Multi-class classifier for ultrasound anatomical detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current medical imaging technologies, such as echocardiography, face challenges in consistently capturing standard cardiac views due to variations in machine properties and sonographer skill, leading to inefficient workflow and suboptimal imaging settings, particularly in identifying multiple objects and evaluating responses from multiple binary classifiers.
Innovation Solution
The implementation of a multi-class classifier, trained using algorithms like LogitBoost, to identify anatomical features in medical images, allowing for automatic adjustment of imaging parameters and simultaneous detection of multiple objects, thereby improving workflow efficiency and consistency by reducing user variability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple binary classifiers are used to detect different anatomical objects, then detection coverage is improved, but training complexity and evaluation time increase linearly
Solution Approach 1:
The patent merges multiple binary classifiers into a single multi-class classifier that can simultaneously detect and distinguish between multiple anatomical objects (e.g., ventricles, atria, valves) in one unified model. This eliminates the need to train and manage separate binary classifiers for each anatomical structure, reducing training complexity while maintaining comprehensive detection coverage.
Solution Approach 2:
The multi-class classifier serves multiple functions simultaneously - it can identify different anatomical objects, their spatial relationships, and their functional characteristics in a single classification framework. This universal approach replaces the need for specialized binary classifiers for each anatomical structure.
2Measurement precision
If separate binary classifiers are trained for each anatomical object, then object-specific detection is improved, but determining actual objects becomes more complex due to difficulty in comparing responses
Solution Approach 1:
The patent combines the detection and classification functions into a single multi-class classifier that outputs both the presence of multiple anatomical objects and their identities simultaneously. This eliminates the need to compare responses from multiple binary classifiers, as the unified model directly provides the classification results.
3Ease of operation
If presets with predetermined imaging parameters are used, then workflow is simplified, but the list becomes unmanageable as it must contain all variations and sub-cases
Solution Approach 1:
The system automatically adjusts imaging parameters based on the detected anatomical structures without requiring manual selection from preset lists. The multi-class classifier analyzes the ultrasound image and autonomously determines optimal imaging settings for the detected anatomy, eliminating the need for users to manage extensive preset lists.
Solution Approach 2:
The imaging parameters are dynamically adjusted based on the specific anatomical structures detected in each image, rather than relying on static preset lists. The system adapts imaging settings in real-time according to the detected ventricles, atria, valves, or other anatomical features.
4Manufacturing precision
If manual adjustment of imaging parameters is performed, then imaging quality can be optimized, but variability increases due to sonographer skill and preference
Solution Approach 1:
The system performs self-adjustment of imaging parameters based on automated anatomical detection, eliminating variability introduced by different sonographers. The multi-class classifier provides consistent, reproducible imaging settings for the same anatomical structures regardless of who operates the system.
Solution Approach 2:
The system uses feedback from the multi-class classifier's anatomical detection to automatically adjust imaging parameters. This closed-loop approach ensures consistent imaging quality by continuously monitoring and adapting to the detected anatomical structures, eliminating human variability.
Data Source
AI summary
Anatomical information is identified from a medical image and/or used for controlling a medical diagnostic imaging system, such as an ultrasound system. To identify anatomical information from a medical image, a processor applies a multi-class classifier. The anatomical information is used to set an imaging parameter of the medical imaging system. The setting or identification may be used in combination or separately.