Adaptive Gaussian Filter for X-Ray CT Noise Reduction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional adaptive filters for reducing noise and streak components in X-ray CT measurements require complex parameter adjustments, which are not clinically efficient and often result in insufficient noise reduction.

Innovation Solution

A noise reduction processor that applies an adaptive Gaussian filter with filter parameters determined by noise variance, using a noise model to estimate variances before and after logarithmic conversion, and generates filters based on after-log variances to uniformly reduce noise and streak components in X-ray intensity data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional adaptive filters are used with individually adjusted parameters, then some noise reduction effect is achieved, but the filter parameters require complex adjustment and the noise reduction effect is insufficient

Engineering Contradiction:

Improvenoise reduction effectVSAvoidparameter adjustment complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The filter automatically determines its own parameters by calculating noise variance from the input data itself, without requiring external parameter adjustment. The standard deviation of the Gaussian filter is set to match the calculated noise standard deviation, enabling the system to self-adapt to different noise conditions in various images

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The filter parameters (standard deviation) are dynamically changed based on the calculated noise characteristics of the input data. Instead of using fixed or manually adjusted parameters, the system automatically adjusts the filter strength to match the actual noise level in each image region

Inventive Principle:

Principle #35Parameter changes

2Ease of operation

If fixed filters are used, then parameter adjustment is simple, but the noise reduction effect is insufficient for varying noise levels

Engineering Contradiction:

Improveparameter adjustment simplicityVSAvoidnoise reduction effect

Core Design Contradiction:

Ease of operationVSMeasurement precision

Solution Approach 1:

The filter transitions from a static fixed filter to a dynamic adaptive filter that automatically adjusts its parameters based on the input data characteristics. The noise variance is calculated for each image, and the filter standard deviation is dynamically set to match this calculated noise level, enabling adaptation to varying noise conditions without manual intervention

Inventive Principle:

Principle #15Dynamics

3Measurement precision

If X-ray dose is increased to improve image quality, then noise and streak components are reduced, but patient safety is compromised

Engineering Contradiction:

Improveimage qualityVSAvoidpatient radiation exposure

Core Design Contradiction:

Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The system converts the harmful noise and streak artifacts present in low-dose images into useful information for filter parameter determination. By calculating noise variance from the low-dose data itself, the system creates an adaptive filter that is specifically tailored to reduce the actual noise present, thereby improving image quality without requiring increased radiation exposure

Inventive Principle:

Principle #22Blessing in disguise (Convert harm into benefit)

Data Source

PatentEP2533197B1Parameter-less noise reduction filter

Publication Date: 2020.10.28 TOSHIBA MEDICAL SYST CORP

Figure 1
Figure 2
Figure 3

AI summary

A noise reduction method comprising the steps of a) deciding a noise variance of each measurement signal based on a predetermined noise model, b) generating a discrete kernel of a filter for each measurement signal, which kernel indicates a frequency response of the filter for each measurement signal, based on the noise variance to the P-th power, c) filtering each measurement signal by applying the corresponding discrete kernel.