Ultrasound Cross-Section Learning With Probe-Constrained Pseudo Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for augmenting learning data in medical images, particularly ultrasound images, do not adequately consider the positional relation between the image acquisition probe and the subject, leading to the generation of images that cannot be acquired in actual scenarios and difficulty in effectively generating variations that align with real-world imaging conditions.
Innovation Solution
An image processing apparatus and method that utilizes a teacher model to train a student model by acquiring pseudo standard cross-section images based on the relation between the imaging plane of a three-dimensional imaging probe and the standard cross-section, generating pseudo ground truth data to augment the learning data, thereby improving estimation precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If conventional data augmentation methods are used to generate learning data variations, then the quantity of learning data is increased, but the reliability of generated images is degraded because images that cannot be acquired in actual ultrasound scenarios are produced
Solution Approach 1:
The patent changes the parameters used for data augmentation by introducing probe position and orientation parameters that reflect actual ultrasound imaging conditions. Instead of arbitrary transformations, the augmentation applies realistic variations based on probe placement geometry, ensuring generated images maintain physical feasibility while increasing data quantity.
Solution Approach 2:
The patent introduces probe position and orientation information as an intermediary element that mediates between the image generation process and reality. This intermediary ensures that all generated cross-section images correspond to physically achievable probe positions, filtering out unrealistic variations while preserving data diversity.
2Adaptability or versatility
If arbitrary cross-section variations are generated for data augmentation, then data diversity is improved, but the adaptability to real-world imaging conditions is degraded
Solution Approach 1:
The patent modifies the augmentation approach by changing from arbitrary cross-section variations to variations constrained by probe geometry parameters. This allows diverse data generation while maintaining adaptability to real-world conditions through physically meaningful parameter transformations.
3Measurement precision
If more learning data with variations is collected, then estimation precision is improved, but the difficulty of data collection is worsened due to the scarcity of medical imaging data
Solution Approach 1:
The patent creates copies of existing medical images with realistic variations by simulating different probe positions and orientations. This copying approach generates synthetic training data that mimics real-world variability without requiring additional physical data collection, thereby improving estimation precision while avoiding data collection difficulties.
Solution Approach 2:
The patent applies parameter changes to existing images by varying probe position and orientation parameters, generating diverse training samples from limited source data. This approach overcomes data scarcity by creating parameter-based variations rather than requiring collection of new physical images.
Data Source
AI summary
An image processing apparatus includes an image acquisition unit that acquires a standard cross-section image from a three-dimensional image including an object, a teacher data acquisition unit that acquires teacher data including the standard cross-section image and ground truth data that is information regarding the object, a first learning unit that constructs a first learning model, a pseudo standard cross-section image acquisition unit that acquires a cross-section that differs from the standard cross-section as a pseudo standard cross-section image, based on a relation between an imaging plane of a three-dimensional imaging probe and the standard cross-section, a pseudo ground truth data acquisition unit that acquires pseudo ground truth data using the first learning model, and a second learning unit that constructs a second learning model by performing training of pseudo teacher data including the pseudo standard cross-section image and the pseudo ground truth data.