Machine Learning Network for Multiphase Collateral Image Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for diagnosing cerebrovascular diseases, such as stroke, face challenges in accurately and efficiently generating multiphase collateral images due to subjective interpretation and increased workload, with existing techniques like multiphase CT angiography and perfusion MRI having limitations in radiation exposure and objectivity.
Innovation Solution
A machine learning method that uses MRI images to generate multiphase collateral images by applying a brain mask and learning these images with a network, enabling rapid and objective generation of diagnostic images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiphase CT angiography is used to evaluate collateral flow, then blood vessel information can be obtained, but radiation exposure increases
Solution Approach 1:
The patent uses MRI images as a copy or alternative representation of blood vessel structures, eliminating the need for direct CT angiography imaging. The machine learning model learns to generate multiphase collateral images from MRI data, providing the necessary vascular information without exposing patients to ionizing radiation from CT scans.
Solution Approach 2:
The patent replaces the mechanical/radiological system of CT angiography with a magnetic resonance-based system. By substituting the physical imaging modality from X-ray based CT to MRI combined with machine learning, the harmful radiation effect is eliminated while maintaining the ability to evaluate collateral flow.
2Measurement precision
If conventional multiphase imaging methods are used, then collateral flow information can be obtained, but imaging time and contrast medium usage increase
Solution Approach 1:
The machine learning model creates a computational copy of the multiphase contrast enhancement process. Instead of physically acquiring multiple phases with contrast medium over time, the model learns the temporal evolution patterns from training data and generates synthetic multiphase images from a single MRI acquisition, dramatically reducing imaging time and contrast medium requirements.
Solution Approach 2:
The machine learning model is pre-trained on multiphase collateral images obtained through conventional methods. This preliminary training allows the model to capture the underlying patterns and dynamics of contrast flow, enabling it to subsequently generate accurate multiphase images from single-phase MRI data without requiring actual multiphase acquisition in clinical practice.
3Ease of manufacture
If experiential technique-based image reconstruction is used, then collateral images can be generated, but interpretation objectivity decreases
Solution Approach 1:
The machine learning model performs the image reconstruction and analysis task autonomously without requiring expert intervention. The model automatically processes MRI images, generates multiphase collateral images, and provides quantitative measurements, eliminating the subjective variability inherent in expert-based interpretation while maintaining ease of image generation.
Solution Approach 2:
The machine learning model is trained using feedback from ground truth multiphase collateral images obtained through conventional methods. During training, the model's predictions are compared against these reference images, and the model parameters are adjusted to minimize the difference. This feedback mechanism ensures the model learns accurate representations of collateral flow patterns, improving both objectivity and reliability.
Data Source
AI summary
A learning method for generating a multiphase collateral image comprises the steps of: receiving inputs of an MRI image of a head part and a multiphase collateral image generated on the basis of the MRI image; generating a brain mask by using the MRI image; generating an MRI image and a multiphase collateral image which are masked by the brain mask; and learning the masked multiphase collateral image for the masked MRI image by using a learning network.