A UBM-Based Big Data Model for Individualized ICL Size Selection in Myopic Eyes

A UBM-Based Big Data Model for Individualized ICL Size Selection in Myopic Eyes | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article A UBM-Based Big Data Model for Individualized ICL Size Selection in Myopic Eyes Jiayu Pan, Yue Wang, Jifan Wang, Ying Li, Xiaohan Chu, Yingshuai Li, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7400290/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background: Accurate prediction of post-operative vault is critical for safe and effective implantation of Implantable Collamer Lenses (ICLs). Traditional sizing methods rely on anterior segment parameters, such as white-to-white (WTW) and anterior chamber depth (ACD), are poorly correlated to the with sulcus-to-sulcus (STS) diameter where the ICL rests. Inappropriate sizing can result in suboptimal vaults, increasing the risk of complications. Ultrasound biomicroscopy (UBM) allows for direct assessment of posterior chamber structures but it has yet to be systematically integrated into predictive models at scale. Methods: We developed and validated a UBM-based, data-driven model for vault prediction using >2,000 ICL implantations performed between January 2023 and January 2025 at a single center. Pre-operative variables extracted from standardized UBM protocols include horizontal and vertical STS (STSH, STSV), lens curvature (LC), Posterior Chamber Angle (CSA) Morphology and Ciliary Process (CP) Size,. A random forest model was trained to identify dominant anatomical predictors. Based on these results, a posterior chamber asymmetry index (CS Diff = |STSH – STSV|) and empirically derived LC thresholds were used to construct a clinically practical ICL size selection chart. This chart was prospectively validated in 286 patients (568 eyes) operated between May and June 2025. Results: Random Forest and SHAP analyses identified STSH, STSV, LCH, and LCV as the most influential predictors. Higher CS Diff were associated with low vaults; greater LC predicted higher vaults. In the validation cohort, the selection map yielded ideal vaults (250–750 μm) in 92.96% of eyes. This represented a significant improvement over the historical WTW+ACD method (ideal rate %, p<0.001). Decision-curve analysis demonstrated superior net clinical benefit across a wide range of threshold probabilities. Subgroup analyses confirmed robust performance, including in those with high-risk phenotypes (older age, shallow ACD, high LC). Conclusions: Posterior chamber geometry—especially STS asymmetry and lens curvature—drives vault variability more than traditional anterior segment metrics. Our validated UBM-guided model enhances individualized ICL sizing and significantly improves clinical outcomes. Adoption of such models into routine practice could reduce re-operations and establish a new standard of care in refractive surgery. Implantable Collamer Lens vault prediction ultrasound biomicroscopy sulcus-to-sulcus posterior chamber anatomy machine learning ICL sizing myopia correction Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Background The Implantable Collamer Lens (ICL) has revolutionised the field of refractive surgery, providing an effective and reversible option for the correction of moderate to high myopia (Kapoor et al., 2025 ; Zhang et al., 2020 ). Compared with corneal refractive procedures such as LASIK and SMILE, ICL implantation preserves corneal biomechanics by avoiding stromal tissue removal, and offers superior visual quality in selected patients (Damgaard et al., 2018 ). As surgical techniques and lens materials continue to evolve, ICL implantation is increasingly being offered to a broader range of patients, including those with thin corneas or high degrees of ametropia (Yao et al., 2024 ; Zhang et al., 2024 ). However, the safety and long-term success of ICL surgery critically depend on the post-operative vault, i.e. the distance between the posterior surface of the implanted ICL and the anterior surface of the crystalline lens (Cui et al., 2023 ; Zhang et al., 2025 ). An ideal vault ensures sufficient space to prevent contact with the natural lens (which could induce anterior subcapsular cataract), while avoiding excessive forward displacement that could increase intraocular pressure or precipitate angle-closure glaucoma (Yang et al., 2023 ). Thus, accurate pre-operative prediction of vault is essential to avoid post-operative complications, re-operations, or ICL explanation. Despite its clinical importance, vault prediction remains an unsolved challenge in ICL surgery. The current FDA-approved sizing approach relies on white-to-white (WTW) corneal diameter and anterior chamber depth (ACD) — parameters easily obtained via optical biometry (Ang et al., 2022 ; Pathak et al., 2024 ). However, these anterior segment measurements are only indirect proxies for the true sulcus-to-sulcus (STS) diameter, where the ICL haptics are actually positioned (Zhu et al., 2022 ). Numerous studies have shown that WTW does not reliably correlate with STS, and that errors in this approximation can lead to significant variability in vault (see in Packer et al., 2016). In our practice, we observed a 2.2% ICL explantation rate during 2023 and 2024, which is consistent with the reported rates of approximately 1–4% in literatures across different centers. Over 65% of our cases were attributable to inaccurate sizing based on the traditional WTW + ACD method (Kane et al., 2016 ). To address these limitations, several alternative formulas have been developed. The NK formula, which combines angle-to-angle (ATA) distance and crystalline lens rise (CLR), has been reported to achieve an accuracy of approximately 80–88% in predicting ideal vault (Nakamura et al., 2023 ). On the other hand, the KS formula focuses on ATA alone; while the ZZ method attempts to incorporate STS and lens thickness (LT) to directly estimate vault (see in Zhong et al., 2024 ). However, these formulas are still rule-based linear estimations that fail to capture the complex inter-dependence of multiple posterior segment structures, especially that of the anatomical asymmetry between horizontal and vertical sulci, lens curvature, and ciliary body configuration. Furthermore, most of the formulas were derived from small datasets and lack external validation across diverse patient populations. Advances in ultrasound biomicroscopy (UBM) have made it possible to directly visualise and quantify posterior chamber anatomy (He et al., 2012 ; Silverman, 2009 ); including STS (horizontal and vertical), lens curvature (LC) — defined as the minimum perpendicular distance from the STS baseline measured by UBM to the anterior surface of the crystalline lens),and ciliary morphology. Yet, UBM data have not been fully leveraged in lens sizing due to a lack in standardised acquisition protocols, inter-operator variability, and limited integration into predictive models. More importantly, no prior studies have systematically combined large-scale UBM data with machine learning approaches to identify dominant anatomical predictors of vault, model their interactions and develop clinically applicable selection strategies. Given this gap, there is a pressing need to develop personalised data-driven ICL sizing algorithms that move beyond WTW-based assumptions and are grounded in real-world imaging and outcome data. Such approaches would enable refractive surgeons to optimise vault prediction on an individualised basis, thereby improving surgical planning and minimising post-operative surprises. This present study aims to address these limitations by constructing and validating a novel vault prediction model based on UBM-derived parameters and big-data analysis. Specifically, we analysed more than 2,000 cases from a high-volume center: (1) to identify the most influential pre-operative variables for vault via machine learning (random forest), (2) to quantify the effect of sulcus asymmetry and lens curvature on vault variability, (3) to develop a clinically applicable ICL selection chart which incorporates horizontal and vertical STS differences (CS Diff) and lens rise (LC); and (4) to prospectively validate the new chart in a cohort of 568 eyes. Our goal is to build a scalable, anatomically grounded and empirically validated system to optimize ICL sizing and enhance safety and predictability in refractive surgery. Methods Study design and population This was a singlecentre, retrospective cohort study with a prospective, withincentre validation phase. All consecutive Implantable Collamer Lens (ICL, Visian V4c) implantations performed at Liaoning Aier Eye Hospital Affiliated to Northeastern University between January 2023 and January 2025 were screened. The full historical cohort (> 2,000 eyes) was used to develop the prediction model and the individualized sizeselection chart. After the chart was ‘locked’, its realworld performance was prospectively evaluated in an independent series of 286 patients (568 eyes) who underwent ICL surgery between May and June 2025. The protocol was in compliance with the Declaration of Helsinki and received approval from the Medical Ethics Committee of Liaoning Aier Eye Hospital/2023-007-01. Written informed consent was obtained from all participants for the use of their deidentified clinical and imaging data. Patients were eligible if they were 18–45 years old, had stable refraction (change < 0.50 D over the preceding year), presented with myopia between − 3.00 and − 18.00 D spherical equivalent, and had an anterior chamber depth (ACD) of at least 2.8 mm, ensuring adequate vaulting space for posterior chamber phakic lens implantation. Eyes were excluded when UBM imaging was of insufficient quality for reliable landmark identification; when previous ocular surgery or trauma was present; or when coronal UBM revealed atypical posterior chamber anatomy (e.g., markedly short ciliary body or extreme posterior chamber angle configurations) that precluded safe standard placement of the ICL. Only the first operated eye was used for model training in sensitivity analyses to address withinsubject correlation; however, the primary model used all eyes with clusterrobust standard errors. Preoperative parameters and imaging protocol All patients underwent a standardized preoperative workup including slitlamp biomicroscopy, dilated fundus examination, corneal topography/tomography, and anterior segment optical biometry. Whitetowhite (WTW) corneal diameter was obtained from Scheimpflug or Placidobased topography. ACD was measured with optical biometry. Highfrequency ultrasound biomicroscopy (UBM; 50 MHz probe) was performed in both the horizontal and vertical meridians to visualise posterior chamber anatomy. From UBM, we extracted sulcustosulcus distances in the horizontal and vertical planes (STSH and STSV respectively), horizontal and vertical lens curvature rises (LCH and LCV respectively; defined as the perpendicular distance from the STS baseline to the anterior crystalline lens surface), and lens thickness (LT). Posterior chamber angle configuration (CSA) was characterised at eight predefined meridians; additional coronal scans of the ciliary body were used to detect short ciliary bodies or focal abnormalities to guide safe lens orientation. For modelling purposes, we prioritized categorizing WTW, STSH, STSV as firstlevel predictors because of their direct geometric relevance to the resting position of ICL haptics. LCH, LCV, LT were considered secondlevel predictors that modulate vault through lens protrusion and posterior chamber crowding. All UBM operators completed an internal training and standardized programme, and adhered to a written scanning standard operating procedure (SOP) detailing patient positioning, probe orientation, gain, depth and focal settings, and landmark identification. All images were centrally reviewed. Scans that failed the predefined quality criteria (offaxis acquisition, invisible ciliary sulcus, poor contrast) were repeated. Interoperator reliability was periodically audited on a 10% random sample; discordant measurements were resolved by consensus. Feature selection, derived indices and model development The entire > 2,000eye historical dataset was used for model derivation. We first screened all candidate variables for implausible values and remeasured any outliers against source images. Continuous variables were standardised to zero mean and unit variance. A randomforest algorithm (1,000 trees, Gini impurity criterion, max_features = √p, outofbag error for internal performance estimation) was trained to predict continuous post-operative vault height. Tenfold crossvalidation with patientlevel partitioning was used to guard against optimistic bias. Variable importance consistently identified STSH, STSV, LCV and LCH as the dominant predictors. Guided by these results and clinical plausibility, we defined a novel posteriorchamber asymmetry index, CS Diff (ΔH–V), calculated as |STSH – STSV|. Exploratory partialdependence and SHAP (SHapley Additive exPlanations) analyses demonstrated that larger CS Diff values were associated with a drift towards low vaults, whereas nearzero differences were associated with excessive vaulting. Lens curvature was then operationalised through empirically derived thresholds: when LC was < 0.5 mm, lens size followed the base chart; when LC was between 0.5 and 0.7 mm, we recommended upsizing by one model-size with oblique ICL placement; when LC was ≥ 0.7 mm, we recommended upsizing by one model-size with horizontal placement. These decision rules, together with CS Diff, CSA morphology and ciliary body length assessments, were embedded into a clinically applicable ICL selection chart. Probability surfaces of achieving an ideal vault (250–750 µm) were generated across the multidimensional space of STSH, STSV and LC using locally weighted regression (LOESS) smoothing and bootstrapped (5,000 iterations) to derive stable cutpoints. The selection chart was ‘locked’ before prospective validation. Postoperative vault measurement and classification Vault was measured at three months post-operatively using anterior segment OCT. Vault was defined as the central linear distance between the posterior surface of the ICL and the anterior crystalline lens surface. Based on widely accepted clinical thresholds, vaults between 250 and 750 µm were considered ideal; vaults below 250 µm were considered low, and vaults above 750 µm deemed high. Two independent, masked graders performed all measurements. Intergrader agreement was quantified using the intraclass correlation coefficient (ICC, twoway random effects, absolute agreement). When the absolute difference between graders exceeded 30 µm, a third senior grader adjudicated the final value. Statistical analysis Continuous variables are presented as means with standard deviations or medians with interquartile ranges, depending on distribution; categorical variables are summarised as counts and percentages. Normality was assessed using the Shapiro–Wilk test and inspection of Q–Q plots. Intergroup comparisons used Student’s t test or the Mann–Whitney U test for continuous variables and the χ² test or Fisher’s exact test for categorical variables. In the derivation cohort, we quantified model fit to continuous vault by the coefficient of determination (R²) and the root mean squared error (RMSE), both estimated with 10fold crossvalidation. For clinical interpretability, we also framed vault prediction as a binary classification problem (ideal vs nonideal vault) and reported accuracy, sensitivity, specificity, balanced accuracy, area under the receiver operating characteristic curve (AUC), Brier score, calibration slope and intercept. Internal uncertainty was quantified with 1,000iteration nonparametric bootstrapping. External (prospective) validation in the 568eye cohort focused on the primary endpoint of the proportion of eyes achieving an ideal vault using the ‘locked’ ICL selection chart. We compared this proportion against the historical FDA WTW + ACD method using risk differences and 95% confidence intervals derived from generalized estimating equations to account for withinpatient correlation. Misclassification patterns (low vs high vault) were descriptively summarised. We further conducted pre-specified subgroup analyses in patients older than 35 years, in eyes with shallow ACD, and in eyes with high CLR, because these profiles were flagged during development as higher risk for deviation. For these analyses, we fitted logistic regression models including interaction terms between subgroup indicators and the predicted vault probability; marginal effects and their 95% confidence intervals were derived by parametric bootstrapping. A twosided p value < 0.05 was considered statistically significant. All analyses were performed in Python 3.10 (scikitlearn 1.2, shap 0.41) and SPSS 27.0 (IBM Corp.). Finally, to gauge clinical usefulness, we performed decisioncurve analysis comparing the net benefit of the new selection map versus the traditional FDA method across a range of threshold probabilities for accepting an ICL size recommendation. Sensitivity analyses included (i) repeating model fitting with only one eye per patient, (ii) excluding outliers defined as vault values beyond 3 SD from the mean, and (iii) reestimating the model with robust regression to downweight influential points. Results of these analyses were qualitatively consistent with the primary findings and are available in Supplementary Material. Results Cohort Flow, Data Integrity, and Baseline Characteristics Between April 2023 and April 2025, a total of N = 2811 eyes underwent Visian ICL V4c implantation at our center. We excluded n = 632 eyes due to incomplete post-operative vault data and n = 155 eyes due to poor-quality UBM images (e.g., off-axis acquisition, non-visualization of the ciliary sulcus, or poor contrast). This resulted in a final derivation dataset of N = 2024 eyes from 1014 patients for model development. A separate prospective validation was conducted on 568 eyes from 286 patients who underwent ICL surgery between May and June 2025. The two cohorts were broadly comparable in demographics and preoperative biometric parameters, including age, sex, spherical equivalent (SE), anterior chamber depth (ACD), white-to-white (WTW), horizontal and vertical sulcus-to-sulcus (STSH, STSV), horizontal and vertical lens curvature (LCH, LCV), lens thickness (LT). In the validation cohort, the inter-rater intraclass correlation coefficient (ICC) for postoperative vault measurements was ICC = 0.89 (95% CI: 0.81–0.93), indicating excellent agreement. Feature Selection and Effect Interpretation (Derivation Cohort) Using a random forest model trained on over 2,000 eyes, the four most influential predictors of postoperative Vaultwere identified as STSH, STSV, LCV, and LCH (Figure 1). The mean decrease in Gini impurity (standardized to 100%) was: STSH = 14.32%, STSV = 12.13%, LCV = 12.65%, and LCH = 8.93%. The marginal contribution of other variables was minimal once these four were included. SHAP analysis confirmed these findings (Figure 2), highlighting the following relationships: (i) The newly defined CS Diff = |STSH – STSV| was negatively associated with vault (i.e., greater asymmetry predicted lower vault); (ii) LC (lens curvature) was positively associated with vault height, independent of STS measurements (i.e., greater vaulting of the lens predicted higher vault). Partial dependence plots revealed clinically intuitive non-linearities near decision thresholds, which were later incorporated into the nomogram-based sizing chart. Construction of the Sizing Chart and Threshold Determination Using the complete derivation dataset, a multidimensional surface was generated to estimate the probability of achieving an ideal vault (250–750 μm) across the space defined by STSH, STSV, CS Diff, and LC (Figure 4). After 5,000 LOESS-smoothed bootstrap iterations, robust cutoffs for CS Diff and LC were determined. Final operational thresholds were: CS Diff (|AH–V|) ≥ 0.963 ± 0.146 mm → significantly increased risk of low vault; consider upsizing or oblique placement. ≤ 0.242 ± 0.114 mm → associated with high vault risk; compensate by downsizing or repositioning. . LC < 0.5 mm → standard sizing; 0.5–0.7 mm → suggest upsizing by one model-size and oblique placement; ≥ 0.7 mm → suggest upsizing and horizontal placement. These rules were further refined using posterior chamber angle (CSA) morphology and coronal ciliary body assessments to avoid short or aberrant insertion points. For continuous vault prediction, internal cross-validation yielded R = 0.1134 (95% CI: 0.1017–0.1358), RMSE = 301.44 μm , and MAE = 186.19 μm . Calibration was good (slope = 0.93, intercept ≈ 0). Prospective Validation and Clinical Utility At three months post-operatively, the proportion of eyes achieving ideal vault in the validation cohort was 92.96% (528/568; 95% CI: 90.8–95.0%). Non-ideal vaults comprised 7.04% (40/568); including 12 low vaults (30%) and 28 high vaults (70%). Four of these patients (3 low, 1 high) had been pre-labeled as high-risk by the model based on the combination of high LC, shallow ACD, and age > 40 — consistent with patterns identified during model development. Most remaining non-ideal cases fell near decision boundaries, with a median absolute deviation from the ideal range (250–750 μm) of 68 μm (IQR: 53–76). Calibration in the external validation was also excellent (slope = 0.93, intercept = 0.07; Figure 6). In the historical cohort (2023–2024), using the FDA-recommended WTW + ACD rule, only 62.3% achieved ideal vault, and the ICL exchange rate was 2.2%, with 65% attributed to sizing errors. Compared with this baseline, the UBM-guided chart improved the ideal vault rate by an absolute risk difference of 22.19% (95% CI: 15–29%, p < 0.001), estimated using GEE to account for within-patient eye correlation. Decision curve analysis demonstrated greater net clinical benefit across a wide range of threshold probabilities (Supplementary Figure S2), supporting superiority over the FDA heuristic. Subgroup Analyses and Sensitivity Checks The model retained good performance across predefined subgroups but showed decreased effectiveness in previously identified high-risk strata. The ideal vault rate was 73.43% among patients ≥ 40 years, versus 92.07% in younger patients (risk difference = 18.64%, 95% CI: 14.43–20.56). For eyes with shallow anterior chambers (ACD < 2.9 mm), the rate was 83.01% compared to 90.45% in deeper chambers. Eyes with high STSL (≥ 0.5 mm) were more likely to experience high vault if upsizing was not followed. Interactions between subgroup indicators and predicted vault probability were significant for age ( p = 0.0045 ) and STSL ( p = 0.0093 ), but not ACD ( p = 0.12 ), suggesting future iterations should refine age- and curvature-sensitive rules (Supplementary Table S1). Re-estimating the model using one randomly selected eye per patient yielded near-identical thresholds and similar internal performance. Excluding outliers (> mean ± 3 SD) had negligible impact on R, RMSE, or classification metrics. Coefficients from robust regression (Huber loss) matched the original CS Diff and LC thresholds. A “leave-one-operator-out” analysis showed no significant performance drop, supporting the adequacy of UBM SOPs and internal quality control procedures. All sensitivity analyses are detailed in the Supplement. No ICL exchanges or serious vault-related adverse events occurred during the 3-month postoperative follow-up in the validation cohort. Transient, clinically insignificant intraocular pressure elevations occurred in <0.352% of eyes and were managed conservatively. Long-term follow-up is ongoing to assess whether early sizing improvements translate into fewer late exchanges or cataract events. Based on over 2,000 real-world UBM cases, we demonstrate that posterior chamber geometry — particularly sulcus asymmetry (CS Diff) and lens curvature (LC) — are primary drivers of vault variability. The data-driven, UBM-guided sizing chart achieved ~93% ideal vault rate in prospective validation, significantly outperforming the traditional WTW + ACD based assumptions. A small number of predictable high-risk phenotypes (older age, shallow ACD, high LC) indicate promising avenues for further refinement and personalization. Discussion Using more than two thousand realworld ICL cases to discover and operationalise the anatomy that truly governs postoperative vault, we demonstrate that a UBMbased, datadriven strategy centred on posteriorchamber geometry—particularly sulcus asymmetry (quantified by CS Diff, |STSH – STSV|) and lens curvature (LC)—can improve the rate of ideal vaults to 92.96% in a prospective validation cohort. This represents a significant clinical improvement over the historical, FDAendorsed WTW + ACD based assumptions - where we experience a 2.2% exchange rate in our centre, of which 65% were attributable to sizing error. Conceptually, our work aim to close the gap between what most sizing calculators currently measure (anterior segment proxies) and where the ICL actually sits (the ciliary sulcus), and it translates those posterior measurements into an implementable, rulebased selection map that surgeons can use at the point of care. A central contribution of this study is to demonstrate that once STSH, STSV and LC are accounted for, the marginal informational value of WTW and ATA for vault prediction becomes small. This aligns with prior evidence that WTW is an inconsistent surrogate for STS, and that the WTW–STS discrepancy widens as WTW deviates from approximately 11.8 mm (Montés-Micó et al., 2020 ). Our analysis—leveraging random forests, SHAP and partialdependence profiling—further reveals how these posterior parameters should be combined: the newly defined CS Diff captures the degree of posterior chamber anisotropy that systematically biases vault upward or downward; in our study, when STS-H was the same (similar values used with an interval of 0.05), vault showed a negative correlation with ΔHsts–Vsts. For example, using the 12.6 model, post-operative vault decreased as ΔHsts–Vsts increased. In contrast, LC offers an orthogonal control knob that directly shifts vault magnitude. These insights offer an explanatory bridge for the inconsistent vault behaviour seen with rulebased formulas such as NK (ATA + CLR), KS (ATAbased) and ZZ (STS + LT): each of them captures part, but not all, of the vaultdetermining geometry, and none explicitly models sulcus asymmetry or provides clear, empirically grounded orientation rules. Clinically, the implications are evident - posterior chamber imaging should become standard in ICL sizing, at least in centres that perform these procedures at scale or in phenotypes with known higher mis-sizing risk (e.g., older age, shallow ACD, high LC). It should also be noted that the current model has been found less applicable to certain special populations; likely due to age-related ciliary muscle functional decline (Alarfaj et al., 2025 ), zonular laxity, and increases in crystalline lens thickness with concomitant loss of transparency (Thompson et al., 2024 ), as supported by previous studies. We are currently developing dedicated models to address these specific patient groups. Our results also argue for reengineering the preoperative pathway, incorporating ICL size selection chart directly into the clinical workflow and making UBM quality control auditable, with predefined thresholds for acceptable image quality and interoperator agreement. From a clinical and surgeon standpoint, refractive surgeons may strongly consider shying away from “WTW + ACD by default” and favouring a “posteriorchamber–informed” sizing. In the long run, UBM is a valuable pre-operative assessment which brings about fewer post-operative complications, reduced rates of ICL exchanges due to mis-sizing and potentially lower longterm surveillance burdens. Our prospective validation is particularly important. Many sizing papers derive “better” formulas retrospectively but stop short of showing how those rules perform prospectively in real patients when locked and used without further refinement. Here, the high calibration slope and the stability of bootstrapped decision boundaries suggest that the rules are both predictive and transportable within the same centre. Nevertheless, genuine external validation—with different devices, technicians, and patient demographics—is imperative. The model’s reliance on UBM makes technician training and SOP fidelity critical; our ‘leaveonetechnicianout’ analysis mitigates, but does not eliminate, concerns about operator dependence. Future work includes standardise UBM acquisition in different centres, define reporting checklists (e.g., minimum set of STS (measured at eight points on a two-dimensional plane), LC values, CSA morphology, and CP size (CP referring to the ciliary process), and preregister analytic pipelines under TRIPOD (for model reporting) and PROBAST (for bias assessment). The study has limitations that delineate a clear agenda for the next phase of research and implementation. First, the data are singlecentre and all lenses were Visian ICL V4c; extrapolation to other models or designs must be empirically tested. Second, our primary validation time point was three months, which is appropriate for early vault assessment but does not capture medium to longterm vault drift due to agerelated lens growth, anterior lens movement, or capsular changes. Third, although random forests with SHAP improve interpretability relative to deep learning, the cutpoints we derived (e.g., LC 0.5/0.7 mm, CS Diff thresholds) are samplespecific and may require recalibration in populations with different biometry distributions or in centres with different UBM systems. Fourth, we did not conduct a formal costeffectiveness analysis; future implementation science work should quantify the incremental cost per avoided exchange or cataract case, the breakeven point for routine UBM deployment, and the sensitivity of those estimates to technician time and imaging tariffs. Several future directions emerge logically from our findings. First, multi-centre external validation with predefined calibration strategies should be tested for transferability and require definition of when and how often the chart needs local recalibration. Second, phenotypespecific extensions for the clearly identifiable highrisk subgroup (older age, shallow ACD, high LC) should be developed, potentially including interaction terms, hierarchical partial pooling, or Bayesian updating that allows centres to adapt thresholds without sacrificing comparability. Third, to overcome operator dependence, AIassisted UBM segmentation and automated extraction of STS, LC, CP and CSA should be pursued. Integrating these tools into a federated learning framework would allow continuous improvement of the prediction model without exposing sensitive patient data. Finally, our data demonstrates protocols that can be easily implemented. Hospitals can incorporate UBMguided sizing for complex anatomies and require that technicians meet predefined competency metrics (e.g., interrater ICC ≥ 0.90, mean absolute CS Diff measurement error ≤ 0.10 mm across repeats). By establishing a standardized recommendation for posteriorchamber–informed sizing and a minimum UBM SOP, multi-centre collaboration can enhance data validation and broaden the model’s utility. In summary, this study provides robust clinical evidence that a posteriorchamber, UBMbased, machinelearninginformed approach to ICL sizing can significantly improve vault accuracy compared to traditional WTW + ACD methods. Our study supports the translation of this approach into a clinically practical ICL sizing tool. Moving forward, we hope refractive surgeons will transition from the conventional WTW + ACD approach to this more precise, posterior-chamber, UBM-based method. Declarations Ethics approval and consent to participate This study was approved by the Medical Ethics Committee of Liaoning Aier Eye Hospital (Ethics Number: 2023-007-01) and complied with the Declaration of Helsinki.We obtained informed consent from all participants to participate Consent for publication Not applicable Availability of data and materials The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. Competing interests The authors declare that they have no competing interests Funding Aier Eye Hospital Research Fund Authors' contributions Pan Jiayu (PJY) was primarily responsible for writing the entire manuscript. Fang Xuejun (FXJ), as the ICL surgeon, also ensured the integrity and accuracy of the entire model establishment. Wang Yue(WY) and Wang Jifan(WJF) participated in the conception and design of the manuscript. Yu Yidan (YYD) was in charge of data acquisition and analysis. Li Ying(LY) and Li Yingshuai (LYS) formulated the relevant UBM operation standards and standardization protocols. Chu Xiaohan (CXH) conducted UBM examinations and data collection. Shu Yee Seow(SYS) revised the final version. All authors reviewed the manuscript. Acknowledgements First and foremost, I would like to express my deepest gratitude to my supervisor Fang Xuejun, for her invaluable guidance, unwavering support, and insightful comments throughout the entire research process. Her expertise and dedication have been instrumental in shaping this manuscript.I am also highly indebted to all the colleagues and research assistants who have provided assistance and shared their valuable References Alarfaj, G., Helayel, H. B., AlSubaie, M., Hariri, J., Alzaher, F., Khan, O., Al-Jindan, M., AlHabash, A., & Sulaimani, N. M. (2025). Posterior chamber phakic intraocular lens adjustment-causes and complications: a retrospective cohort study. International journal of ophthalmology , 18 (5), 883–888. https://doi.org/10.18240/ijo.2025.05.14 Ang, R.E.T., Reyes, E.K.F., Ayuyao, F.A.J. et al. Comparison of white-to-white measurements using four devices and their determination of ICL sizing. Eye and Vis 9 , 36 (2022). https://doi.org/10.1186/s40662-022-00308-z Cui, W., Wu, X., Ren, Q., Liu, K., Kong, F., & Wu, J. (2023). A new formula based on new parameters for predicting postoperative vault after posterior chamber intraocular lens implantation: a retrospective study. Quantitative imaging in medicine and surgery , 13 (9), 5502–5510. https://doi.org/10.21037/qims-22-1425 Damgaard, I. B., Reffat, M., & Hjortdal, J. (2018). Review of Corneal Biomechanical Properties Following LASIK and SMILE for Myopia and Myopic Astigmatism. The open ophthalmology journal , 12 , 164–174. https://doi.org/10.2174/1874364101812010164 He, M., Wang, D., & Jiang, Y. (2012). Overview of Ultrasound Biomicroscopy. Journal of current glaucoma practice , 6 (1), 25–53. https://doi.org/10.5005/jp-journals-10008-1105 Kane, J. X., Van Heerden, A., Atik, A., & Petsoglou, C. (2016). Intraocular lens Kapoor, G., Sharma, V. K., Singh, P., Sahran, D., Seth, S., Ambiya, V., Kalra, N., & Mishra, A. (2025). Objective and subjective outcomes following implantable collamer lens (ICL) implantation for the correction of myopia. Oman journal of ophthalmology , 18 (2), 187–192. https://doi.org/10.4103/ojo.ojo_171_24 Montés-Micó, R., Tañá-Rivero, P., Aguilar-Córcoles, S. et al. Angle-to-angle and spur-to-spur distance analysis with high-resolution optical coherence tomography. Eye and Vis 7 , 42 (2020). https://doi.org/10.1186/s40662-020-00208-0 Nakamura, T., Nishida, T., Isogai, N., Kojima, T., Sugiyama, Y., & Yoshida, Y. (2023). Evaluation of implantable collamer lens sizing developed by reviewing the horizontal compression-vault coefficient. Journal of cataract and refractive surgery , 49 (5), 525–530. https://doi.org/10.1097/j.jcrs.0000000000001140 Packer M. (2016). Meta-analysis and review: effectiveness, safety, and central port design of the intraocular collamer lens. Clinical ophthalmology (Auckland, N.Z.) , 10 , 1059–1077. https://doi.org/10.2147/OPTH.S111620 Pathak, M., Sahu, V., Kumar, A., Kaur, K., & Gurnani, B. (2024). Current Concepts and Recent Updates of Optical Biometry- A Comprehensive Review. Clinical ophthalmology (Auckland, N.Z.) , 18 , 1191–1206. https://doi.org/10.2147/OPTH.S464538 power formula accuracy: Comparison of 7 formulas. Journal of Cataract & Refractive Silverman R. H. (2009). High-resolution ultrasound imaging of the eye - a review. Clinical & experimental ophthalmology , 37 (1), 54–67. https://doi.org/10.1111/j.1442-9071.2008.01892.x Surgery , 42 (10), 1490–1500. https://doi.org/10.1016/j.jcrs.2016.07.021 Thompson, V., Cummings, A. B., & Wang, X. (2024). Implantable Collamer Lens Procedure Planning: A Review of Global Approaches. Clinical ophthalmology (Auckland, N.Z.) , 18 , 1033–1043. https://doi.org/10.2147/OPTH.S456397 Yang, J., Li, H., Wu, M., He, R., Nong, Y., Zou, Z., Zhang, C., & Zhou, S. (2023). A vault-prediction formula for implantable collamer lens based on preoperative parameters: a retrospective clinical study. BMC ophthalmology , 23 (1), 350. https://doi.org/10.1186/s12886-023-03096-9 Yao, J., Feng, J., Li, W. et al. SMILE and ICL implantation on the ocular surface and meibomian glands in patients with postoperative myopia. BMC Ophthalmol 24 , 522 (2024). https://doi.org/10.1186/s12886-024-03790-2 Zhang, J., He, F., Liu, Y., & Fan, X. (2020). Implantable collamer lens with a central hole for residual refractive error correction after corneal refractive surgery. Experimental and therapeutic medicine , 20 (6), 160. https://doi.org/10.3892/etm.2020.9289 Zhang, Q., Gong, D., Li, K., Dang, K., Wang, Y., Pan, C., Yan, Z., & Yang, W. (2024). From inception to innovation: bibliometric analysis of the evolution, hotspots, and trends in implantable collamer lens surgery research. Frontiers in medicine , 11 , 1432780. https://doi.org/10.3389/fmed.2024.1432780 Zhang, Y., Xi, R., Higashita, R., Okamoto, K., Kamiya, K., Miyata, K., Igarashi, A., Hata, S., Nakamura, T., & Liu, J. (2025). Prior Anatomical Knowledge-guided GAN for ICL surgery postoperative prediction based on AS-OCT image. Medical image analysis , 105 , 103689. Advance online publication. https://doi.org/10.1016/j.media.2025.103689 Zhong, X., Li, Y., Li, Y., Wang, G., Du, Y., & Zhang, M. (2024). Comparison of Predictability in Vault Using NK Formula and KS Formula for the Implantable Collamer Lens Surgery. Journal of ophthalmology , 2024 , 4256371. https://doi.org/10.1155/2024/4256371 Zhu, Q. J., Zhu, W. J., Chen, W. J., Ma, L., & Yuan, Y. (2022). A prediction model for sulcus-to-sulcus diameter in myopic eyes: a 1466-sample retrospective study. BMC ophthalmology , 22 (1), 307. https://doi.org/10.1186/s12886-022-02535-3 Supplementary Table Supplementary Table S1 is not available with this version. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7400290","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":507745141,"identity":"fe6d74a5-347e-4ffe-8538-6cc6089c580a","order_by":0,"name":"Jiayu Pan","email":"","orcid":"","institution":"Liaoning Aier Eye Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jiayu","middleName":"","lastName":"Pan","suffix":""},{"id":507745144,"identity":"e2cb969b-9612-4ce2-9722-600979002c30","order_by":1,"name":"Yue Wang","email":"","orcid":"","institution":"Liaoning Aier Eye Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yue","middleName":"","lastName":"Wang","suffix":""},{"id":507745148,"identity":"e165eaa3-d0e7-4ca3-b7e1-981cceb5cc95","order_by":2,"name":"Jifan Wang","email":"","orcid":"","institution":"Liaoning Aier Eye Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jifan","middleName":"","lastName":"Wang","suffix":""},{"id":507745149,"identity":"df961d74-4fd0-4dcf-b7be-f79904335780","order_by":3,"name":"Ying Li","email":"","orcid":"","institution":"Liaoning Aier Eye Hospital","correspondingAuthor":false,"prefix":"","firstName":"Ying","middleName":"","lastName":"Li","suffix":""},{"id":507745151,"identity":"97ffba13-307f-4b2c-be3c-d0187047fa09","order_by":4,"name":"Xiaohan Chu","email":"","orcid":"","institution":"Liaoning Aier Eye Hospital","correspondingAuthor":false,"prefix":"","firstName":"Xiaohan","middleName":"","lastName":"Chu","suffix":""},{"id":507745153,"identity":"dffd60bc-df21-4899-9b1a-ad344f2c7a9d","order_by":5,"name":"Yingshuai Li","email":"","orcid":"","institution":"Liaoning Aier Eye Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yingshuai","middleName":"","lastName":"Li","suffix":""},{"id":507745155,"identity":"25004401-e51f-4529-b504-ded47905ddf7","order_by":6,"name":"Yidan Yu","email":"","orcid":"","institution":"Liaoning Aier Eye Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yidan","middleName":"","lastName":"Yu","suffix":""},{"id":507745156,"identity":"d8bf641b-5acd-498c-94c9-bec2af4f957b","order_by":7,"name":"Shu Yee Seow","email":"","orcid":"","institution":"International Specialist Eye Centre (ISEC)","correspondingAuthor":false,"prefix":"","firstName":"Shu","middleName":"Yee","lastName":"Seow","suffix":""},{"id":507745159,"identity":"cc5ffb6e-5d40-4016-8495-06fe6ac30838","order_by":8,"name":"Xuejun Fang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABBElEQVRIiWNgGAWjYBAC9gYIzQxEBxgSgCwDZgJaeA7AtbAlkKYFxDQAUwaEHMbDfvbwa942G3aD2z0fPzyouCdvzs578ANDjU00Ti08eWmWM9vSmA3unN0skXCm2HBnM1+yBMOxtNwGHFrsGXLMDD62HWY2uJG7QSKxLSHB4DCPgQRjw2GcWnj435gZJLb9B2rJefwj8R9Yi/EPvFokcowffGw7ANLCJpHYANZiht8WiTdmjDPOJTNL3kgzs0g4lmC4AagFyMDtFx7+HOPPPGV2yXw3kh/f/FGTIG9w/ozxjQ81Nji1AAGbBJBIRhVLwK0cBJg/AAk7/GpGwSgYBaNgRAMAFE1YsTBSksgAAAAASUVORK5CYII=","orcid":"","institution":"Aier Academy of Ophthalmology, Central South University","correspondingAuthor":true,"prefix":"","firstName":"Xuejun","middleName":"","lastName":"Fang","suffix":""}],"badges":[],"createdAt":"2025-08-18 13:38:28","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7400290/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7400290/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":90383511,"identity":"dd3901bb-014e-4afd-b78e-5bc0900897b5","added_by":"auto","created_at":"2025-09-02 07:00:26","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":38560,"visible":true,"origin":"","legend":"\u003cp\u003eFeature Importance of Vault Predictors in Random Forest Model\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eNote.\u003c/em\u003e Mean decrease in Gini impurity (scaled to 100%) for all preoperative anatomical variables used to predict postoperative vault. STSH (horizontal sulcus-to-sulcus distance), STSV (vertical sulcus-to-sulcus), LCH (horizontal lens curvature), and LCV (vertical lens curvature) emerged as the top four predictors. All remaining features, including WTW,ACD and ACA(anterior chamber angle), had insignificant contribution after these were accounted for\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7400290/v1/6ff80f926a2f4ccb6387d0c9.png"},{"id":90382283,"identity":"a31db9c4-442c-4d24-95fa-dd8c43188910","added_by":"auto","created_at":"2025-09-02 06:52:26","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":275878,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP Analysis of Vault Height Predictors\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eNote.\u003c/em\u003e SHAP (SHapley Additive exPlanations) values illustrating the marginal contribution of each variable to the predicted vault. CS Diff (|STSH – STSV|) shows a negative association with vault, reflecting the effect of posterior chamber asymmetry. LC (lens curvature) exhibits a positive association, indicating that greater lens protrusion leads to higher vaults. STSH and STSV are displayed to show their relative influence and directional effect.\u003c/p\u003e","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7400290/v1/65434bf96962759db61ccbd6.jpeg"},{"id":90382281,"identity":"97c721af-1227-499b-ba73-a85ec86ba476","added_by":"auto","created_at":"2025-09-02 06:52:26","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":391790,"visible":true,"origin":"","legend":"\u003cp\u003eProbability Surface of Achieving Ideal Vault by STSH, STSV, and LC\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eNote\u003c/em\u003e. Three-dimensional LOESS-generated probability surface mapping the likelihood of achieving an ideal vault (250–750 μm) based on STSH, STSV, and LC. Higher symmetry (low CS Diff) and moderate LC values are associated with the highest probability of ideal vaults. The surface is bootstrapped over 5,000 iterations to ensure stability of thresholds and cut points.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7400290/v1/af8d5ede9d068d17618442ef.png"},{"id":90383512,"identity":"339d5228-b79e-487c-8525-355110b7349e","added_by":"auto","created_at":"2025-09-02 07:00:26","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":115834,"visible":true,"origin":"","legend":"\u003cp\u003eDecision Curve Analysis Comparing UBM-Guided and WTW+ACD Methods\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eNote.\u003c/em\u003e This decision curve analysis illustrates the net clinical benefit of four strategies—UBM-guided sizing, WTW+ACD sizing, treat-all, and treat-none—across a range of threshold probabilities. The UBM-guided approach maintains the highest net benefit over most clinically relevant thresholds, indicating superior decision-making performance compared to WTW+ACD. Treat-all shows declining net benefit at higher thresholds, while treat-none remains constant at zero net benefit, serving as a reference line.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7400290/v1/72c2714ad92c9e022d54e88c.png"},{"id":90382293,"identity":"a7ecc234-2b6a-4ea1-bfbf-2193c6a16c19","added_by":"auto","created_at":"2025-09-02 06:52:26","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":485355,"visible":true,"origin":"","legend":"\u003cp\u003eProbability Surface of Achieving Ideal Vault Based on STS Measurements\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eNote.\u003c/em\u003e This 3D probability surface illustrates the predicted likelihood of achieving an ideal postoperative vault as a function of horizontal sulcus-to-sulcus (STSH) and vertical sulcus-to-sulcus (STSV) distances. The peak probability region, approaching 100%, is located near the optimal combination of STSH and STSV, with likelihoods decreasing symmetrically as measurements deviate from this range. The color gradient represents probability levels, highlighting the narrow parameter window associated with the highest surgical success rates.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7400290/v1/106fa8ff75dcb5114d001a21.png"},{"id":91332452,"identity":"4544270e-bf8b-4e12-ad08-716f3aa8fa9b","added_by":"auto","created_at":"2025-09-15 11:11:55","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1775888,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7400290/v1/949cf58e-d53f-48e7-9955-13027889dff2.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A UBM-Based Big Data Model for Individualized ICL Size Selection in Myopic Eyes","fulltext":[{"header":"Background","content":"\u003cp\u003eThe Implantable Collamer Lens (ICL) has revolutionised the field of refractive surgery, providing an effective and reversible option for the correction of moderate to high myopia (Kapoor et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Zhang et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Compared with corneal refractive procedures such as LASIK and SMILE, ICL implantation preserves corneal biomechanics by avoiding stromal tissue removal, and offers superior visual quality in selected patients (Damgaard et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). As surgical techniques and lens materials continue to evolve, ICL implantation is increasingly being offered to a broader range of patients, including those with thin corneas or high degrees of ametropia (Yao et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Zhang et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). However, the safety and long-term success of ICL surgery critically depend on the post-operative vault, i.e. the distance between the posterior surface of the implanted ICL and the anterior surface of the crystalline lens (Cui et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Zhang et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). An ideal vault ensures sufficient space to prevent contact with the natural lens (which could induce anterior subcapsular cataract), while avoiding excessive forward displacement that could increase intraocular pressure or precipitate angle-closure glaucoma (Yang et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Thus, accurate pre-operative prediction of vault is essential to avoid post-operative complications, re-operations, or ICL explanation.\u003c/p\u003e\u003cp\u003eDespite its clinical importance, vault prediction remains an unsolved challenge in ICL surgery. The current FDA-approved sizing approach relies on white-to-white (WTW) corneal diameter and anterior chamber depth (ACD) \u0026mdash; parameters easily obtained via optical biometry (Ang et al., \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Pathak et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). However, these anterior segment measurements are only indirect proxies for the true sulcus-to-sulcus (STS) diameter, where the ICL haptics are actually positioned (Zhu et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Numerous studies have shown that WTW does not reliably correlate with STS, and that errors in this approximation can lead to significant variability in vault (see in Packer et al., 2016). In our practice, we observed a 2.2% ICL explantation rate during 2023 and 2024, which is consistent with the reported rates of approximately 1\u0026ndash;4% in literatures across different centers. Over 65% of our cases were attributable to inaccurate sizing based on the traditional WTW\u0026thinsp;+\u0026thinsp;ACD method (Kane et al., \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2016\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eTo address these limitations, several alternative formulas have been developed. The NK formula, which combines angle-to-angle (ATA) distance and crystalline lens rise (CLR), has been reported to achieve an accuracy of approximately 80\u0026ndash;88% in predicting ideal vault (Nakamura et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). On the other hand, the KS formula focuses on ATA alone; while the ZZ method attempts to incorporate STS and lens thickness (LT) to directly estimate vault (see in Zhong et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). However, these formulas are still rule-based linear estimations that fail to capture the complex inter-dependence of multiple posterior segment structures, especially that of the anatomical asymmetry between horizontal and vertical sulci, lens curvature, and ciliary body configuration. Furthermore, most of the formulas were derived from small datasets and lack external validation across diverse patient populations.\u003c/p\u003e\u003cp\u003eAdvances in ultrasound biomicroscopy (UBM) have made it possible to directly visualise and quantify posterior chamber anatomy (He et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Silverman, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2009\u003c/span\u003e); including STS (horizontal and vertical), lens curvature (LC) \u0026mdash; defined as the minimum perpendicular distance from the STS baseline measured by UBM to the anterior surface of the crystalline lens),and ciliary morphology. Yet, UBM data have not been fully leveraged in lens sizing due to a lack in standardised acquisition protocols, inter-operator variability, and limited integration into predictive models. More importantly, no prior studies have systematically combined large-scale UBM data with machine learning approaches to identify dominant anatomical predictors of vault, model their interactions and develop clinically applicable selection strategies.\u003c/p\u003e\u003cp\u003eGiven this gap, there is a pressing need to develop personalised data-driven ICL sizing algorithms that move beyond WTW-based assumptions and are grounded in real-world imaging and outcome data. Such approaches would enable refractive surgeons to optimise vault prediction on an individualised basis, thereby improving surgical planning and minimising post-operative surprises. This present study aims to address these limitations by constructing and validating a novel vault prediction model based on UBM-derived parameters and big-data analysis. Specifically, we analysed more than 2,000 cases from a high-volume center: (1) to identify the most influential pre-operative variables for vault via machine learning (random forest), (2) to quantify the effect of sulcus asymmetry and lens curvature on vault variability, (3) to develop a clinically applicable ICL selection chart which incorporates horizontal and vertical STS differences (CS Diff) and lens rise (LC); and (4) to prospectively validate the new chart in a cohort of 568 eyes. Our goal is to build a scalable, anatomically grounded and empirically validated system to optimize ICL sizing and enhance safety and predictability in refractive surgery.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eStudy design and population\u003c/h2\u003e\u003cp\u003eThis was a singlecentre, retrospective cohort study with a prospective, withincentre validation phase. All consecutive Implantable Collamer Lens (ICL, Visian V4c) implantations performed at Liaoning Aier Eye Hospital Affiliated to Northeastern University between January 2023 and January 2025 were screened. The full historical cohort (\u0026gt;\u0026thinsp;2,000 eyes) was used to develop the prediction model and the individualized sizeselection chart. After the chart was \u0026lsquo;locked\u0026rsquo;, its realworld performance was prospectively evaluated in an independent series of 286 patients (568 eyes) who underwent ICL surgery between May and June 2025. The protocol was in compliance with the Declaration of Helsinki and received approval from the Medical Ethics Committee of Liaoning Aier Eye Hospital/2023-007-01. Written informed consent was obtained from all participants for the use of their deidentified clinical and imaging data.\u003c/p\u003e\u003cp\u003ePatients were eligible if they were 18\u0026ndash;45 years old, had stable refraction (change\u0026thinsp;\u0026lt;\u0026thinsp;0.50 D over the preceding year), presented with myopia between \u0026minus;\u0026thinsp;3.00 and \u0026minus;\u0026thinsp;18.00 D spherical equivalent, and had an anterior chamber depth (ACD) of at least 2.8 mm, ensuring adequate vaulting space for posterior chamber phakic lens implantation. Eyes were excluded when UBM imaging was of insufficient quality for reliable landmark identification; when previous ocular surgery or trauma was present; or when coronal UBM revealed atypical posterior chamber anatomy (e.g., markedly short ciliary body or extreme posterior chamber angle configurations) that precluded safe standard placement of the ICL. Only the first operated eye was used for model training in sensitivity analyses to address withinsubject correlation; however, the primary model used all eyes with clusterrobust standard errors.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003ePreoperative parameters and imaging protocol\u003c/h3\u003e\n\u003cp\u003eAll patients underwent a standardized preoperative workup including slitlamp biomicroscopy, dilated fundus examination, corneal topography/tomography, and anterior segment optical biometry. Whitetowhite (WTW) corneal diameter was obtained from Scheimpflug or Placidobased topography. ACD was measured with optical biometry. Highfrequency ultrasound biomicroscopy (UBM; 50 MHz probe) was performed in both the horizontal and vertical meridians to visualise posterior chamber anatomy. From UBM, we extracted sulcustosulcus distances in the horizontal and vertical planes (STSH and STSV respectively), horizontal and vertical lens curvature rises (LCH and LCV respectively; defined as the perpendicular distance from the STS baseline to the anterior crystalline lens surface), and lens thickness (LT). Posterior chamber angle configuration (CSA) was characterised at eight predefined meridians; additional coronal scans of the ciliary body were used to detect short ciliary bodies or focal abnormalities to guide safe lens orientation.\u003c/p\u003e\u003cp\u003eFor modelling purposes, we prioritized categorizing WTW, STSH, STSV as firstlevel predictors because of their direct geometric relevance to the resting position of ICL haptics. LCH, LCV, LT were considered secondlevel predictors that modulate vault through lens protrusion and posterior chamber crowding. All UBM operators completed an internal training and standardized programme, and adhered to a written scanning standard operating procedure (SOP) detailing patient positioning, probe orientation, gain, depth and focal settings, and landmark identification. All images were centrally reviewed. Scans that failed the predefined quality criteria (offaxis acquisition, invisible ciliary sulcus, poor contrast) were repeated. Interoperator reliability was periodically audited on a 10% random sample; discordant measurements were resolved by consensus.\u003c/p\u003e\n\u003ch3\u003eFeature selection, derived indices and model development\u003c/h3\u003e\n\u003cp\u003eThe entire\u0026thinsp;\u0026gt;\u0026thinsp;2,000eye historical dataset was used for model derivation. We first screened all candidate variables for implausible values and remeasured any outliers against source images. Continuous variables were standardised to zero mean and unit variance. A randomforest algorithm (1,000 trees, Gini impurity criterion, max_features = \u0026radic;p, outofbag error for internal performance estimation) was trained to predict continuous post-operative vault height. Tenfold crossvalidation with patientlevel partitioning was used to guard against optimistic bias. Variable importance consistently identified STSH, STSV, LCV and LCH as the dominant predictors.\u003c/p\u003e\u003cp\u003eGuided by these results and clinical plausibility, we defined a novel posteriorchamber asymmetry index, CS Diff (ΔH\u0026ndash;V), calculated as |STSH \u0026ndash; STSV|. Exploratory partialdependence and SHAP (SHapley Additive exPlanations) analyses demonstrated that larger CS Diff values were associated with a drift towards low vaults, whereas nearzero differences were associated with excessive vaulting. Lens curvature was then operationalised through empirically derived thresholds: when LC was \u0026lt;\u0026thinsp;0.5 mm, lens size followed the base chart; when LC was between 0.5 and 0.7 mm, we recommended upsizing by one model-size with oblique ICL placement; when LC was \u0026ge;\u0026thinsp;0.7 mm, we recommended upsizing by one model-size with horizontal placement. These decision rules, together with CS Diff, CSA morphology and ciliary body length assessments, were embedded into a clinically applicable ICL selection chart. Probability surfaces of achieving an ideal vault (250\u0026ndash;750 \u0026micro;m) were generated across the multidimensional space of STSH, STSV and LC using locally weighted regression (LOESS) smoothing and bootstrapped (5,000 iterations) to derive stable cutpoints. The selection chart was \u0026lsquo;locked\u0026rsquo; before prospective validation.\u003c/p\u003e\n\u003ch3\u003ePostoperative vault measurement and classification\u003c/h3\u003e\n\u003cp\u003eVault was measured at three months post-operatively using anterior segment OCT. Vault was defined as the central linear distance between the posterior surface of the ICL and the anterior crystalline lens surface. Based on widely accepted clinical thresholds, vaults between 250 and 750 \u0026micro;m were considered ideal; vaults below 250 \u0026micro;m were considered low, and vaults above 750 \u0026micro;m deemed high. Two independent, masked graders performed all measurements. Intergrader agreement was quantified using the intraclass correlation coefficient (ICC, twoway random effects, absolute agreement). When the absolute difference between graders exceeded 30 \u0026micro;m, a third senior grader adjudicated the final value.\u003c/p\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003eStatistical analysis\u003c/h2\u003e\u003cp\u003eContinuous variables are presented as means with standard deviations or medians with interquartile ranges, depending on distribution; categorical variables are summarised as counts and percentages. Normality was assessed using the Shapiro\u0026ndash;Wilk test and inspection of Q\u0026ndash;Q plots. Intergroup comparisons used Student\u0026rsquo;s t test or the Mann\u0026ndash;Whitney U test for continuous variables and the χ\u0026sup2; test or Fisher\u0026rsquo;s exact test for categorical variables. In the derivation cohort, we quantified model fit to continuous vault by the coefficient of determination (R\u0026sup2;) and the root mean squared error (RMSE), both estimated with 10fold crossvalidation. For clinical interpretability, we also framed vault prediction as a binary classification problem (ideal vs nonideal vault) and reported accuracy, sensitivity, specificity, balanced accuracy, area under the receiver operating characteristic curve (AUC), Brier score, calibration slope and intercept. Internal uncertainty was quantified with 1,000iteration nonparametric bootstrapping.\u003c/p\u003e\u003cp\u003eExternal (prospective) validation in the 568eye cohort focused on the primary endpoint of the proportion of eyes achieving an ideal vault using the \u0026lsquo;locked\u0026rsquo; ICL selection chart. We compared this proportion against the historical FDA WTW\u0026thinsp;+\u0026thinsp;ACD method using risk differences and 95% confidence intervals derived from generalized estimating equations to account for withinpatient correlation. Misclassification patterns (low vs high vault) were descriptively summarised. We further conducted pre-specified subgroup analyses in patients older than 35 years, in eyes with shallow ACD, and in eyes with high CLR, because these profiles were flagged during development as higher risk for deviation. For these analyses, we fitted logistic regression models including interaction terms between subgroup indicators and the predicted vault probability; marginal effects and their 95% confidence intervals were derived by parametric bootstrapping. A twosided p value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was considered statistically significant. All analyses were performed in Python 3.10 (scikitlearn 1.2, shap 0.41) and SPSS 27.0 (IBM Corp.).\u003c/p\u003e\u003cp\u003eFinally, to gauge clinical usefulness, we performed decisioncurve analysis comparing the net benefit of the new selection map versus the traditional FDA method across a range of threshold probabilities for accepting an ICL size recommendation. Sensitivity analyses included (i) repeating model fitting with only one eye per patient, (ii) excluding outliers defined as vault values beyond 3 SD from the mean, and (iii) reestimating the model with robust regression to downweight influential points. Results of these analyses were qualitatively consistent with the primary findings and are available in Supplementary Material.\u003c/p\u003e\u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003eCohort Flow, Data Integrity, and Baseline Characteristics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eBetween April 2023 and April 2025, a total of \u003cem\u003eN = 2811\u003c/em\u003e eyes underwent Visian ICL V4c implantation at our center. We excluded \u003cem\u003en = 632\u003c/em\u003e eyes due to incomplete post-operative vault data and \u003cem\u003en = 155\u003c/em\u003e eyes due to poor-quality UBM images (e.g., off-axis acquisition, non-visualization of the ciliary sulcus, or poor contrast). This resulted in a final derivation dataset of \u003cem\u003eN = 2024\u003c/em\u003e eyes from \u003cem\u003e1014\u003c/em\u003e patients for model development. A separate prospective validation was conducted on 568 eyes from 286 patients who underwent ICL surgery between May and June 2025.\u003c/p\u003e\n\u003cp\u003eThe two cohorts were broadly comparable in demographics and preoperative biometric parameters, including age, sex, spherical equivalent (SE), anterior chamber depth (ACD), white-to-white (WTW), horizontal and vertical sulcus-to-sulcus (STSH, STSV), horizontal and vertical lens curvature (LCH, LCV), lens thickness (LT). In the validation cohort, the inter-rater intraclass correlation coefficient (ICC) for postoperative vault measurements was \u003cem\u003eICC = 0.89\u003c/em\u003e (95% CI: 0.81–0.93), indicating excellent agreement.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFeature Selection and Effect Interpretation (Derivation Cohort)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eUsing a random forest model trained on over 2,000 eyes, the four most influential predictors of postoperative Vaultwere identified as STSH, STSV, LCV, and LCH (Figure 1). The mean decrease in Gini impurity (standardized to 100%) was: STSH = 14.32%, STSV = 12.13%, LCV = 12.65%, and LCH = 8.93%. The marginal contribution of other variables was minimal once these four were included.\u003c/p\u003e\n\u003cp\u003eSHAP analysis confirmed these findings (Figure 2), highlighting the following relationships:\u003c/p\u003e\n\u003cul type=\"disc\"\u003e\n \u003cli\u003e(i) The newly defined \u003cem\u003eCS Diff = |STSH – STSV|\u003c/em\u003e was negatively associated with vault (i.e., greater asymmetry predicted lower vault);\u003c/li\u003e\n \u003cli\u003e(ii) LC (lens curvature) was positively associated with vault height, independent of STS measurements (i.e., greater vaulting of the lens predicted higher vault).\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003ePartial dependence plots revealed clinically intuitive non-linearities near decision thresholds, which were later incorporated into the nomogram-based sizing chart.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConstruction of the Sizing Chart and Threshold Determination\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eUsing the complete derivation dataset, a multidimensional surface was generated to estimate the probability of achieving an ideal vault (250–750 μm) across the space defined by STSH, STSV, CS Diff, and LC (Figure 4). After 5,000 LOESS-smoothed bootstrap iterations, robust cutoffs for CS Diff and LC were determined. Final operational thresholds were:\u003c/p\u003e\n\u003cul type=\"disc\"\u003e\n \u003cli\u003e\u003cem\u003eCS Diff (|AH–V|)\u003c/em\u003e\u003cbr\u003e\u0026nbsp;≥ 0.963 ± 0.146 mm → significantly increased risk of low vault; consider upsizing or oblique placement.\u003cbr\u003e\u0026nbsp;≤ 0.242 ± 0.114 mm → associated with high vault risk; compensate by downsizing or repositioning. .\u003c/li\u003e\n \u003cli\u003e\u003cem\u003eLC\u003c/em\u003e\u003cbr\u003e\u0026nbsp;\u0026lt; 0.5 mm → standard sizing;\u003cbr\u003e\u0026nbsp;0.5–0.7 mm → suggest upsizing by one model-size and oblique placement;\u003cbr\u003e\u0026nbsp;≥ 0.7 mm → suggest upsizing and horizontal placement.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThese rules were further refined using posterior chamber angle (CSA) morphology and coronal ciliary body assessments to avoid short or aberrant insertion points. For continuous vault prediction, internal cross-validation yielded \u003cem\u003eR = 0.1134\u003c/em\u003e (95% CI: 0.1017–0.1358), \u003cem\u003eRMSE = 301.44 μm\u003c/em\u003e, and \u003cem\u003eMAE = 186.19 μm\u003c/em\u003e. Calibration was good (slope = 0.93, intercept ≈ 0).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eProspective Validation and Clinical Utility\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAt three months post-operatively, the proportion of eyes achieving ideal vault in the validation cohort was \u003cem\u003e92.96%\u003c/em\u003e (528/568; 95% CI: 90.8–95.0%). Non-ideal vaults comprised 7.04% (40/568); including 12 low vaults (30%) and 28 high vaults (70%). Four of these patients (3 low, 1 high) had been pre-labeled as high-risk by the model based on the combination of high LC, shallow ACD, and age \u0026gt; 40 — consistent with patterns identified during model development.\u003c/p\u003e\n\u003cp\u003eMost remaining non-ideal cases fell near decision boundaries, with a median absolute deviation from the ideal range (250–750 μm) of 68 μm (IQR: 53–76). Calibration in the external validation was also excellent (slope = 0.93, intercept = 0.07; Figure 6).\u003c/p\u003e\n\u003cp\u003eIn the historical cohort (2023–2024), using the FDA-recommended WTW + ACD rule, only \u003cem\u003e62.3%\u003c/em\u003e achieved ideal vault, and the ICL exchange rate was 2.2%, with 65% attributed to sizing errors. Compared with this baseline, the UBM-guided chart improved the ideal vault rate by an absolute risk difference of \u003cem\u003e22.19%\u003c/em\u003e (95% CI: 15–29%, \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001), estimated using GEE to account for within-patient eye correlation. Decision curve analysis demonstrated greater net clinical benefit across a wide range of threshold probabilities (Supplementary Figure S2), supporting superiority over the FDA heuristic.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSubgroup Analyses and Sensitivity Checks\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe model retained good performance across predefined subgroups but showed decreased effectiveness in previously identified high-risk strata. The ideal vault rate was \u003cem\u003e73.43%\u003c/em\u003e among patients ≥ 40 years, versus \u003cem\u003e92.07%\u003c/em\u003e in younger patients (risk difference = 18.64%, 95% CI: 14.43–20.56). For eyes with shallow anterior chambers (ACD \u0026lt; 2.9 mm), the rate was \u003cem\u003e83.01%\u003c/em\u003e compared to \u003cem\u003e90.45%\u003c/em\u003e in deeper chambers. Eyes with high STSL (≥ 0.5 mm) were more likely to experience high vault if upsizing was not followed. Interactions between subgroup indicators and predicted vault probability were significant for age (\u003cem\u003ep = 0.0045\u003c/em\u003e) and STSL (\u003cem\u003ep = 0.0093\u003c/em\u003e), but not ACD (\u003cem\u003ep = 0.12\u003c/em\u003e), suggesting future iterations should refine age- and curvature-sensitive rules (Supplementary Table S1).\u003c/p\u003e\n\u003cp\u003eRe-estimating the model using one randomly selected eye per patient yielded near-identical thresholds and similar internal performance. Excluding outliers (\u0026gt; mean ± 3 SD) had negligible impact on R, RMSE, or classification metrics. Coefficients from robust regression (Huber loss) matched the original CS Diff and LC thresholds. A “leave-one-operator-out” analysis showed no significant performance drop, supporting the adequacy of UBM SOPs and internal quality control procedures. All sensitivity analyses are detailed in the Supplement.\u003c/p\u003e\n\u003cp\u003eNo ICL exchanges or serious vault-related adverse events occurred during the 3-month postoperative follow-up in the validation cohort. Transient, clinically insignificant intraocular pressure elevations occurred in \u0026lt;0.352% of eyes and were managed conservatively. Long-term follow-up is ongoing to assess whether early sizing improvements translate into fewer late exchanges or cataract events.\u003c/p\u003e\n\u003cp\u003eBased on over 2,000 real-world UBM cases, we demonstrate that posterior chamber geometry — particularly sulcus asymmetry (CS Diff) and lens curvature (LC) — are primary drivers of vault variability. The data-driven, UBM-guided sizing chart achieved ~93% ideal vault rate in prospective validation, significantly outperforming the traditional WTW + ACD based assumptions. A small number of predictable high-risk phenotypes (older age, shallow ACD, high LC) indicate promising avenues for further refinement and personalization.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eUsing more than two thousand realworld ICL cases to discover and operationalise the anatomy that truly governs postoperative vault, we demonstrate that a UBMbased, datadriven strategy centred on posteriorchamber geometry\u0026mdash;particularly sulcus asymmetry (quantified by CS Diff, |STSH \u0026ndash; STSV|) and lens curvature (LC)\u0026mdash;can improve the rate of ideal vaults to 92.96% in a prospective validation cohort. This represents a significant clinical improvement over the historical, FDAendorsed WTW\u0026thinsp;+\u0026thinsp;ACD based assumptions - where we experience a 2.2% exchange rate in our centre, of which 65% were attributable to sizing error. Conceptually, our work aim to close the gap between what most sizing calculators currently measure (anterior segment proxies) and where the ICL actually sits (the ciliary sulcus), and it translates those posterior measurements into an implementable, rulebased selection map that surgeons can use at the point of care.\u003c/p\u003e\u003cp\u003eA central contribution of this study is to demonstrate that once STSH, STSV and LC are accounted for, the marginal informational value of WTW and ATA for vault prediction becomes small. This aligns with prior evidence that WTW is an inconsistent surrogate for STS, and that the WTW\u0026ndash;STS discrepancy widens as WTW deviates from approximately 11.8 mm (Mont\u0026eacute;s-Mic\u0026oacute; et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Our analysis\u0026mdash;leveraging random forests, SHAP and partialdependence profiling\u0026mdash;further reveals how these posterior parameters should be combined: the newly defined CS Diff captures the degree of posterior chamber anisotropy that systematically biases vault upward or downward; in our study, when STS-H was the same (similar values used with an interval of 0.05), vault showed a negative correlation with ΔHsts\u0026ndash;Vsts. For example, using the 12.6 model, post-operative vault decreased as ΔHsts\u0026ndash;Vsts increased. In contrast, LC offers an orthogonal control knob that directly shifts vault magnitude. These insights offer an explanatory bridge for the inconsistent vault behaviour seen with rulebased formulas such as NK (ATA\u0026thinsp;+\u0026thinsp;CLR), KS (ATAbased) and ZZ (STS\u0026thinsp;+\u0026thinsp;LT): each of them captures part, but not all, of the vaultdetermining geometry, and none explicitly models sulcus asymmetry or provides clear, empirically grounded orientation rules.\u003c/p\u003e\u003cp\u003eClinically, the implications are evident - posterior chamber imaging should become standard in ICL sizing, at least in centres that perform these procedures at scale or in phenotypes with known higher mis-sizing risk (e.g., older age, shallow ACD, high LC). It should also be noted that the current model has been found less applicable to certain special populations; likely due to age-related ciliary muscle functional decline (Alarfaj et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), zonular laxity, and increases in crystalline lens thickness with concomitant loss of transparency (Thompson et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), as supported by previous studies. We are currently developing dedicated models to address these specific patient groups.\u003c/p\u003e\u003cp\u003eOur results also argue for reengineering the preoperative pathway, incorporating ICL size selection chart directly into the clinical workflow and making UBM quality control auditable, with predefined thresholds for acceptable image quality and interoperator agreement. From a clinical and surgeon standpoint, refractive surgeons may strongly consider shying away from \u0026ldquo;WTW\u0026thinsp;+\u0026thinsp;ACD by default\u0026rdquo; and favouring a \u0026ldquo;posteriorchamber\u0026ndash;informed\u0026rdquo; sizing. In the long run, UBM is a valuable pre-operative assessment which brings about fewer post-operative complications, reduced rates of ICL exchanges due to mis-sizing and potentially lower longterm surveillance burdens.\u003c/p\u003e\u003cp\u003eOur prospective validation is particularly important. Many sizing papers derive \u0026ldquo;better\u0026rdquo; formulas retrospectively but stop short of showing how those rules perform prospectively in real patients when locked and used without further refinement. Here, the high calibration slope and the stability of bootstrapped decision boundaries suggest that the rules are both predictive and transportable within the same centre. Nevertheless, genuine external validation\u0026mdash;with different devices, technicians, and patient demographics\u0026mdash;is imperative. The model\u0026rsquo;s reliance on UBM makes technician training and SOP fidelity critical; our \u0026lsquo;leaveonetechnicianout\u0026rsquo; analysis mitigates, but does not eliminate, concerns about operator dependence. Future work includes standardise UBM acquisition in different centres, define reporting checklists (e.g., minimum set of STS (measured at eight points on a two-dimensional plane), LC values, CSA morphology, and CP size (CP referring to the ciliary process), and preregister analytic pipelines under TRIPOD (for model reporting) and PROBAST (for bias assessment).\u003c/p\u003e\u003cp\u003eThe study has limitations that delineate a clear agenda for the next phase of research and implementation. First, the data are singlecentre and all lenses were Visian ICL V4c; extrapolation to other models or designs must be empirically tested. Second, our primary validation time point was three months, which is appropriate for early vault assessment but does not capture medium to longterm vault drift due to agerelated lens growth, anterior lens movement, or capsular changes. Third, although random forests with SHAP improve interpretability relative to deep learning, the cutpoints we derived (e.g., LC 0.5/0.7 mm, CS Diff thresholds) are samplespecific and may require recalibration in populations with different biometry distributions or in centres with different UBM systems. Fourth, we did not conduct a formal costeffectiveness analysis; future implementation science work should quantify the incremental cost per avoided exchange or cataract case, the breakeven point for routine UBM deployment, and the sensitivity of those estimates to technician time and imaging tariffs.\u003c/p\u003e\u003cp\u003eSeveral future directions emerge logically from our findings. First, multi-centre external validation with predefined calibration strategies should be tested for transferability and require definition of when and how often the chart needs local recalibration. Second, phenotypespecific extensions for the clearly identifiable highrisk subgroup (older age, shallow ACD, high LC) should be developed, potentially including interaction terms, hierarchical partial pooling, or Bayesian updating that allows centres to adapt thresholds without sacrificing comparability. Third, to overcome operator dependence, AIassisted UBM segmentation and automated extraction of STS, LC, CP and CSA should be pursued. Integrating these tools into a federated learning framework would allow continuous improvement of the prediction model without exposing sensitive patient data.\u003c/p\u003e\u003cp\u003eFinally, our data demonstrates protocols that can be easily implemented. Hospitals can incorporate UBMguided sizing for complex anatomies and require that technicians meet predefined competency metrics (e.g., interrater ICC\u0026thinsp;\u0026ge;\u0026thinsp;0.90, mean absolute CS Diff measurement error\u0026thinsp;\u0026le;\u0026thinsp;0.10 mm across repeats). By establishing a standardized recommendation for posteriorchamber\u0026ndash;informed sizing and a minimum UBM SOP, multi-centre collaboration can enhance data validation and broaden the model\u0026rsquo;s utility.\u003c/p\u003e\u003cp\u003eIn summary, this study provides robust clinical evidence that a posteriorchamber, UBMbased, machinelearninginformed approach to ICL sizing can significantly improve vault accuracy compared to traditional WTW\u0026thinsp;+\u0026thinsp;ACD methods. Our study supports the translation of this approach into a clinically practical ICL sizing tool. Moving forward, we hope refractive surgeons will transition from the conventional WTW\u0026thinsp;+\u0026thinsp;ACD approach to this more precise, posterior-chamber, UBM-based method.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was approved by the Medical Ethics Committee of Liaoning Aier Eye Hospital (Ethics Number: 2023-007-01) and complied with the Declaration of Helsinki.We obtained informed consent from all participants to participate\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAier Eye Hospital Research Fund\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026apos; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePan Jiayu (PJY) was primarily responsible for writing the entire manuscript. Fang Xuejun (FXJ), as the ICL surgeon, also ensured the integrity and accuracy of the entire model establishment. Wang Yue(WY) and Wang Jifan(WJF) participated in the conception and design of the manuscript. Yu Yidan (YYD) was in charge of data acquisition and analysis. Li Ying(LY) and Li Yingshuai (LYS) formulated the relevant UBM operation standards and standardization protocols. Chu Xiaohan (CXH) conducted UBM examinations and data collection. Shu Yee Seow(SYS) revised the final version. All authors reviewed the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFirst and foremost, I would like to express my deepest gratitude to my supervisor Fang Xuejun, for her \u0026nbsp; invaluable guidance, unwavering support, and insightful comments throughout the entire research process. Her expertise and dedication have been instrumental in shaping this manuscript.I am also highly indebted to all the colleagues and research assistants who have provided assistance and shared their valuable\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003e\u003cu\u003eAlarfaj, G., Helayel, H. B., AlSubaie, M., Hariri, J., Alzaher, F., Khan, O., Al-Jindan, M., AlHabash, A., \u0026amp; Sulaimani, N. M. (2025). Posterior chamber phakic intraocular lens adjustment-causes and complications: a retrospective cohort study. \u003cem\u003eInternational journal of ophthalmology\u003c/em\u003e, \u003cem\u003e18\u003c/em\u003e(5), 883\u0026ndash;888. https://doi.org/10.18240/ijo.2025.05.14\u003c/u\u003e\u003c/li\u003e\n\u003cli\u003eAng, R.E.T., Reyes, E.K.F., Ayuyao, F.A.J. \u003cem\u003eet al.\u003c/em\u003e Comparison of white-to-white measurements using four devices and their determination of ICL sizing. \u003cem\u003eEye and Vis\u003c/em\u003e \u003cstrong\u003e9\u003c/strong\u003e, 36 (2022). https://doi.org/10.1186/s40662-022-00308-z\u003c/li\u003e\n\u003cli\u003eCui, W., Wu, X., Ren, Q., Liu, K., Kong, F., \u0026amp; Wu, J. (2023). A new formula based on new parameters for predicting postoperative vault after posterior chamber intraocular lens implantation: a retrospective study. \u003cem\u003eQuantitative imaging in medicine and surgery\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e(9), 5502\u0026ndash;5510. https://doi.org/10.21037/qims-22-1425\u003c/li\u003e\n\u003cli\u003eDamgaard, I. B., Reffat, M., \u0026amp; Hjortdal, J. (2018). Review of Corneal Biomechanical Properties Following LASIK and SMILE for Myopia and Myopic Astigmatism. \u003cem\u003eThe open ophthalmology journal\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e, 164\u0026ndash;174. https://doi.org/10.2174/1874364101812010164\u003c/li\u003e\n\u003cli\u003eHe, M., Wang, D., \u0026amp; Jiang, Y. (2012). Overview of Ultrasound Biomicroscopy. \u003cem\u003eJournal of current glaucoma practice\u003c/em\u003e, \u003cem\u003e6\u003c/em\u003e(1), 25\u0026ndash;53. https://doi.org/10.5005/jp-journals-10008-1105\u003c/li\u003e\n\u003cli\u003eKane, J. X., Van Heerden, A., Atik, A., \u0026amp; Petsoglou, C. (2016). Intraocular lens\u003c/li\u003e\n\u003cli\u003eKapoor, G., Sharma, V. K., Singh, P., Sahran, D., Seth, S., Ambiya, V., Kalra, N., \u0026amp; Mishra, A. (2025). Objective and subjective outcomes following implantable collamer lens (ICL) implantation for the correction of myopia. \u003cem\u003eOman journal of ophthalmology\u003c/em\u003e, \u003cem\u003e18\u003c/em\u003e(2), 187\u0026ndash;192. https://doi.org/10.4103/ojo.ojo_171_24\u003c/li\u003e\n\u003cli\u003eMont\u0026eacute;s-Mic\u0026oacute;, R., Ta\u0026ntilde;\u0026aacute;-Rivero, P., Aguilar-C\u0026oacute;rcoles, S. \u003cem\u003eet al.\u003c/em\u003e Angle-to-angle and spur-to-spur distance analysis with high-resolution optical coherence tomography. \u003cem\u003eEye and Vis\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 42 (2020). https://doi.org/10.1186/s40662-020-00208-0\u003c/li\u003e\n\u003cli\u003eNakamura, T., Nishida, T., Isogai, N., Kojima, T., Sugiyama, Y., \u0026amp; Yoshida, Y. (2023). Evaluation of implantable collamer lens sizing developed by reviewing the horizontal compression-vault coefficient. \u003cem\u003eJournal of cataract and refractive surgery\u003c/em\u003e, \u003cem\u003e49\u003c/em\u003e(5), 525\u0026ndash;530. https://doi.org/10.1097/j.jcrs.0000000000001140\u003c/li\u003e\n\u003cli\u003ePacker M. (2016). Meta-analysis and review: effectiveness, safety, and central port design of the intraocular collamer lens. \u003cem\u003eClinical ophthalmology (Auckland, N.Z.)\u003c/em\u003e, \u003cem\u003e10\u003c/em\u003e, 1059\u0026ndash;1077. https://doi.org/10.2147/OPTH.S111620\u003c/li\u003e\n\u003cli\u003ePathak, M., Sahu, V., Kumar, A., Kaur, K., \u0026amp; Gurnani, B. (2024). Current Concepts and Recent Updates of Optical Biometry- A Comprehensive Review. \u003cem\u003eClinical ophthalmology (Auckland, N.Z.)\u003c/em\u003e, \u003cem\u003e18\u003c/em\u003e, 1191\u0026ndash;1206. https://doi.org/10.2147/OPTH.S464538\u003c/li\u003e\n\u003cli\u003epower formula accuracy: Comparison of 7 formulas. \u003cem\u003eJournal of Cataract \u0026amp; Refractive\u003c/em\u003e\u003c/li\u003e\n\u003cli\u003eSilverman R. H. (2009). High-resolution ultrasound imaging of the eye - a review. \u003cem\u003eClinical \u0026amp; experimental ophthalmology\u003c/em\u003e, \u003cem\u003e37\u003c/em\u003e(1), 54\u0026ndash;67. https://doi.org/10.1111/j.1442-9071.2008.01892.x\u003c/li\u003e\n\u003cli\u003e\u003cem\u003eSurgery\u003c/em\u003e, \u003cem\u003e42\u003c/em\u003e(10), 1490\u0026ndash;1500. https://doi.org/10.1016/j.jcrs.2016.07.021\u003c/li\u003e\n\u003cli\u003e\u003cu\u003eThompson, V., Cummings, A. B., \u0026amp; Wang, X. (2024). Implantable Collamer Lens Procedure Planning: A Review of Global Approaches. \u003cem\u003eClinical ophthalmology (Auckland, N.Z.)\u003c/em\u003e, \u003cem\u003e18\u003c/em\u003e, 1033\u0026ndash;1043. https://doi.org/10.2147/OPTH.S456397\u003c/u\u003e\u003c/li\u003e\n\u003cli\u003eYang, J., Li, H., Wu, M., He, R., Nong, Y., Zou, Z., Zhang, C., \u0026amp; Zhou, S. (2023). A vault-prediction formula for implantable collamer lens based on preoperative parameters: a retrospective clinical study. \u003cem\u003eBMC ophthalmology\u003c/em\u003e, \u003cem\u003e23\u003c/em\u003e(1), 350. https://doi.org/10.1186/s12886-023-03096-9\u003c/li\u003e\n\u003cli\u003eYao, J., Feng, J., Li, W. \u003cem\u003eet al.\u003c/em\u003e SMILE and ICL implantation on the ocular surface and meibomian glands in patients with postoperative myopia. \u003cem\u003eBMC Ophthalmol\u003c/em\u003e \u003cstrong\u003e24\u003c/strong\u003e, 522 (2024). https://doi.org/10.1186/s12886-024-03790-2\u003c/li\u003e\n\u003cli\u003eZhang, J., He, F., Liu, Y., \u0026amp; Fan, X. (2020). Implantable collamer lens with a central hole for residual refractive error correction after corneal refractive surgery. \u003cem\u003eExperimental and therapeutic medicine\u003c/em\u003e, \u003cem\u003e20\u003c/em\u003e(6), 160. https://doi.org/10.3892/etm.2020.9289\u003c/li\u003e\n\u003cli\u003eZhang, Q., Gong, D., Li, K., Dang, K., Wang, Y., Pan, C., Yan, Z., \u0026amp; Yang, W. (2024). From inception to innovation: bibliometric analysis of the evolution, hotspots, and trends in implantable collamer lens surgery research. \u003cem\u003eFrontiers in medicine\u003c/em\u003e, \u003cem\u003e11\u003c/em\u003e, 1432780. https://doi.org/10.3389/fmed.2024.1432780\u003c/li\u003e\n\u003cli\u003eZhang, Y., Xi, R., Higashita, R., Okamoto, K., Kamiya, K., Miyata, K., Igarashi, A., Hata, S., Nakamura, T., \u0026amp; Liu, J. (2025). Prior Anatomical Knowledge-guided GAN for ICL surgery postoperative prediction based on AS-OCT image. \u003cem\u003eMedical image analysis\u003c/em\u003e, \u003cem\u003e105\u003c/em\u003e, 103689. Advance online publication. https://doi.org/10.1016/j.media.2025.103689\u003c/li\u003e\n\u003cli\u003eZhong, X., Li, Y., Li, Y., Wang, G., Du, Y., \u0026amp; Zhang, M. (2024). Comparison of Predictability in Vault Using NK Formula and KS Formula for the Implantable Collamer Lens Surgery. \u003cem\u003eJournal of ophthalmology\u003c/em\u003e, \u003cem\u003e2024\u003c/em\u003e, 4256371. https://doi.org/10.1155/2024/4256371\u003c/li\u003e\n\u003cli\u003eZhu, Q. J., Zhu, W. J., Chen, W. J., Ma, L., \u0026amp; Yuan, Y. (2022). A prediction model for sulcus-to-sulcus diameter in myopic eyes: a 1466-sample retrospective study. \u003cem\u003eBMC ophthalmology\u003c/em\u003e, \u003cem\u003e22\u003c/em\u003e(1), 307. https://doi.org/10.1186/s12886-022-02535-3\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Supplementary Table ","content":"\u003cp\u003eSupplementary Table S1 is not available with this version.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Implantable Collamer Lens, vault prediction, ultrasound biomicroscopy, sulcus-to-sulcus, posterior chamber anatomy, machine learning, ICL sizing, myopia correction","lastPublishedDoi":"10.21203/rs.3.rs-7400290/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7400290/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground:\u003c/strong\u003e\u003cbr\u003e\nAccurate prediction of post-operative vault is critical for safe and effective implantation of Implantable Collamer Lenses (ICLs). Traditional sizing methods rely on anterior segment parameters, such as white-to-white (WTW) and anterior chamber depth (ACD), are poorly correlated to the with sulcus-to-sulcus (STS) diameter where the ICL rests. Inappropriate sizing can result in suboptimal vaults, increasing the risk of complications. Ultrasound biomicroscopy (UBM) allows for direct assessment of posterior chamber structures but it has yet to be systematically integrated into predictive models at scale.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods:\u003c/strong\u003e\u003cbr\u003e\nWe developed and validated a UBM-based, data-driven model for vault prediction using \u0026gt;2,000 ICL implantations performed between January 2023 and January 2025 at a single center. Pre-operative variables extracted from standardized UBM protocols include horizontal and vertical STS (STSH, STSV), lens curvature (LC), Posterior Chamber Angle (CSA) Morphology and Ciliary Process (CP) Size,. A random forest model was trained to identify dominant anatomical predictors. Based on these results, a posterior chamber asymmetry index (CS Diff = |STSH – STSV|) and empirically derived LC thresholds were used to construct a clinically practical ICL size selection chart. This chart was prospectively validated in 286 patients (568 eyes) operated between May and June 2025.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults:\u003c/strong\u003e\u003cbr\u003e\nRandom Forest and SHAP analyses identified STSH, STSV, LCH, and LCV as the most influential predictors. Higher CS Diff were associated with low vaults; greater LC predicted higher vaults. In the validation cohort, the selection map yielded ideal vaults (250–750 μm) in 92.96% of eyes. This represented a significant improvement over the historical WTW+ACD method (ideal rate \u0026lt;76.51\u0026gt;%, p\u0026lt;0.001). Decision-curve analysis demonstrated superior net clinical benefit across a wide range of threshold probabilities. Subgroup analyses confirmed robust performance, including in those with high-risk phenotypes (older age, shallow ACD, high LC).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusions:\u003c/strong\u003e\u003cbr\u003e\nPosterior chamber geometry—especially STS asymmetry and lens curvature—drives vault variability more than traditional anterior segment metrics. Our validated UBM-guided model enhances individualized ICL sizing and significantly improves clinical outcomes. Adoption of such models into routine practice could reduce re-operations and establish a new standard of care in refractive surgery.\u003c/p\u003e","manuscriptTitle":"A UBM-Based Big Data Model for Individualized ICL Size Selection in Myopic Eyes","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-09-02 06:52:21","doi":"10.21203/rs.3.rs-7400290/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"cd2fabd4-2883-475e-9026-3bc86d092094","owner":[],"postedDate":"September 2nd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-09-15T05:09:00+00:00","versionOfRecord":[],"versionCreatedAt":"2025-09-02 06:52:21","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7400290","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7400290","identity":"rs-7400290","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}