Graefe's Archive of Clinical and Experimental Ophthalmology | 03 October 2025
Development and validation of machine learning for pathological changes in high myopia patients with tessellated fundus: a cross-sectional study
Pathological myopia causes irreversible damage, yet early changes in high myopia patients with tessellated fundus are often overlooked. The identification of such patients relies on OCT and fundus imaging, but early detection is difficult. Therefore, we developed and validated machine learning models using lifestyle and clinical parameters.
Authors: Huiyi Zuo1· Hai Huang2 · Baoyu Huang1 · Jian He1 · Xin Liu1 · Lijia Huang1 · Fulan Bi1 · Minli Huang1 ✅
Executive Impact Assessment
Early detection of pathological changes in high myopia patients with tessellated fundus is critical for preventing irreversible vision damage, yet current methods are often insufficient or overlooked in primary care settings.
Failure to identify early pathological changes can lead to delayed intervention, progression to severe forms of pathological myopia, and ultimately, irreversible low vision or blindness for patients, increasing healthcare burden.
0.0
Max AUC Achieved
0
Eyes Analyzed
0
ML Models Validated
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
Study Overview
This study focuses on leveraging machine learning to detect early pathological changes in high myopia patients with tessellated fundus. Traditional methods often overlook these subtle changes, making early intervention difficult. The researchers built models based on lifestyle and clinical parameters, which showed promising accuracy in identifying at-risk individuals.
Methodology Insights
A cross-sectional study design was used, collecting data from high myopia patients with tessellated fundus. Data was split into training and test sets (7:3 ratio). LASSO regression with five-fold cross-validation was employed for variable selection. Four machine learning models (Random Forest, SVM, Linear SVM, XGBoost) were constructed and evaluated using AUC, sensitivity, specificity, calibration curves, and decision curve analysis.
Key Findings
Out of 529 analyzed eyes, 329 were pathological myopia (PM) and 200 non-PM. Key variables identified for modeling included education level, daily iPad usage, reading time, lamp light source, outdoor activity, spherical equivalent, axial length, corneal/lens thickness, PPA extent/radius, and disc tilt ratio. AUCs in the test set ranged from 0.765 to 0.839 across the models, demonstrating good discrimination and clinical applicability.
0.839
Highest AUC in Test Set (XGBoost)
The XGBoost model demonstrated the highest discriminative power in the test set, indicating its strong potential for identifying pathological changes.
Machine Learning Model Development Workflow
Data Collection (High Myopia Patients with Tessellated Fundus)
→
Data Split (Training/Test)
→
LASSO Regression (Variable Selection)
→
ML Model Construction (RF, SVM, XGBoost)
→
Performance Evaluation (AUC, Sensitivity, Specificity)
→
Clinical Applicability Assessment (Calibration, Decision Curves)
| Model | AUC (95% CI) | Sensitivity | Specificity |
|---|---|---|---|
| Random Forest | 0.834 (0.769–0.899) | 0.667 | 0.889 |
| XGBoost | 0.839 (0.776–0.902) | 0.743 | 0.815 |
| SVM-Kernel | 0.815 (0.748–0.882) | 0.676 | 0.833 |
| SVM-Linear | 0.765 (0.690–0.840) | 0.619 | 0.778 |
Note: XGBoost showed the highest AUC, while Random Forest achieved the best specificity. The models provide a balanced performance for early screening.
Impact of Early Detection in Primary Care
Scenario: A 35-year-old patient with high myopia and tessellated fundus visits a primary care clinic. Traditional fundus photography might miss subtle early pathological changes. Using the developed ML model, the patient's lifestyle habits (e.g., extensive daily reading time) and clinical parameters (e.g., axial length, disc tilt ratio) are input.
Outcome: The ML model identifies the patient as high-risk for developing pathological myopia with a 78% probability. Based on this, the clinician refers the patient for advanced OCT imaging, revealing nascent retinal thinning and early PPA expansion not visible on standard fundus photos. Early intervention, including lifestyle modifications and closer monitoring, is initiated, significantly reducing the risk of irreversible vision loss from conditions like myopic maculopathy.
Benefit: This proactive approach, enabled by the ML model, demonstrates a significant improvement over traditional screening methods, allowing for timely and targeted interventions that preserve vision and improve patient outcomes.
Advanced ROI Calculator
Estimate the potential return on investment for integrating AI-powered early detection models into your clinical practice.
Implementation Roadmap
Our structured approach ensures a seamless integration of AI models into your existing workflows, maximizing clinical and operational benefits.
Data Preparation & Preprocessing
Gather comprehensive patient data, including lifestyle and clinical parameters, and ensure data quality and formatting for model input.
Model Development & Training
Construct and train various machine learning models (e.g., Random Forest, XGBoost) using selected variables and cross-validation techniques.
Validation & Refinement
Rigorously test model performance using AUC, sensitivity, and specificity on independent test sets, and refine models for optimal clinical applicability.
Clinical Integration & Monitoring
Deploy the validated models as an auxiliary tool in clinical settings, continuously monitor performance, and gather feedback for ongoing improvements.
Ready to Transform Your Enterprise?
Schedule a personalized consultation to explore how our AI solutions can address your specific challenges and drive tangible results.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat