Cardiac Echo View Synthesis for Foreshortening Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Foreshortening in cardiac echo imaging leads to geometric distortion of the left ventricle, causing inaccurate measurements of clinical parameters such as ejection fraction and global strain, and existing methods fail to prevent or correct this issue during image acquisition.
Innovation Solution
Utilizing view synthesis to generate a synthetic non-foreshortened image by inferring the heart's pose from a foreshortened view using a trained image segmentation model and heart model, and transferring this pose to a new image plane to correct the view.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the ultrasound probe is positioned directly on the apex of the heart to obtain an optimal A4C view, then image quality and measurement accuracy are improved, but the procedure becomes more difficult to acquire due to factors such as obstruction by patient's ribs, low image contrast, and non-ideal image conditions
Solution Approach 1:
The system creates a synthetic copy of the heart structure from the foreshortened view by inferring 3D pose and generating a virtual non-foreshortened image. This synthetic copy allows accurate measurements to be obtained without requiring the operator to achieve the difficult optimal probe position, thus resolving the contradiction between measurement precision and ease of operation.
Solution Approach 2:
The system introduces an intermediary computational model that bridges the foreshortened view and the desired non-foreshortened view. By inferring 3D pose from the difficult-to-obtain view and synthesizing the ideal view computationally, the intermediary enables accurate measurements without requiring direct optimal positioning during acquisition.
2Ease of operation
If the ultrasound probe is not positioned directly on the apex of the heart, then the acquisition becomes easier under non-ideal conditions, but the view becomes foreshortened causing geometric distortion of the left ventricle
Solution Approach 1:
The system generates a synthetic copy of the heart in the correct non-foreshortened orientation from the foreshortened input view. This virtual copy preserves the true geometric shape of the left ventricle without the distortion present in the original acquired image, allowing operators to easily acquire images while maintaining measurement accuracy through the synthesized correct geometry.
Solution Approach 2:
The system transforms the 2D foreshortened image into a 3D pose representation, then renders a new 2D image from the correct orientation. By moving to a 3D intermediate representation, the system can correct geometric distortion while maintaining ease of acquisition from the original difficult-to-obtain view.
3Measurement precision
If existing methods detect foreshortening after acquisition to exclude inaccurate measurements, then measurement accuracy is maintained by excluding bad data, but the process becomes more complex and time-consuming
Solution Approach 1:
Instead of detecting and excluding foreshortened images after acquisition, the system performs preliminary action by automatically correcting the foreshortening in real-time during acquisition. The synthetic non-foreshortened image is generated immediately from the input view, eliminating the need for subsequent detection and exclusion processes, thus reducing complexity while maintaining accuracy.
Solution Approach 2:
The system provides immediate feedback by generating the corrected non-foreshortened image from the input view, allowing operators to see the correction in real-time. This feedback mechanism eliminates the need for post-acquisition detection and manual review, simplifying the overall process while ensuring measurement accuracy through automated correction rather than exclusion.
Data Source
AI summary
Systems and methods for automatic correction of view foreshortening in cardiac echo using view synthesis. A trained image segmentation method is used to infer a 2D pose from the input image. The pose is transferred to a new view representing a non-foreshortened view plane. A non-foreshortened image is then rendered.