Discriminator for Infarction Region Discrimination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
It is challenging to accurately discriminate infarction regions in CT images, especially when a large amount of data indicating the correct infarction region is difficult to prepare, which can lead to delayed thrombolytic therapy and increased bleeding risks.
Innovation Solution
A deep learning method that trains a discriminator using a common learning unit and multiple neural networks to differentiate between CT images with and without contrast medium, as well as between different imaging modalities like CT and MR images, to accurately extract and quantify infarction regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning is applied to automatically extract infarction regions, then discrimination accuracy is improved, but the requirement for large amounts of labeled training data increases
Solution Approach 1:
The patent introduces an intermediary model (second neural network) that generates synthetic training data by translating images from a first modality to a second modality. This intermediary approach allows the system to create artificial labeled datasets without requiring actual annotated medical images, thereby resolving the contradiction between achieving high discrimination accuracy and the scarcity of labeled training data.
Solution Approach 2:
The patent creates copies of existing medical images through image translation between different modalities. By generating synthetic images that mimic the appearance and characteristics of target modality images, the system expands the available training data without needing to collect and annotate additional real medical images, thus addressing the data quantity requirement for deep learning.
2Adaptability or versatility
If multiple neural networks are used to handle different image modalities, then adaptability is improved, but device complexity increases
Solution Approach 1:
The patent implements a universal image translation framework that can handle multiple image modalities (CT, MRI, PET, etc.) through a single second neural network. This network is trained to translate between any pair of modalities, making the system versatile without requiring separate specialized networks for each modality combination, thus reducing overall system complexity while maintaining adaptability.
Solution Approach 2:
The patent employs dynamic routing mechanisms where the system can flexibly switch between different processing paths depending on the input modality and desired output. The neural network architecture dynamically adjusts its behavior based on the specific translation task required, allowing the system to handle diverse modalities efficiently without being overwhelmed by static complexity.
Data Source
AI summary
A discriminator includes a common learning unit and a plurality of learning units that are connected to an output unit of the common learning unit. The discriminator is trained, using a plurality of data sets of a first image obtained by capturing an image of a subject that has developed a disease and an image data of a disease region in the first image, such that information indicating the disease region is output from a first learning unit in a case in which the first image is input to the common learning unit. In addition, the discriminator is trained, using a plurality of data sets of an image set obtained by registration between the first image and a second image whose type is different from the type of the first image, such that an estimated image of the second image is output from an output unit of a second learning unit.