implantation procedures. Furthermore, the introduction of the novel majority-vote model demonstrates its capability to combine the advantages of multiple models, yielding superior accuracy. Importantly, this study will empower ophthalmologists to use a precise tool for vault prediction, facilitating informed ICL size selection in clinical practice. [J Refract Surg. 2024;40(3):e126-e132.].

Predictability comparison of sizing parameters for postoperative vault after implantable Collamer lens implantation

PURPOSE

This study aims to assess the accuracy of three parameters (white-to-white distance [WTW], angle-to-angle [ATA], and sulcus-to-sulcus [STS]) in predicting postoperative vault and to formulate an optimized predictive model.

METHODS

In this retrospective study, a cohort of 465 patients (comprising 769 eyes) who underwent the implantation of the V4c implantable Collamer lens with a central port (ICL) for myopia correction was examined. Least absolute shrinkage and selection operator (LASSO) regression and classification models were used to predict postoperative vault. The influences of WTW, ATA, and STS on predicting the postoperative vault and ICL size were analyzed and compared.

RESULTS

The dataset was randomly divided into training (80%) and test (20%) sets, with no significant differences observed between them. The screened variables included only seven variables which conferred the largest signal in the model, namely, lens thickness (LT, estimated coefficients for logistic least absolute shrinkage of -0.20), STS (-0.04), size (0.08), flat K (-0.006), anterior chamber depth (0.15), spherical error (-0.006), and cylindrical error (-0.0008). The optimal prediction model depended on STS (R2=0.419, RMSE=0.139), whereas the least effective prediction model relied on WTW (R2=0.395, RMSE=0.142). In the classified prediction models of the vault, classification prediction of the vault based on STS exhibited superior accuracy compared to ATA or WTW.

CONCLUSIONS

This study compared the capabilities of WTW, ATA, and STS in predicting postoperative vault, demonstrating that STS exhibits a stronger correlation than the other two parameters.

Predicting Implantable Collamer Lens Vault Using Machine Learning Based on Various Preoperative Biometric Factors

Purpose

To predict the vault size after Implantable Collamer Lens (ICL) V4c implantation using machine learning methods and to compare the predicted vault with the conventional manufacturer's nomogram.

Methods

This study included 707 patients (707 eyes) who underwent ICL V4c implantation at the Department of Ophthalmology, Peking Union Medical College Hospital, from September 2019 to January 2022. Random Forest Regression (RFR), XGBoost, and linear regression (LR) were used to predict the vault size 1 week after ICL V4c implantation. The mean absolute error (MAE), median absolute error (MedAE), root mean square error (RMSE), symmetric mean absolute percentage error (SMAPE), and Bland-Altman plot were utilized to compare the prediction performance of these machine learning methods.

Results

The dataset was divided into a training set of 180 patients (180 eyes) and a test set of 527 patients (527 eyes). XGBoost had the lowest prediction error, with mean MAE, RMSE, and SMAPE values of 121.70 µm, 148.87 µm, and 19.13%, respectively. The Bland‒Altman plots of RFR and XGBoost showed better prediction consistency than LR. However, XGBoost showed narrower 95% limits of agreement (LoA) than RFR, ranging from -307.12 to 256.59 µm.

Conclusions

XGBoost demonstrated better predictive performance than RFR and LR, as it had the lowest prediction error and the narrowest 95% LoA. Machine learning may be applicable for vault prediction, and it might be helpful for reducing the complications and the secondary surgery rate.

Translational Relevance

Using the proposed machine learning model, surgeons can consider the postoperative vault to reduce the surgical complications.

Comparison of the Accuracy of Seven Vault Prediction Formulae for Implantable Collamer Lens Implantation

INTRODUCTION

This study aimed to compare the accuracy of seven implantable collamer lens (ICL) implantation vault prediction formulae.

METHODS

We retrospectively analyzed 328 patients (328 eyes) who underwent ICL implantation and the prediction accuracy of seven formulae: NK, KS, WH, Luo, Zhu, Hun, and ZZ were compared. Moreover, the accuracy of the seven formulae for different ICL sizes was compared. The formulae were tested using mean absolute prediction error (MAE), median absolute prediction error (MedAE), prediction error (PE) percentages at ± 50 µm, ± 100 µm, ± 200 µm, and ± 300 µm, and Bland-Altman analysis.

RESULTS

The PE of the seven formulae were statistically significant (P < 0.001). The KS (101.00 µm) and WH formulae (116.65 µm) had the smallest MedAE, followed by the Luo (123.62 µm), NK (141.50 µm), Hun (152.68 µm), ZZ (196.00 µm) and Zhu formula (225.98 µm). The highest percentage of PE in the range of ± 300µm was 94.3% and 93% for the KS and WH formulae, respectively. Among the different ICL size groupings, the KS formula predicted the smallest MedAE for 12.1 mm and 12.6 mm, whereas the Luo and WH formulae predicted the smallest MedAE for 13.2 mm and 13.7 mm, respectively.

CONCLUSIONS

The KS and WH formulae provided better outcomes by predicting the vault with higher accuracy than of the NK, Hun, Luo, ZZ, and Zhu formulae.

TRIAL REGISTRATION

ChiCTR2200065501.

Ocular features in myopic eyes with longer horizontal ciliary sulcus diameters for intended implantable collamer lens surgery

PURPOSE

To assess the ocular anterior segment characteristics in myopic eyes intended for ICL surgery with horizontal ciliary sulcus-to-sulcus (STS) diameters being greater than vertical STS diameters.

METHODS

This retrospective, comparative case study included 1230 eyes of patients who underwent ICL implantation for the treatment of myopia or myopic astigmatism at the Zhongshan Ophthalmic Center from September 2020 to November 2021. The myopic eyes were divided into two groups according to the relatively long diameter of the ciliary sulcus. General parameters and anterior chamber parameters were compared between the two groups.

RESULTS

1230 eyes of 694 patients were included. The proportion of myopic eyes with longer horizontal STS diameters was 4.63%. Horizontal STS distances exceeding vertical meridians in these eyes were mainly attributed to the shortening of vertical STS distances (horizontal STS: P = 0.112; vertical STS: P < 0.001). Eyes with longer horizontal meridians of the ciliary sulcus displayed larger steep keratometry value (P = 0.001), larger corneal volume (P = 0.002), larger corneal astigmatism (P < 0.001), larger ocular residual astigmatism (P = 0.017), worse visual acuity (logMAR UDVA: P = 0.021; logMAR CDVA: P = 0.001), and more iridociliary cysts (P = 0.017) compared to eyes with vertically oval shapes.

CONCLUSION

Myopic eyes with longer horizontal STS diameters are commonly accompanied by a change in corneal morphology and more iridociliary cysts. The anatomical features of the ciliary sulcus should be given sufficient consideration to ICL size and placement selection, contributing to more personalized and precise surgery.