Synthesizing Missing CT Phases for Cancer Diagnosis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing cancer diagnosis methods face challenges in accurately classifying cancer subtypes using incomplete CT images, as they often lack one or more phases necessary for precise diagnosis.
Innovation Solution
A system and method that synthesize missing phases in CT images using a generative adversarial network (GAN), generating a full-phase CT image set, extracting lesion-level features, and predicting cancer subtypes based on these features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If CT images are repeated to acquire all four phases for accurate cancer diagnosis, then diagnostic accuracy is improved, but additional cost and radiation exposure increase
Solution Approach 1:
The patent uses a generative adversarial network (GAN) to synthesize missing CT phases by copying and transforming existing complete-phase images. The GAN generator creates virtual CT images of the missing phases based on learned patterns from complete datasets, allowing accurate cancer diagnosis without repeating actual CT scans and thereby reducing radiation exposure while maintaining diagnostic accuracy
2Measurement precision
If CT images are repeated to acquire all four phases for accurate cancer diagnosis, then diagnostic accuracy is improved, but additional cost increases
Solution Approach 1:
The patent applies GAN technology to generate synthetic CT images of missing phases by copying and transforming existing complete-phase images. This virtual synthesis approach eliminates the need for repeated physical CT scans, thereby reducing healthcare costs while maintaining the diagnostic accuracy required for accurate cancer subtype classification
3Measurement precision
If multi-modal medical images are acquired with all modalities for accurate cancer diagnosis, then diagnostic accuracy is improved, but image acquisition difficulty increases
Solution Approach 1:
The patent uses GANs to synthesize missing CT phases by copying patterns from complete-phase images and transforming them to match the required phase characteristics. This virtual generation approach simplifies the acquisition process by eliminating the need to coordinate multiple complex imaging modalities and procedures, while still achieving accurate cancer diagnosis through the synthesized complete-phase CT image sets
Data Source
AI summary
A cancer diagnosis system that performs cancer diagnosis from an incomplete set of CT images having at least one missing phase includes an input unit that receives the incomplete set of CT images, a full-phase CT image set generation unit that synthesizes CT images for the at least one missing phase to generate a full-phase CT image set, a lesion-level feature extraction unit that extracts a feature map and a segmentation map from the full-phase CT image set, and extracts lesion-level features from the feature map and the segmentation map, and a cancer subtype prediction unit that predicts a subtype of cancer based on the extracted lesion-level features. Therefore, it may be possible to synthesize CT images with missing phases, and perform accurate classification of the pathological subtype of the tumor in consideration of the synthesized CT images.