Predictive Model Shape Tracking for Low-Resolution Medical Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current medical imaging technologies face challenges in generating high-resolution segmentations of targets, such as the heart, from low-resolution images due to issues like inter-slice shifts, cardiac motion, and noisy ECG gating, which affect the accuracy and efficiency of diagnosis, particularly in dynamic organs like the heart and lung.
Innovation Solution
A method and system that utilize predictive models, including convolutional neural networks, to determine shape parameters and pose transformations from low-resolution medical images, enabling the generation of high-resolution segmentations and tracking of target shapes over time, using shape and motion models to improve image resolution and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution volumetric CMR imaging is performed to obtain rich diagnosis information, then diagnostic accuracy is improved, but acquisition time increases and patient compliance decreases
Solution Approach 1:
The system performs preliminary action by acquiring low-resolution images first, then uses predictive models to generate high-resolution segmentations retrospectively. This allows the actual high-resolution imaging to be done in advance or offline, reducing the patient's breath-hold time during the actual scan while still providing diagnostic-quality images.
Solution Approach 2:
The patent creates a copy of the target organ's shape and segmentation from low-resolution images using predictive models. Instead of directly acquiring high-resolution images, the system generates a high-resolution representation (copy) of the target by applying learned shape priors and transformations to the low-resolution input, achieving diagnostic quality without the time penalty of direct high-resolution scanning.
2Productivity
If low-resolution multi-slice imaging is used to reduce acquisition time, then scanning speed is improved, but image quality deteriorates due to inter-slice shifts and motion artifacts
Solution Approach 1:
The patent introduces an intermediary predictive model that acts as a mediator between low-resolution images and high-resolution segmentations. This model learns the relationship between low-resolution inputs and high-resolution outputs from training data, then uses this learned knowledge to generate accurate segmentations from fast low-resolution scans, effectively bridging the quality gap without sacrificing scanning speed.
Solution Approach 2:
The system changes parameters by learning transformations that map low-resolution image characteristics to high-resolution segmentation properties. The predictive model adjusts shape parameters, pose transformations, and segmentation boundaries based on the low-resolution input, effectively transforming the quality parameters of the output without changing the acquisition speed of the input.
3Measurement precision
If 2D segmentation algorithms are applied to low-resolution images, then in-plane segmentation resolution is improved, but out-of-plane accuracy deteriorates due to slice artifacts and motion
Solution Approach 1:
The patent transitions from 2D segmentation to 3D segmentation by incorporating volumetric information and shape priors. The predictive model learns 3D shape representations and applies them across multiple slices, enabling accurate out-of-plane segmentation while maintaining in-plane resolution. This dimensional extension allows the system to overcome the limitations of 2D algorithms that cannot handle inter-slice variations effectively.
4Measurement precision
If direct 3D segmentation is performed to achieve accurate target representation, then segmentation accuracy is improved, but computational complexity increases due to out-plane interpolation
Solution Approach 1:
The system performs preliminary action by pre-training predictive models on large datasets of paired low-resolution and high-resolution images. This offline training phase captures the complex relationships between different resolutions and segmentation qualities. During actual use, the pre-trained model quickly generates segmentations from low-resolution inputs without requiring computationally intensive real-time 3D interpolation, thus reducing online computational complexity while maintaining accuracy.
Data Source
AI summary
Systems and methods for generating and tracking shapes of a target may be provided. The method may include obtaining at least one first resolution image corresponding to at least one of a sequence of time frames of a medical scan. The method may include determining, according to a predictive model, one or more shape parameters regarding a shape of a target from the at least one first resolution image. The method may include determining, based on the one or more shape parameters and a shape model, at least one shape of the target from the at least one first resolution image. The method may further include generating a second resolution visual representation of the target by rendering the determined shape of the target.