X-Ray Object Enhancement Using Temporal Background Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Medical images, particularly X-ray fluoroscopic images, suffer from low contrast and noise due to limitations in imaging speed and radiation dosage, making it difficult to clearly visualize thin tubular structures such as blood vessels, catheters, and stents.
Innovation Solution
An apparatus utilizing machine learning models to enhance X-ray images by detecting objects, determining a background layer, and generating an output image that improves noise level, clarity, and contrast through residual image averaging and alignment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If radiation dosage is increased to improve image quality, then image contrast and clarity improve, but patient exposure to harmful radiation increases
Solution Approach 1:
The system performs preliminary actions by capturing multiple frames before the current frame and using machine learning models to predict and reconstruct the current frame. This allows image enhancement through temporal averaging and background subtraction without requiring increased radiation dosage in the current frame, thus improving image quality while maintaining safe radiation levels.
Solution Approach 2:
The system creates multiple copies of the imaging process by capturing multiple frames over time. These temporal copies are then processed through machine learning models to reconstruct an enhanced current frame. This approach allows the system to achieve high image quality equivalent to higher radiation dosage without actually increasing the radiation exposure.
2Speed
If imaging speed is increased to capture real-time motion, then temporal resolution improves, but image contrast and signal-to-noise ratio deteriorate
Solution Approach 1:
The system performs preliminary capture of multiple frames at the required high imaging speed to maintain temporal resolution. Then, through machine learning-based temporal averaging and background subtraction, it enhances the contrast and signal-to-noise ratio of the reconstructed frame without requiring slower imaging speeds, thus resolving the contradiction between speed and image quality.
3Object-affected harmful factors
If radiation dosage is reduced to minimize patient exposure, then harmful radiation exposure decreases, but image noise increases and contrast decreases
Solution Approach 1:
The system captures multiple frames at reduced radiation dosage in advance and uses machine learning models to reconstruct an enhanced current frame. The temporal averaging and background subtraction processes reduce noise and improve contrast in the reconstructed image, allowing safe radiation levels to be maintained while achieving high image quality.
Solution Approach 2:
The system creates multiple temporal copies at low radiation dosage and combines them through machine learning processing. This allows the noise and contrast issues of individual low-dosage frames to be overcome by the collective information from multiple copies, achieving high image quality without high radiation exposure.
4Measurement precision
If temporal averaging is performed over more frames to reduce noise, then noise level decreases, but motion artifacts and misalignment increase
Solution Approach 1:
The system performs preliminary detection of object locations, orientations, and deformations in each frame before temporal averaging. This allows the machine learning model to align and register frames based on actual object motion, reducing motion artifacts and misalignment while still benefiting from noise reduction through temporal averaging of properly registered frames.
Data Source
AI summary
In some embodiments, a method for enhancing objects in an X-ray video may include receiving an image frame in an X-ray video and detecting, using a first machine learning model, one or more objects in the image frame, wherein the detection is performed based on the image frame, a sequence of image frames preceding the image frame in the X-ray video, and data indicating one or more objects in the sequence of image frames. The method may further include determining, using a second machine learning model, a background image layer, based on the image frame and one or more image frames from the sequence of image frames that precedes the image frame in the X-ray video. The method may further generate an output image containing an enhanced view of the one or more objects in the image frame based in part on the background image layer.