Neural Network Ejection Fraction Determination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current ejection fraction measurement methods, such as echocardiography, are inaccurate and subjective due to manual tracing of ventricle borders and reliance on elliptical cylinder models, leading to delayed and worsened disease diagnosis.
Innovation Solution
A neural network is trained with cardiac imaging data from different hearts with known ejection fractions, allowing for automated ejection fraction determination without manual tracing, using pre-processing and filtering of imaging data to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If manual tracing of ventricle borders is performed to measure ejection fraction, then the measurement can be obtained using traditional echocardiography methods, but the measurement accuracy deteriorates due to human subjectivity and error
Solution Approach 1:
The patent replaces the manual mechanical tracing process with an automated image processing system that uses edge detection algorithms and contour analysis. The system automatically identifies ventricle borders in echocardiogram images through computational methods, eliminating human subjectivity while maintaining ease of operation. This substitution of manual mechanical tracing with automated computational analysis directly resolves the contradiction between operational simplicity and measurement precision.
2Ease of manufacture
If elliptical cylinder models are used to compute ventricle volume, then the calculation process is simplified and convenient, but the measurement accuracy deteriorates due to model approximation errors
Solution Approach 1:
The patent extracts the ventricle border contours directly from the imaging data without fitting them to predefined geometric models like elliptical cylinders. By taking out the actual boundary information from the images through automated edge detection and contour tracing, the system eliminates the approximation errors inherent in model-based approaches while maintaining computational efficiency through direct pixel-based area and volume calculations.
3Measurement precision
If exact end systolic and end diastolic image frames are identified to compute fractional difference, then the ejection fraction can be calculated using Simpson Biplane methodology, but the process becomes complex and error-prone
Solution Approach 1:
The patent applies preliminary action by automatically detecting and marking the end-systolic and end-diastolic frames during the image acquisition and preprocessing stage. The system uses motion analysis and boundary change detection to identify these critical phases before the actual measurement is performed. This preliminary identification eliminates the need for complex manual frame-by-frame analysis by the operator, reducing process complexity while ensuring accurate timing for the ejection fraction calculation.
4Measurement precision
If automated neural network methods are used to determine ejection fraction, then measurement accuracy and reproducibility are improved, but the computational complexity and processing requirements increase
Solution Approach 1:
The patent applies preliminary action by performing extensive preprocessing of the training data and the input images before neural network processing. This includes image normalization, enhancement, feature extraction, and annotation of ground truth data during the training phase. For inference, the system pre-processes input images to standardize them and extract relevant features before feeding them to the neural network. This preliminary preparation reduces the computational burden during actual measurement while maintaining high accuracy, as the neural network receives optimized, pre-processed data rather than raw images.
Data Source
AI summary
Embodiments of the invention provide a method, system and computer program product for artificially intelligent ejection fraction determination. In a method for artificially intelligent ejection fraction determination, a neural network is loaded into memory of a computer, that has been trained with different sets of cardiac imaging data acquired during imaging of a ventricle for different hearts and a known ejection fraction for each of the sets. Then, a contemporaneous set of imaging data is acquired of a ventricle of a heart and the contemporaneous set of imaging data is provided to the neural network. Finally, an ejection fraction determination output by the neural network is displayed in a display of the computer without tracing a ventricle boundary of the heart.