Neural Network for CT Projection Data Denoising

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

X-ray systems face a trade-off between resolution and image quality due to the limitations of focal spot size, where larger focal spots improve signal-to-noise ratio but reduce spatial resolution, and smaller focal spots enhance resolution but lower signal-to-noise ratio.

Innovation Solution

The system employs a neural network trained on data from both large and small focal spot sizes to generate images with high spatial resolution and signal-to-noise ratio, using a combination of convolutional neural networks and residual networks to denoise and enhance projection images.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If the focal spot size of the X-ray tube is increased, then the X-ray flux and image quality (SNR) are improved, but the spatial resolution is reduced

Engineering Contradiction:
Improveimage quality (SNR)VSAvoidspatial resolution
Core Design Contradiction:
ReliabilityVSMeasurement precision

Solution Approach 1:

The system performs preliminary action by acquiring projection data using both large and small focal spot sizes before image reconstruction. The neural network is trained in advance on paired data from both focal spot sizes, enabling it to learn the transformation from large-spot (high SNR) to small-spot (high resolution) characteristics. This preliminary data acquisition and model training resolves the contradiction by preparing the necessary information beforehand to achieve both high SNR and high resolution in the final image.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The neural network acts as an intermediary between the large focal spot data (providing high SNR) and the desired small focal spot image quality (providing high resolution). The network learns the mapping relationship between data from different focal spot sizes and transforms the large-spot projection data into images with small-spot characteristics, effectively mediating the trade-off between SNR and spatial resolution.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If the focal spot size of the X-ray tube is decreased, then the spatial resolution is improved, but the X-ray flux and image quality (SNR) are reduced

Engineering Contradiction:
Improvespatial resolutionVSAvoidimage quality (SNR)
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The system merges the advantages of both large and small focal spot sizes by acquiring data from both configurations and combining them through neural network processing. The large focal spot provides high SNR data while the small focal spot provides high resolution reference data. The neural network merges these complementary characteristics to generate images that simultaneously achieve both high SNR and high spatial resolution, resolving the contradiction between these two opposing requirements.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The system changes the focal spot size parameter during data acquisition, using both large and small focal spots to collect projection data. By varying this critical parameter and processing the combined data through a trained neural network, the system achieves image quality that corresponds to small focal spot resolution while maintaining the high SNR characteristics of large focal spot data, thus resolving the trade-off.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3764325B1System and program for avoiding focal spot blurring
Publication Date: 2024.05.15 CANON MEDICAL SYST CORP
  • EP3764325B1 patent drawingFigure 1A~1B
  • EP3764325B1 patent drawingFigure 2
  • EP3764325B1 patent drawingFigure 3A~3C

AI summary

A system according to an embodiment includes circuitry. The circuitry inputs, to a first trained model third projection data to generate fourth projection data, receives the third projection data, the third projection data obtained from a CT (computed tomography) scan, and reconstructs a first image based on the fourth projection data. The first trained model has been trained using first projection data as an input and second projection data or first subtraction data between the second projection data and the first projection data as an output, the first projection data being acquired using an X-ray source having a first focal spot size, the second projection data being acquired using an X-ray source having a second focal spot size that is smaller than the first focal spot size, and the third projection data being acquired using an X-ray source having a third focal spot size that is larger than the second focal spot size.