AI-Based Arterial Input Function Estimation From Saturated MR Scans
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for quantifying myocardial blood flow using cardiovascular magnetic resonance (CMR) face challenges such as T1-saturation effects in the left ventricular cavity, low contrast-to-noise ratio, and the complexity and cost of dual-bolus or dual-sequence protocols, which are not universally available for clinical use.
Innovation Solution
A machine learning system trained on a large dataset of dual-bolus or dual-sequence scans predicts the arterial input function (AIF) from a single high-dose gadolinium bolus, allowing for accurate myocardial perfusion parameter estimation without the need for dual-bolus protocols and enabling conversion between different data types.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If dual-bolus protocol is used to overcome T1-saturation effects, then AIF accuracy is improved, but scan complexity and cost increase
Solution Approach 1:
The patent uses a machine learning model trained on dual-bolus data to create a virtual copy of the dual-bolus AIF measurement process. The model learns the relationship between single-bolus and dual-bolus measurements, then applies this learned mapping to predict accurate AIF from simpler single-bolus scans, eliminating the need for actual dual-bolus protocol execution while preserving measurement accuracy
Solution Approach 2:
The patent replaces the physical dual-bolus injection mechanism with a computational machine learning system. Instead of mechanically administering two separate contrast boluses with precise timing, the system uses an trained AI model to computationally derive the equivalent AIF information from a single bolus scan, substituting physical complexity with algorithmic processing
2Measurement precision
If dual-bolus protocol is used to maintain signal linearity, then AIF measurement accuracy is improved, but scan time and cost increase
Solution Approach 1:
The machine learning model creates a virtual representation of the dual-bolus measurement process by learning from training data. It copies the essential measurement relationships without requiring the actual time-consuming dual-bolus protocol execution, providing accurate AIF measurements from faster single-bolus scans
Solution Approach 2:
The patent performs preliminary training of the machine learning model using dual-bolus data before actual clinical scans. This preliminary action captures the complex relationships between contrast concentration and signal intensity, allowing the model to quickly predict accurate AIF during routine single-bolus scans without requiring the time-intensive dual-bolus protocol at the time of scanning
3Measurement precision
If low-dose contrast is used to avoid T1-saturation, then AIF accuracy is improved, but myocardial CNR deteriorates
Solution Approach 1:
The machine learning model learns the relationship between low-dose and high-dose contrast effects from training data. It copies the AIF measurement benefits of low-dose protocols while compensating for the myocardial enhancement deficiency through computational prediction, allowing accurate AIF derivation without being constrained by low contrast dosing
Solution Approach 2:
The patent introduces a machine learning model as an intermediary between the raw single-bolus signal data and the final AIF measurement. This intermediary processes the signal data, applying learned corrections and transformations that compensate for T1-saturation effects and low-dose limitations, producing accurate AIF measurements without requiring actual dual-bolus administration
4Ease of operation
If single-bolus protocol is used to simplify scanning, then ease of operation is improved, but AIF accuracy deteriorates due to T1-saturation
Solution Approach 1:
The patent replaces the complex mechanical dual-bolus injection system with a simpler single-bolus mechanical protocol combined with a computational correction system. The machine learning model compensates for the simplified protocol's limitations, allowing accurate AIF measurement with easier single-bolus administration
Solution Approach 2:
The machine learning model creates a computational copy of the accurate dual-bolus AIF measurement process. By learning from training data, it replicates the measurement accuracy of complex protocols while operating on simpler single-bolus scan data, bridging the gap between ease of operation and measurement precision
Data Source
AI summary
Examples of the present disclosure include a method and system for deriving a computer implemented trained artificial intelligence (AI) model that is capable of predicting arterial input function (AIF) from blood and myocardial signal intensity curves of a subject obtained using a magnetic resonance (MR) scanner during injection of a single high dosage bolus of contrast agent that would otherwise cause signal saturation. The method and system requires an input data set of a large number of prior obtained dual bolus or dual sequence sets of scan measurements to be used as training data for the AI model. Once the computer implemented trained AI model has been obtained, it is deployed in a further method and system which receives MR data for a particular subject for which myocardial perfusion parameters such as the AIF is to found, the MR data being obtained from an MR scan taken during injection of a single bolus of contrast agent at a concentration sufficient to give MR signal saturation. The trained AI model is then able to predict, from the single bolus saturated MR data, myocardial perfusion parameters such as the AIF for the subject. The trained AI model is also able to convert between dual bolus datasets and duals sequence datasets.