3D Image Processing via 2D Slice Conversion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing 3D image processing techniques require significant computing resources and are challenging to train, especially when applying neural networks to 3D images, often resulting in high computational demands and reduced efficiency.
Innovation Solution
A system and method that process 3D images by obtaining slices of different orientations, applying conversion models to these slices, and reconstructing converted 3D images, which can be trained using 2D sample slices to reduce computational requirements and improve image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If a neural network is trained using 3D sample images, then 3D image processing quality is improved, but computing resource consumption increases significantly
Solution Approach 1:
The patent divides the 3D image processing task into multiple 2D slice processing tasks. Instead of processing the entire 3D volume at once, the system slices the 3D image into multiple 2D cross-sectional images and processes each slice independently using a 2D neural network. This segmentation approach maintains processing quality while significantly reducing computational complexity and resource consumption.
Solution Approach 2:
The patent creates 2D copies (slices) of the 3D image data and processes these copies using a 2D neural network model. By working with 2D representations rather than the full 3D structure, the system achieves comparable processing quality with reduced computational demands, as the 2D network can be applied to each slice independently and then reconstructed into the final 3D result.
2Reliability
If a neural network is trained using 3D sample images, then 3D image processing capability is improved, but training difficulty increases
Solution Approach 1:
The training process is segmented into 2D slice processing tasks rather than requiring the network to learn from complete 3D volumes. Each 2D slice can be processed and trained independently, simplifying the training algorithm and reducing the complexity of convergence. The 2D neural network requires fewer parameters and less complex optimization compared to a 3D network, making training more manageable while still achieving reliable 3D processing capability through slice reconstruction.
3Use of energy by moving object
If 2D image processing is applied to 3D images on a slice-by-slice basis, then computing resource consumption is reduced, but processing efficiency may be affected
Solution Approach 1:
The patent merges multiple processed 2D slices back into a single 3D image structure. After each 2D slice is processed independently by the neural network, the results are combined and reconstructed to form the complete processed 3D image. This merging approach allows the system to leverage efficient 2D processing while maintaining the integrity and utility of the full 3D data structure, achieving both resource efficiency and processing effectiveness.
Data Source
AI summary
The present disclosure provides systems and methods for processing a 3D image. A method may include for an orientation of the 3D image, obtaining, from the 3D image, a plurality of slices of the orientation; processing the plurality of slices of the orientation by applying a conversion model; and determining a converted 3D image of the orientation based on the plurality of processed slices of the orientation.