AI Semantic Segmentation for Selective X-ray Visibility Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current medical imaging technologies face challenges in effectively enhancing or reducing the visibility of specific anatomical structures in X-ray images, often relying on global adjustments that affect all structures equally, rather than semantically relevant features, which can obscure important details like medical instruments or implants.
Innovation Solution
The method employs an artificial intelligence algorithm to generate partial images based on semantic classes (anatomical structures, medical implants, instruments, or fastening elements) and combines these with imaging data using adjustable functions and parameters, allowing for user-defined enhancements or reductions in visibility, particularly for structures like bones or surgical materials, to create an output image that prioritizes relevant information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Illumination intensity
If global scaling of image intensities is applied to enhance visibility of structures, then the overall contrast is improved, but all structures are affected equally which obscures specific important details
Solution Approach 1:
The patent segments the image into different semantic classes (anatomical structures, medical implants, instruments, etc.) and applies different visibility adjustments to each class. This allows specific structures to be enhanced or suppressed independently while maintaining overall image quality, resolving the contradiction between global contrast enhancement and specific structure detection.
Solution Approach 2:
The patent applies local quality by assigning different rendering parameters to different semantic classes within the image. Instead of uniform global scaling, each semantic class receives tailored visibility adjustments, enabling important details like medical instruments to stand out while maintaining appropriate context from anatomical structures.
2Illumination intensity
If frequency or bandpass filters are used to compress image dynamics and increase local contrast, then local contrast is improved, but image noise increases relative to remaining image content
Solution Approach 1:
The patent segments the image into semantic classes and applies selective visibility adjustments rather than using broad frequency filters. This targeted approach allows local contrast enhancement in specific regions of interest while preserving overall image information and reducing noise amplification that occurs with aggressive filtering.
Solution Approach 2:
The patent changes visibility parameters (brightness, contrast, opacity) for different semantic classes rather than applying global frequency filtering. This parameter-based approach enables local contrast enhancement without the noise amplification and information loss associated with traditional bandpass filtering methods.
3Difficulty of detecting and measuring
If semantic class-based visibility adjustment is applied to enhance specific structures, then visibility of relevant structures is improved, but device complexity increases
Solution Approach 1:
The patent uses pre-trained machine learning models that have been copied and applied to new images. The model weights and architectural framework are reused, and only the visibility parameters need to be adjusted for different semantic classes. This copying approach enables complex semantic understanding without proportionally increasing computational complexity for each new application.
Solution Approach 2:
The patent performs preliminary action by pre-training the machine learning model on annotated medical images before actual use. The semantic classification capability is built in advance, so during actual imaging procedures, the system only needs to apply pre-computed class labels and corresponding visibility parameters, significantly reducing real-time computational complexity.
Data Source
AI summary
A method for setting the visibility of objects in a projection image generated by radiation of an anatomical region includes loading imaging data representing a projection image generated by radiation into a memory of a computer, generating a first partial image from the imaging data using a first semantic class by an artificial intelligence device executing an artificial intelligence algorithm, combining the first partial image and the imaging data to generate an output image using a first adjustable function with at least one first parameter such that the first function determines a rendering of the first partial image in the output image, displaying an output image on a display device; and generating training data for training the artificial intelligence algorithm of the artificial intelligence device.