Adaptive Image Segmentation Model for Nested Tissues
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image segmentation methods, such as cascaded anisotropic convolutional neural networks (ACNN), are limited in their generality and applicability for segmenting nested human tissues, requiring manual redesign for each tissue type, which hampers their practicality and effectiveness for images other than brain tumors.
Innovation Solution
An image segmentation method that trains an initial model using sample images to determine the number of image segmentation modules based on the types of pixels, allowing for automatic adaptation to different human tissue images, enabling segmentation of various nested tissues without manual intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a cascaded ACNN is used for brain tumor segmentation, then segmentation accuracy for nested tissues is improved, but the model requires manual redesign for each tissue type, reducing generality and applicability
Solution Approach 1:
The patent transforms the specialized brain tumor segmentation model into a universal image segmentation model that can automatically adapt to different nested tissue types. The model uses automatic module determination based on pixel type analysis and generalized training procedures, enabling it to handle various nested tissues (liver lesions, lung nodules, etc.) without manual redesign, thus achieving both high segmentation accuracy and broad applicability
2Measurement precision
If segmentation models are manually redesigned for different tissue types, then segmentation performance for specific tissues is optimized, but the time and complexity required for model development increases
Solution Approach 1:
The patent implements a self-service mechanism where the segmentation model automatically determines the number of segmentation modules based on the pixel type information from training images. The system autonomously performs parameter determination, module configuration, and model training without requiring manual intervention for each new tissue type, thereby maintaining high segmentation performance while dramatically reducing model development time
Solution Approach 2:
The patent performs preliminary analysis of pixel types and automatically determines the optimal number of segmentation modules before actual model training. This preliminary configuration step enables the model to be quickly adapted to new tissue types by simply providing training images, eliminating the need for time-consuming manual model redesign and accelerating the overall development process
3Device complexity
If the number of segmentation modules is fixed, then model structure is simplified, but the model cannot adapt to different numbers of tissue types in various images
Solution Approach 1:
The patent introduces dynamic adaptability into the model structure by automatically determining the number of segmentation modules based on the specific characteristics of each image's pixel types. The model can dynamically adjust its architecture from one to multiple segmentation modules depending on the complexity of the nested tissues in the input images, achieving both structural simplicity and high adaptability without requiring manual configuration
Data Source
AI summary
An image segmentation method is provided for a computer device. The method includes obtaining a plurality of sample images, calling an initial model to input the plurality of sample images into the initial model and to train the initial model based on the plurality of sample images to obtain an image segmentation model and, based on the initial model, determining a number of image segmentation modules according to a number of types of pixels of the plurality of sample images. Different image segmentation modules are used for segmenting different regions of an image. The method also includes calling the image segmentation model in response to obtaining a first image to be segmented, and segmenting the first image by using the image segmentation model based on a plurality of image segmentation modules, to output a second image. Computer device and non-transitory computer-readable storage medium counterparts are also contemplated.