Synthetic X-Ray View Generation for Lower-Dose Diagnosis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing medical imaging workflows often require additional X-ray views, such as lateral views, which are not always available, leading to organizational overhead and negative patient experience due to the need for rescheduling examinations.
Innovation Solution
A machine learning-based approach generates synthetic X-ray images from acquired images, allowing radiologists to assess pathologies in different views without requiring additional scans, using a pre-trained encoder-decoder architecture and discriminator model to enhance the generation of realistic additional views.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If additional X-ray views are acquired to facilitate diagnosis, then diagnostic accuracy is improved, but patient dose and organizational overhead increase
Solution Approach 1:
The patent generates synthetic X-ray images as copies of actual X-ray views using machine learning models. These synthetic images replicate the appearance and diagnostic information of real X-ray projections without requiring additional physical exposures, thereby providing alternative views while avoiding additional patient radiation dose
Solution Approach 2:
The patent replaces the mechanical/physical process of acquiring additional X-ray views with a computational process. Instead of physically repositioning the patient or X-ray source to obtain different projections, the system uses trained neural networks to synthesize alternative views from existing images, substituting physical measurement with information processing
2Measurement precision
If additional X-ray views are scheduled for patients, then diagnostic capability is improved, but patient experience and workflow efficiency deteriorate
Solution Approach 1:
The patent performs the action of generating additional views in advance, immediately after acquiring the initial X-ray images. By synthesizing alternative projections computationally rather than scheduling separate examination sessions, the system makes diagnostic information available without delaying the examination workflow or requiring additional patient appointments
3Productivity
If synthetic X-ray images are generated using machine learning, then additional views are obtained without additional scans, but computational complexity increases
Solution Approach 1:
The patent performs the complex computational task of training machine learning models in advance, before actual image generation is needed. The pre-trained models can then rapidly generate synthetic views during clinical use, shifting the computational burden from the real-time operation to a preliminary training phase
Solution Approach 2:
The patent uses trained machine learning models to copy the transformation patterns learned from paired X-ray images. Once the model learns the mapping between different views during training, it can efficiently generate synthetic images by applying this learned transformation, reducing the computational complexity of real-time image generation
Data Source
Figure 1
Figure 2~3
Figure 4~5
AI summary
The present invention relates to X-ray imaging. In order to improve X-ray imaging workflow, an image processing apparatus (10) is proposed that comprises an input (12), a processor (14), and an output (16). The input (12) is configured to receive a first X-ray image obtained in an image acquisition. The first X-ray image has a first view of a body part of a patient. The processor (14) is configured to generate, based on the received first X-ray image, a second X-ray image having a second view of the body part of the patient using a pre-trained machine-learning model. The second view is different from the first view. The output (16) is configured to output the generated second X-ray image.