TEE Standard View Detection Using 0-Degree Omniplane Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional transesophageal echocardiography (TEE) ultrasound imaging requires a large number of standard views, making it challenging to train artificial intelligence algorithms accurately and efficiently for automatic detection.
Innovation Solution
A method and system that automatically detect standard two-dimensional views in TEE ultrasound images by analyzing images at a zero-degree omniplane angle and adjusting the angle to identify other views using a deep neural network and view detection processor.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the number of standard views in TEE ultrasound is increased to provide detailed heart views, then the comprehensiveness of cardiac imaging is improved, but the number of images required for training AI algorithms increases substantially, reducing training efficiency and accuracy
Solution Approach 1:
The patent uses a deep neural network to learn the mapping relationship between standard views at 0-degree omniplane angle and other standard views at non-zero angles. Instead of training the AI on all 22 standard views, the system trains on a reduced subset (9 views at 0-degree angle) and uses the learned angular transformation relationships to generate or infer the remaining views, significantly reducing training data requirements while maintaining comprehensive cardiac imaging capability
Solution Approach 2:
The patent makes the 0-degree omniplane angle views serve multiple functions: they are used directly for diagnosis and also serve as training data for AI algorithms that learn to generate or identify other standard views at different angles. This multi-functional use of a reduced image set improves both training efficiency and maintains comprehensive cardiac assessment
2Reliability
If the number of standard views in TEE ultrasound is increased to provide detailed heart views, then the comprehensiveness of cardiac imaging is improved, but the training time for AI algorithms increases substantially
Solution Approach 1:
The patent extracts and utilizes only the essential feature set from the complete TEE view set by focusing on the 9 standard views at 0-degree omniplane angle. The deep neural network learns the angular transformation relationships from this extracted subset, eliminating the need to process all 22 standard views during training, thereby substantially reducing training time while preserving comprehensive cardiac imaging capability
Solution Approach 2:
The system performs preliminary training on a reduced, representative subset of TEE views (0-degree angle views) before deployment. The deep neural network is pre-trained on this manageable dataset to learn angular transformation patterns, enabling rapid inference and view identification without requiring extensive training on the complete set of 22 standard views
3Reliability
If conventional methods are used to detect standard views in TEE ultrasound, then all 22 standard views must be processed, but the accuracy of AI detection algorithms decreases due to the large number of images
Solution Approach 1:
The patent segments the complex task of detecting all 22 TEE standard views into two manageable components: (1) detecting views at the 0-degree omniplane angle using a deep neural network, and (2) inferring or generating views at other angles using learned angular transformation relationships. This segmentation reduces the immediate processing burden and improves detection accuracy by focusing the AI on a smaller, more manageable view subset
Data Source
AI summary
Systems and methods for automatically detecting two-dimensional standard views in transesophageal ultrasound images are provided. The method includes acquiring a first ultrasound image at a first depth and a first omniplane angle. The first ultrasound image corresponds with a first standard view. The first omniplane angle is an angle different from zero (0) degrees. The method includes adjusting the first omniplane angle to a second omniplane angle of 0 degrees. The method includes acquiring a second ultrasound image at the first depth with the second omniplane angle. The method includes automatically analyzing the second ultrasound image to identify a second standard view corresponding with the second omniplane angle. The method includes automatically determining the first standard view based on the second standard view and the first omniplane angle. The method includes causing a display system to present the first ultrasound image with a first identifier of the first standard view.