Slice Image Alignment for Accurate Multi-Slice Region Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image data detection methods face inefficiencies due to information asymmetry caused by differences in resolution and regions of interest resulting from varying medical appliances and scanning protocols, leading to low computational speed and detection accuracy.
Innovation Solution
An image data detection method utilizing a three-dimensional alignment and aggregation network to align and aggregate context information across slice images, incorporating a two-dimensional backbone network, three-dimensional alignment network, and target detection network to enhance feature extraction and detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If resampling is used to resolve information asymmetry in medical images, then detection accuracy is improved, but computational speed decreases
Solution Approach 1:
The medical image is divided into multiple slice images along the scanning direction. A slice image group is constructed by selecting the current slice and adjacent slices within a preset range. This segmentation allows processing of local regions with context information, improving detection accuracy without requiring resampling of the entire image, thus maintaining computational efficiency.
Solution Approach 2:
The patent introduces a third dimension (scanning direction) to the traditional 2D slice image processing. By constructing slice image groups that include adjacent slices, the method utilizes spatial context information across multiple dimensions, resolving information asymmetry without resorting to resampling operations that would reduce computational speed.
2Ease of operation
If conventional detection methods are used on slice images, then processing simplicity is maintained, but information asymmetry between different resolutions and regions persists
Solution Approach 1:
The patent introduces a slice image group as an intermediary structure between individual slice images and the final detection result. This intermediary incorporates context information from adjacent slices, allowing the detection model to access multi-resolution and multi-region information without complicating the overall processing pipeline. The group-based approach maintains operational simplicity while reducing information loss.
3Use of energy by moving object
If only single slice images are processed, then computational resources are conserved, but detection accuracy for lesions requiring context information is reduced
Solution Approach 1:
Instead of processing all adjacent slices indefinitely, the patent applies a preset range limit to the slice image group construction. This partial action approach includes enough context information from adjacent slices to improve detection accuracy for lesions requiring spatial context, while avoiding the excessive computational resource consumption that would result from processing unlimited slices. The preset range optimizes the balance between accuracy and resource usage.
Data Source
Figure 1~2
Figure 3~5
Figure 6
AI summary
The present application relates to an image data testing method and apparatus in the technical field of artificial intelligence. The method comprises: for each target image in a slice image sequence, determining a corresponding slice image group, wherein the slice image group comprises target images and neighboring images, which have a contextual relationship with the target images, in the slice image sequence; for each slice image in the slice image group, respectively extracting a corresponding slice feature image; performing an alignment operation on the correspondingly extracted slice feature images in the slice image group; aggregating context information of each slice image in the slice image group by utilizing the aligned feature images; and performing target region detection on the aggregated feature images to obtain detection results corresponding to the target images, and merging the detection results corresponding to each target image to generate a detection result corresponding to an image to be detected. By using the method, not only can the problem of information asymmetry of an image to be detected be solved, but the testing efficiency of image data can also be effectively improved.