Meta-Image Deepnet Pipeline for Tumor Detection in CT Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computer-aided methods for diagnosing tumors in CT images face challenges due to tumors with high moisture content having similar gray levels to normal muscle or nearby organs, leading to low diagnosis capacity and false diagnoses.
Innovation Solution
A meta-image-based deepnet pipeline is developed, incorporating expert knowledge to enhance tumor detection by transforming domain knowledge into a form processable by deepnets, using both deepnet-based and analytics-based approaches to generate meta-images that augment feature space and improve diagnosis capacity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If deep network technologies are used to process abdominal CT images, then image processing capability is improved, but diagnosis accuracy remains insufficient due to tumors having similar gray levels to normal organs
Solution Approach 1:
The patent introduces meta-images as an intermediary representation that bridges the gap between raw CT images and deep network processing. These meta-images are generated by transforming domain knowledge (expert rules about tumor characteristics, organ boundaries, and pathological patterns) into image format, serving as a mediator that enhances the input information for the deep network without requiring changes to the network architecture itself.
Solution Approach 2:
The patent transforms domain knowledge into meta-images by changing the parameter representation from raw pixel intensity values to knowledge-enriched features. This involves encoding expert knowledge about tumor gray level ranges, organ boundary constraints, and pathological patterns into the meta-image parameters, thereby changing the input parameter space to include diagnostically relevant information that was previously inaccessible to the deep network.
2Adaptability or versatility
If manual model creation of sub-nets and complicated fusion process are used to incorporate knowledge, then knowledge representation capability is improved, but system complexity increases
Solution Approach 1:
The patent merges the knowledge representation and image processing functions into a unified meta-image framework. Instead of creating separate sub-nets for different knowledge aspects and then fusing them, the patent combines all domain knowledge into the meta-image generation process, producing a single integrated representation that can be processed by a standard deep network architecture.
Solution Approach 2:
The patent creates meta-images as copies or transformations of the original CT images that embed domain knowledge. Rather than building complex sub-nets to represent knowledge, the patent generates simplified meta-image copies that capture essential diagnostic information, making the system easier to implement and maintain while preserving knowledge representation capabilities.
Data Source
AI summary
The present invention relates to a novel meta-image-based tumor detection deepnet pipeline to increase the diagnosis capacity by cooperating with experts' knowledge for accurate tumor recognition in medical images.