3D Image Region Detection via Multi-Planar Slice Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for interpreting 3D medical images struggle to provide both accurate classification and human-interpretable visual explanations, particularly in detecting differences between image cohorts.
Innovation Solution
A device and method for training a machine learning model that generates stacks of 2D slices from 3D images with varying orientations, allowing for the detection of regions of interest through slice-level prediction scores and subsequent tomographic reconstruction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If segmentation models are trained to provide visual explanations, then human interpretability is improved, but labelling efforts and device complexity increase significantly
Solution Approach 1:
The patent segments the 3D medical image into multiple 2D slices with different orientations (axial, sagittal, coronal planes). This segmentation allows the machine learning model to process and provide explanations for specific 2D slices independently, reducing the overall complexity compared to segmenting and labeling entire 3D volumes. The slice-level predictions can be aggregated to provide comprehensive 3D explanations.
Solution Approach 2:
The patent transforms the problem from 3D image classification to 2D slice classification by extracting multiple 2D slices from the 3D volume at different orientations. This dimensional reduction allows the use of well-established 2D convolutional neural networks while still capturing 3D information through multi-planar viewing. The slice-level predictions naturally provide spatial localization explanations.
2Area of stationary object
If 3D medical images are processed directly, then comprehensive visualization is achieved, but detection of differences between images becomes difficult
Solution Approach 1:
The patent divides the 3D medical image into multiple 2D slices across different orientations (axial, sagittal, coronal planes). This segmentation makes it easier to detect and compare differences between images, as each 2D slice can be independently processed and compared. The multi-planar approach allows radiologists and algorithms to identify abnormalities more easily than in single-plane 2D images.
Solution Approach 2:
The patent converts 3D volumetric data into multiple 2D representations at different orientations. This transformation enables the use of well-established 2D image processing and machine learning techniques to detect differences and patterns that may be obscured in 3D volume rendering. The multi-view approach provides comprehensive visualization while facilitating difference detection.
3Measurement precision
If slice-level predictions are generated from multiple orientations, then region of interest detection accuracy is improved, but processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary extraction of 2D slices from the 3D image before feeding them to the machine learning model. By pre-processing the image into standardized 2D slices at common orientations (axial, sagittal, coronal), the system optimizes the input format for the neural network, improving detection accuracy while controlling processing time through efficient slice extraction algorithms.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention relates to a device and a method for training a machine learning model for detecting the presence of at least one region of interest in a 3D image of a subject. The present invention further relates to a device and a method for detecting and reconstructing at least one region of interest in a 3D image of a subject using a trained machine learning model obtained with the method and device for training.