Image Analysis Using Multiple Local Models for Feature Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for identifying anatomical features in images, especially when anatomical features are in a diseased state, face challenges due to variance in images from different patients and the limitations of statistical models trained with a small number of images, leading to inaccurate feature identification.
Innovation Solution
A computer-implemented method that fits multiple models to an image, where each part of the region of interest is modeled by multiple second models, allowing for a more accurate fit and using weighted averages based on the distance from a model's center to combine data from multiple models for feature identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single statistical model is used to identify anatomical features in images, then the model fitting process is simple and fast, but the identification accuracy is insufficient especially when anatomical features are in a diseased state
Solution Approach 1:
The patent divides a single statistical model into multiple local statistical models, where each model focuses on a specific local region of the anatomical structure. This segmentation allows each model to capture detailed local variations more accurately, particularly in diseased states, while the overall system remains manageable through modular organization of these local models.
Solution Approach 2:
The patent combines multiple local statistical models to form a comprehensive modeling system. By merging the results from multiple local models that each model specific regions, the system achieves higher overall accuracy in feature identification while maintaining computational efficiency through the distributed nature of the modeling approach.
2Reliability
If a statistical model is trained with a small number of images, then the training process is fast and resource-efficient, but the model fails to capture true variation and identifies features inaccurately
Solution Approach 1:
By segmenting the training process into multiple local models trained on different subsets of the limited training images, the system can better utilize the available data. Each local model focuses on specific regions and variations, allowing the system to capture true anatomical variation more effectively even with few training images overall.
Solution Approach 2:
The patent applies local quality by training each local statistical model with specific characteristics tailored to its region. This allows different parts of the anatomical structure to be modeled with appropriate detail and variation capture, improving overall model robustness without requiring proportional increases in training data quantity.
3Measurement precision
If multiple models are fitted to an image with overlapping regions, then the feature identification becomes more accurate through redundancy, but the computational complexity and processing time increase
Solution Approach 1:
The patent segments the image into multiple local regions, each modeled by a dedicated local statistical model. This segmentation enables parallel processing of different regions, reducing overall computation time compared to a single comprehensive model, while the overlapping regions provide redundant information for improved accuracy through combination of multiple model results.
Solution Approach 2:
The patent applies partial action by fitting multiple local models to different regions of the image, where each model processes only its specific region rather than the entire image. This partial processing approach reduces the computational burden on each individual model while the aggregation of multiple partial results achieves comprehensive and accurate feature identification.
Data Source
Figure 1~1A
Figure 2
Figure 3
AI summary
A computer-implemented method for identifying features in an image. The method comprises fitting a plurality of second models to the image, the plurality of second models together modelling a region of interest, wherein each part of the region of interest is modelled by at least two of the plurality of second models; and identifying the features in the image based upon the fit of the plurality of second models.