Dynamic Image Prediction for PET Radiation Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current PET examination methods require significant time for tracer binding and distribution, leading to delayed imaging that can increase radiation dose and limit diagnostic accuracy, especially when trying to differentiate between blood flow and disease-specific information.
Innovation Solution
A method and computing device that utilize dynamic image data to predict early and delay image data by extracting blood flow and disease-specific information, allowing for the generation of training data that normalizes image data based on brightness and predicts new image data using machine learning models, reducing the need for extensive radiation exposure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If PET images are obtained after a specific time (2-3 hours) when tracer binding reaches stable state, then diagnostic accuracy is improved, but radiation dose increases and examination time is extended
Solution Approach 1:
The patent applies preliminary action by capturing dynamic image data during the early phase (blood flow phase) before the tracer fully distributes. The machine learning model is trained in advance using both early dynamic data and delayed images to predict the delayed image characteristics from early data alone, eliminating the need for patients to wait 2-3 hours and reducing radiation exposure while maintaining diagnostic accuracy.
2Measurement precision
If PET images are obtained after a specific time (2-3 hours) when tracer binding reaches stable state, then disease-specific information is maximized, but examination time is extended
Solution Approach 1:
The system performs preliminary capture of dynamic image data during the blood flow phase (early phase), then uses a pre-trained machine learning model to predict the delayed image characteristics. This eliminates the need for patients to undergo prolonged 2-3 hour examinations while still obtaining disease-specific information, as the model has been trained in advance on paired early and delayed images.
Solution Approach 2:
The patent creates a predictive copy of the delayed image characteristics using machine learning. Instead of requiring actual delayed images to be captured, the system generates a predicted copy that replicates the disease-specific information found in delayed images, based on patterns learned from training data that correlates early dynamic phases with subsequent delayed phases.
3Speed
If dynamic image data is captured during blood flow phase, then early information is obtained, but blood flow influence obscures disease-specific information
Solution Approach 1:
The machine learning model serves as an intermediary that bridges the early dynamic phase and delayed phase information. It learns the complex relationship between blood flow patterns and subsequent disease-specific tracer distribution, then uses this learned relationship to extract and predict disease-specific information from early dynamic data that would otherwise be obscured by blood flow effects.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enhances diagnostic accuracy by reducing radiation dose and improving image quality, enabling earlier prediction of disease states with reduced tracer amounts and shorter image acquisition times.
Implementation Method 1
A positron emission computed tomography (PET) examination is a cutting-edge nuclear medicine imaging method for administering a positron-emitting radioisotope and obtaining radiation emitted outside the human body
Data Source
AI summary
The present invention relates to a method for predicting a state of an object on the basis of dynamic image data and a computing device performing same, the method enabling initial dynamic image data and delay image data to be predicted by performing learning on the basis of dynamic image data captured at a time point when both blood flow image information and disease-specific biological information are included, and furthermore, enabling blood flow image information and disease-specific biological information of the object to be provided.