Hybrid IOL Refractive Power Prediction via Physics-Constrained ML
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining the refractive power of intraocular lenses (IOLs) rely on approximations and are limited in accuracy, as they fail to fully replicate the complexity of the biological eye, and require extensive clinical data for training, which can be impractical and prone to measurement errors.
Innovation Solution
A computer-implemented method using a machine learning system trained with clinical ophthalmological data and a physical model, where the loss function incorporates both clinical data and physical model limitations to ensure physically consistent predictions, allowing for fewer required data points and improved robustness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If physical models are used to determine IOL refractive power, then physical knowledge is incorporated into the determination, but the models are only approximations and cannot reproduce the full complexity of the biological eye
Solution Approach 1:
The patent combines physical models with machine learning models into a hybrid system. The physical model provides a baseline determination of refractive power based on established optical principles, while the machine learning model learns from clinical data to capture complex biological variations. The two models work together to produce a final prediction that benefits from both physical consistency and data-driven accuracy, resolving the contradiction between relying on approximate physical models and capturing full biological complexity.
Solution Approach 2:
The machine learning model serves as an intermediary between the physical model and the complex biological reality. It translates clinical measurements into corrections or adjustments to the physical model's output, bridging the gap between simplified physical approximations and the full complexity of the biological eye without requiring complete abandonment of physical principles.
2Measurement precision
If ray tracing methods are used to improve model accuracy, then accuracy is enhanced beyond paraxial approximation, but approximations are still included and flexibility is restricted by the chosen model structure
Solution Approach 1:
The patent employs a dynamic hybrid modeling approach where the machine learning model can adapt its predictions based on the specific input characteristics. Rather than being constrained to a fixed physical model structure, the system dynamically adjusts by applying data-driven corrections to the physical model output, allowing flexibility in how different patient anatomies are handled while maintaining computational efficiency.
3Measurement precision
If extensive clinical training data is used for machine learning, then the model can be trained to determine refractive power, but it requires large amounts of data which is impractical and prone to measurement errors
Solution Approach 1:
The physical model acts as an intermediary that generates synthetic training data with known ground truth values. This synthetic data serves as a proxy for extensive clinical data, allowing the machine learning model to be trained on physically consistent examples without requiring large volumes of actual patient measurements, thereby reducing both data quantity requirements and exposure to measurement errors in clinical datasets.
Solution Approach 2:
The system performs preliminary generation of synthetic training data using the physical model before actual machine learning training begins. This preliminary action creates a robust foundation of training examples that are free from measurement errors, reducing the need for extensive clinical data collection and preprocessing.
4Adaptability or versatility
If purely data-driven machine learning is used, then the model can adapt to clinical variations, but it requires many data points for training and lacks physical constraints
Solution Approach 1:
The patent merges data-driven machine learning with physics-based constraints through a hybrid architecture. The machine learning model captures clinical variations by learning from diverse patient data, while the physical model provides constraints and generates synthetic training data. This combination allows the system to adapt to clinical variations with fewer actual training data points, as the physical model supplements the training dataset with physically plausible examples.
Data Source
AI summary
A computer-implemented method for determining the refractive power of an intraocular lens includes providing a physical model for determining refractive power and training a machine learning system with clinical ophthalmological training data and associated desired results to form a learning model for determining the refractive power. A loss function for training includes: a first component taking into account clinical ophthalmological training data and associated and desired results and a second component taking into account limitations of the physical model wherein a loss function component value is greater the further a predicted value of the refractive power during the training is from results of the physical model with the same clinical ophthalmological training data as input values. Moreover, the method includes providing ophthalmological data of a patient and predicting the refractive power of the intraocular lens to be used by means of the trained machine learning system.