Machine Learning Training Data Volume Expansion for IOL Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The limited availability of training data in ophthalmology for machine learning systems used in cataract operations leads to subcritical data volumes, resulting in low prediction precision, and existing solutions for updating these systems are complicated and inefficient.
Innovation Solution
A computer-implemented method and system that measures ophthalmological biometry data and determines initial refractive power values for intraocular lenses using a trained machine learning system, forming new training data records and calculating an importance indicator value to assess the value of additional data, thereby incentivizing continuous data updates and improving prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning systems are trained with limited training data available in ophthalmology, then the system can be deployed, but the prediction precision remains low
Solution Approach 1:
The system implements a feedback mechanism where postoperative refractive results values are measured and fed back to form new training data records. This continuous feedback loop allows the machine learning system to progressively improve prediction precision by incorporating actual clinical outcomes into the training data, directly addressing the limitation of initially limited training data volume.
Solution Approach 2:
The system enables self-service by automatically collecting postoperative results and generating new training data records without requiring manual data collection efforts. The machine learning system serves itself by using its own predictions and the measured actual outcomes to continuously expand and improve its training dataset, thereby enhancing prediction precision autonomously.
2Quantity of substance
If additional training data from other clinics are collected, then the training data volume increases, but the data exchange process becomes complicated and error-prone
Solution Approach 1:
The system achieves universality by designing a standardized data format that can be used across different clinics and systems. The training data records follow a consistent structure containing ophthalmological biometry data, initial refractive power values, and postoperative refractive results values, enabling seamless data exchange and aggregation from multiple sources without requiring complex custom integration for each clinic.
Solution Approach 2:
The system applies parameter changes by standardizing the data representation format and structure. By defining uniform parameters for storing and exchanging training data (biometry data, refractive power values, results values), the system eliminates the complexity and errors associated with exchanging data in different formats between clinics, while still allowing volume expansion.
3Measurement precision
If the machine learning system is retrained regularly with updated data, then the prediction accuracy improves, but the time and resources required for updates increase
Solution Approach 1:
The system applies partial action by incrementally updating the training data with new postoperative results records rather than requiring complete retraining with all data. The machine learning system can process and learn from individual new training data records or small batches, reducing the time investment required for each update cycle while still improving prediction accuracy progressively.
4Quantity of substance
If clinics manually exchange patient data to update machine learning systems, then the training data can be updated, but the process is susceptible to errors and organizational complications
Solution Approach 1:
The system replaces the manual mechanical process of data exchange with an automated electronic system. Training data records are automatically generated from measured postoperative refractive results values and stored in a standardized format, eliminating the need for manual data collection, formatting, and exchange between clinics. This automation significantly reduces errors and organizational complications while maintaining data freshness.
Data Source
AI summary
A computer-implemented method for increasing a training data volume for a machine learning system for determining an initial refractive power value for an intraocular lens to be inserted is described. The method includes measuring a group of ophthalmological biometry data of a patient and determining an initial refractive power value for the intraocular lens to be inserted by a trained machine learning system. The measured ophthalmological biometry data and a postoperative target refraction value are used as input data for the trained machine learning system. The method also includes measuring a postoperative refractive results value, assigning the postoperative refractive results value to the measured ophthalmological biometry data of the patient, and determining an importance indicator value for the new training data record.