Generative Neural Network Lung Function Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current lung imaging techniques, such as CT scans, require multiple volume scans to estimate regional lung function, which is impractical for patients with advanced COPD and increases radiation exposure, and existing methods are computationally costly and difficult to scale.
Innovation Solution
A neural network-based generative modeling system that can directly convert a single CT scan to local measures of lung function, such as Jacobian and PRM, eliminating the need for multiple scans and reducing radiation exposure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple volume CT scans are used to estimate regional lung function, then measurement precision is improved, but radiation exposure increases
Solution Approach 1:
The patent creates a synthetic copy of the expiratory CT scan by translating it to represent an inspiratory volume. This synthetic inspiratory image allows computation of lung function measures (such as regional volume change and ventilation) without requiring a actual inspiratory scan, thereby eliminating additional radiation exposure while preserving measurement capability
Solution Approach 2:
The method performs preliminary image translation of the expiratory scan to inspiratory volume before any lung function analysis is conducted. This preliminary transformation enables subsequent computation of regional lung function measures from a single scan, avoiding the need for multiple scans and their associated radiation exposure
2Reliability
If multiple volume CT scans are performed, then reliability of lung function estimation is improved, but loss of time increases
Solution Approach 1:
By creating a synthetic inspiratory image from the expiratory scan through neural network-based translation, the method enables reliable lung function estimation without requiring the patient to undergo multiple scan acquisitions at different respiratory volumes, thus eliminating the time loss associated with multiple scans
3Measurement precision
If traditional image registration methods are used to compute lung function measures, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent replaces the mechanical image registration process with a neural network-based image translation approach. Instead of performing complex deformable registration between inspiratory and expiratory scans to compute Jacobian determinants and ventilation measures, the method uses a trained neural network to directly translate the expiratory scan to an inspiratory-equivalent image, from which lung function measures can be computed using simpler intensity-based methods
Solution Approach 2:
The synthetic inspiratory image created by the neural network serves as a substitute for the actual inspiratory scan, eliminating the need for complex registration operations. The translation process inherently captures the volumetric transformation, allowing direct computation of lung function measures without explicit registration steps
Data Source
AI summary
A machine implemented method includes acquiring at a processor a first image of a patent by performing a computed tomography (CT) scan of a patient and applying at the processor a generative neural network model to the first image of the patient to generate a second image of the patient. The method may further include performing, at the processor, an analysis using the first image and the second image. The first image may be a first pulmonary image and the second image may be a second pulmonary image. The first pulmonary image may show lungs at a first volume and the second pulmonary image may show the lungs at a second volume, the first volume different from the second volume. The first pulmonary image may be a scan of the lungs at expiration and the second pulmonary image may be an inspiratory image of the lungs.