Machine Learning IOL Position Prediction via CNN and Graph Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining the final position and orientation of an intraocular lens (IOL) after surgery are inefficient, relying on manual extraction of geometric properties from OCT images, which limits the prediction accuracy and requires multiple steps.
Innovation Solution
A machine learning-supported method using a combination of deep neural networks and graph-based neural networks to directly determine the final location of the IOL by processing digital data from eye scans, including OCT images, to predict the IOL position and orientation after a growing-in phase.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual extraction of geometric properties from OCT images is used, then the process is simple and straightforward, but the prediction accuracy of IOL position is limited and important information is lost
Solution Approach 1:
A deep neural network (CNN) is introduced as an intermediary between the OCT scan data and the final IOL position prediction. The CNN processes the raw scan data to extract relevant features and transform them into a format suitable for the graph-based neural network, thereby improving prediction accuracy while managing complexity through specialized intermediate processing
Solution Approach 2:
The system combines multiple types of data (OCT scan data, biometric parameters, geometric properties) and multiple processing approaches (CNN for feature extraction, graph-based neural network for relationship modeling) into a composite prediction system that leverages the strengths of each component to achieve superior accuracy
2Loss of information
If manual extraction of geometric properties is performed, then only a few predefined variables are obtained, but a large proportion of important information in the digital images is lost
Solution Approach 1:
The graph-based neural network selectively extracts and processes only the most relevant features and relationships from the comprehensive scan data, rather than processing all raw data uniformly. This allows the system to retain important information while improving processing efficiency by focusing computational resources on critical predictors of IOL position
Solution Approach 2:
The information processing is segmented into distinct stages: the CNN handles initial feature extraction from raw scan data, while the graph-based neural network handles relationship modeling and final prediction. This segmentation allows each component to optimize for its specific task, improving both information retention and processing speed
3Reliability
If multiple processing steps are used for IOL position determination, then comprehensive analysis is performed, but the process becomes time-consuming and error-prone
Solution Approach 1:
The system merges the feature extraction function and the prediction function into a unified deep learning pipeline. The CNN and graph-based neural network work together in an integrated architecture that processes data through multiple stages automatically, reducing manual intervention and minimizing opportunities for human error while maintaining comprehensive analysis
Solution Approach 2:
The deep neural network performs preliminary processing and feature extraction from the raw scan data before the main prediction algorithm is applied. This preliminary action prepares the data in an optimized format, reducing the computational burden and time required for the final prediction step while ensuring that all relevant information is properly processed
Data Source
AI summary
The invention relates to a computer-assisted method for position determination for an intraocular lens supported by machine learning. The method comprises providing a scan result for an eye. The scan result here represents an image of an anatomical structure of the eye. The method further comprises use of a trained machine learning system for the direct determination of a final location of an intraocular lens to be fitted, wherein digital data of the scan of the eye is used as the input data for the machine learning system.