Angiography Image Blending With Sensitivity-Specificity Model Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing angiographic imaging techniques require high doses of chemical contrast agents and x-ray radiation, posing health risks and trade-offs between sensitivity and specificity in image detection.
Innovation Solution
A method involving multiple machine learning models, each optimized for either sensitivity or specificity, with user-adjustable controls to combine their outputs, enhancing angiographic image analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the dose of chemical contrast agent and x-ray radiation is increased to improve image quality, then image quality is improved, but patient harm increases due to toxic side effects and radiation injury
Solution Approach 1:
The patent creates virtual copies of angiographic images through machine learning models that simulate what the images would look like at different contrast doses. The GAN-based system generates synthetic high-dose images from low-dose input images, allowing clinicians to evaluate image quality without actually administering high doses of contrast agents or radiation to patients
Solution Approach 2:
The patent replaces the physical mechanism of increasing contrast agent dosage and radiation exposure with a computational mechanism. Instead of mechanically injecting more contrast and delivering more radiation, the system uses deep learning algorithms to computationally enhance image quality, substituting a digital processing approach for a physical dosage approach
2Reliability
If machine learning models focus on optimizing sensitivity, then detection of relevant features improves, but detection of irrelevant features increases reducing specificity
Solution Approach 1:
The patent segments the machine learning task into two distinct models: one optimized for sensitivity (detecting all potential vessels including false positives) and another optimized for specificity (filtering out false positives). By dividing the detection task into separate specialized models, the system achieves both high sensitivity and high specificity that cannot be achieved by a single model
Solution Approach 2:
The patent introduces an intermediary component that combines the outputs of the sensitivity-optimized model and the specificity-optimized model. This intermediary integration mechanism reconciles the conflicting outputs from the two specialized models, allowing the system to leverage both high sensitivity detection and high specificity filtering capabilities
Data Source
AI summary
Systems, methods, and computer program products for generating and displaying angiography images from multiple machine learning models that provide different sensitivity and specificity performance in the segmentation of vascular structures in an angiogram. A graphical user interface with control elements provides user control over the mixture of the multiple machine learning models and other settings, including an a pseudo brightness control and a zoom control. The displayed angiogram image is based on the settings of the control elements, and adjusts in response to changes thereto. The pseudo brightness control controls the mixture of the high specificity model versus the high sensitivity model in the displayed image. The zoom control modifies the magnification of the displayed image and increases the proportional mixture of the high sensitivity model in the displayed image. A widget or other type of control element may also be provided for controlling the mixture of the empirical data versus machine learning model output in the displayed image.