CT image ring artifact removal method based on stripe classification

Abstract

The invention discloses a CT image ring artifact removal method based on stripe classification, and belongs to the field of CT image reconstruction. According to the method, aiming at annular artifacts caused by inconsistent detector response, mechanical errors and ray scattering, the artifacts are efficiently removed through a single / multi-pixel stripe classification processing mechanism. The method comprises the following steps of: firstly, analyzing and positioning a single-pixel stripe based on a one-dimensional median filtering residual error, identifying an artifact type by adopting double-threshold segmentation, and eliminating through two-dimensional median filtering; secondly, constructing a sinogram column direction accumulation sum curve, positioning multi-pixel stripes by combining multiphase decomposition and high-frequency filtering, and repairing abnormal data by adopting linear interpolation; and finally, residual artifacts are removed through secondary filtering. According to the method, classification processing and a two-party method collaborative architecture are creatively fused, the limitation that a traditional method is poor in type adaptability and high in parameter sensitivity is broken through, the method has the advantages of being high in parameter robustness, low in algorithm weight and capable of maintaining the structure, and the CT image quality and the quantitative analysis precision can be improved in the fields of medical diagnosis, industrial detection and the like.

Description

Technical Field

[0001] The present invention relates to the field of CT image reconstruction, and a method for removing CT image annular artifacts based on stripe classification. Background Art

[0002] X-ray computed tomography (CT) technology is widely used in medical diagnosis, industrial inspection, and materials science because it can provide high-resolution three-dimensional structural information. However, affected by multiple factors such as inconsistent detector responses, mechanical assembly errors, ray scattering, and data acquisition noise, annular artifacts are common in CT images. Such artifacts appear as concentric circular interferences in the reconstructed images, seriously reducing the image contrast and quantitative analysis accuracy, and even leading to misdiagnosis or missed detection of industrial defects, becoming the core problem restricting high-precision imaging. Generally, annular artifacts are mainly divided into two categories: one is the annular artifact composed of a single stripe, and the main reasons for the generation of such annular artifacts are detector dead pixels, inconsistent responses between detector grid elements, and discontinuities between detectors. The other is the banded annular artifact composed of multiple pixels, and its generation reason is usually the discontinuity between different pixel set blocks of the detector.

[0003] Existing technologies are either limited by computational costs and model construction, or the imaging hardware is not compatible with the annular artifact removal algorithm, and deep learning methods still have shortcomings in generalization and real-time performance. Therefore, it is of great significance to develop an efficient, lightweight, and robust annular artifact removal algorithm to break through the hardware dependence of traditional methods and the data bottleneck of deep learning. Summary of the Invention

[0004] The present invention proposes a method for removing CT image annular artifacts based on stripe classification for different X-ray CT scenarios and data. By studying the pixel width of the stripes, we divide the bar artifacts into two types: single-pixel stripes and multi-pixel stripes. They require different identification and removal methods.

[0005] In the present invention, we select two improved methods to process the two types of stripes respectively: (1) For the identification and removal of single-pixel stripes, we choose to use threshold segmentation to identify the position of the artifacts, and then apply median filtering at the corresponding position of the artifacts to remove the artifacts and noise; (2) For the identification and removal of multi-pixel stripes, in the present invention, we choose to use a method combining polynomial decomposition and curve and high-frequency filtering to identify the position of local extrema to determine the position of the artifacts, and then use the interpolation method to remove the bar artifacts. This algorithm can be used not only in the CT reconstructed images but also in the sinograms in the projection domain.

[0006] In order to achieve the purpose of removing annular artifacts in CT images without losing the original structure of the CT images, the present invention provides an annular artifact removal algorithm based on stripe classification.

[0007] CT Ring Artifact Correction Method, the method comprising the following steps:

[0008] Step 1: Locate the ring artifacts of single-pixel stripes. Locate the single-pixel stripes in the sinogram by means of one-dimensional median filtering and thresholding;

[0009] Step 2: Process the identified single-pixel stripes. Remove the single-pixel stripes located in Step 1 by two-dimensional median filtering;

[0010] Step 3: Sinogram cumulative sum curve. Accumulate the sinogram along columns to obtain the pixel sum curve;

[0011] Step 4: Select an appropriate number of polyphase decomposition levels L and locate the positions of multi-pixel stripes. Perform polyphase decomposition on the sum curve obtained in Step 3 to obtain sub-curves, then perform high-frequency filtering on the sub-curves to obtain local extrema, and calculate the positions of the multi-pixel stripes through the positions of the local extrema;

[0012] Step 5: Remove the multi-pixel stripes. Remove the multi-pixel stripes by using an interpolation algorithm according to the positions of the multi-pixel stripes in Step 4;

[0013] Step 6: Remove the remaining stripe artifacts in the image. After Steps 1-5, due to improper selection of polyphase decomposition parameters, the stripe artifacts may not be completely removed, but the remaining stripe artifacts are all single-pixel stripes, so Steps 1-2 are executed again to remove the remaining single-pixel stripes.

[0014] Advantages of the above one or more technical solutions:

[0015] Compared with the existing algorithms, the advantages of the present invention are as follows:

[0016] (1) Classification processing and precise adaptation; based on the continuous number distribution of pixels, the present invention classifies bar artifacts into two categories: single-pixel stripes and multi-pixel stripes, and designs processing flows respectively according to their characteristics, significantly improving the accuracy of artifact removal.

[0017] (2) Strong parameter robustness and wide applicability; through adaptive threshold design and normalization preprocessing, the present invention makes the parameter selection have strong robustness, is insensitive to data fluctuations within a certain range, can adapt to different imaging scenarios, and reduces the cost of manual parameter adjustment.

[0018] (3) Lightweight and easy implementation; compared with the high computing power requirements and complex post-processing methods of deep learning models, the present invention is optimized based on a classical filtering and interpolation framework, without complex network training or high-dimensional transformation calculations. The algorithm structure is simple, the computing efficiency is high, and it is easy to be integrated into a real-time imaging system or a low-resource hardware platform. Through an innovative architecture of classification processing and dual-method collaboration, the present invention solves the pain points in the prior art such as poor adaptability to artifact types, high computational complexity, and strong parameter sensitivity. While ensuring high efficiency and ease of use, it significantly improves the accuracy and robustness of removing circular artifacts, providing a more universal solution for optimizing the quality of CT imaging. Brief Description of the Drawings

[0019] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0020] Figure 1 It is the overall algorithm flowchart of the present invention.

[0021] Figure 2 It is the result of the present invention in processing different circular artifacts under simulated data.

Claims

1. A method for removing circular artifacts from CT images based on stripe classification, characterized in that, Including the following steps: S1. Locate single-pixel stripe artifacts in the sinogram, and identify the positions of single-pixel stripes through one-dimensional median filtering and threshold segmentation; S2. Apply two-dimensional median filtering to the single-pixel stripes for removal; S3. Generate a sinogram cumulative sum curve, and accumulate the sinogram along columns to obtain a pixel sum curve; S4. Perform polyphase decomposition on the pixel sum curve, decompose the sum curve into multiple sub-curves, perform high-frequency filtering on the sub-curves, and locate the positions of multi-pixel stripes; S5. Based on the positions of the multi-pixel stripes, use an interpolation algorithm for removal; S6. Re-execute S1 and S2 on the remaining single-pixel stripes to complete the final removal of annular artifacts.

2. The method according to claim 1, characterized in that, Step 1 specifically includes: processing the original sine pattern through one-dimensional horizontal median filtering and calculating the difference function Combined with a high threshold and a low threshold to screen the positions of enhanced single-pixel fringes and weakened single-pixel fringes.

3. The method according to claim 1, wherein In S2, the window width of the two-dimensional median filtering is adapted to stripes with single-pixel or double-pixel widths.

4. The method according to claim 1, wherein In S4: The number of layers L of polyphase decomposition is dynamically determined according to the ratio of the artifact width to the number of columns of the sinogram; after decomposition, local maxima or minima of each sub-curve are extracted through high-frequency filtering to achieve multi-pixel stripe positioning.

5. The method according to claim 4, wherein The selection of the number of multi-phase decomposition layers L is based on the ratio of the artifact width to the number of columns in the sinogram, and the decomposition expressions are y1(s), y2(s), …, y L (s), where k = 1, 2, …, L.

6. The method according to claim 1, wherein The interpolation algorithm in S5 adopts linear interpolation, and the interpolation data is selected from the projection values of normal pixels in y k (j).

7. The method according to claim 1, wherein The secondary removal of the remaining single-pixel stripes in S6 includes: re-performing threshold segmentation and median filtering on the residual stripes after polyphase decomposition.

8. The method according to claim 1, characterized in that, The method is applicable to the sinogram in the projection domain or the reconstructed CT image domain.

9. The method according to claim 1, characterized in that, The thresholds Th and Tl are adaptively adjusted based on the standard deviation of single-pixel stripes.

10. The method according to claim 1, wherein The generation of the sinogram cumulative sum curve in step 3 includes pre-normalizing the projection data to reduce the influence of inconsistent detector responses.

Images