A method and application for enhancing weak signals in medical images

Abstract

The present invention discloses a method and application for enhancing weak signals in medical images, comprising performing a preprocessing operation of suppressing mean shift on an input medical image according to preprocessing parameters to obtain a preprocessed image; and performing a filtering operation on the preprocessed image using a Z-wave modulated wavelet filter function to obtain an enhanced medical image. By analyzing the statistical characteristics of the input medical image and combining the Z-wave modulated wavelet filter group proposed by the present invention, the present invention can effectively enhance the contrast between a small or extremely small part of pixels such as lesions or blood vessels in the image and the background, while retaining the detail information in the image to the maximum extent. The algorithm proposed by the present invention is particularly sensitive to weak signals and can detect extremely small lesion areas. The medical image processing method proposed by the present invention can assist radiologists in reading and diagnosing films, and can also be embedded in the structure of an artificial intelligence model as a feature extraction method, or used as an optimization target to guide image processing results.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence-assisted image processing technology, and in particular to a method for enhancing weak signals in medical images and its application. Background Art

[0002] In recent years, with the rapid development of artificial intelligence and image analysis and processing technologies, medical image processing using computers has become widely used in various disciplines and fields closely related to medical imaging. By leveraging technologies such as image feature enhancement, computer vision, and deep learning, medical image processing and analysis can effectively visualize relevant images, pathologies, and anatomical structures, assisting physicians in diagnosis and clinical operations, effectively reducing their workload and achieving accurate diagnosis and precise treatment.

[0003] Currently, deep learning-based medical image diagnosis methods require extensive labeling of image data, consuming significant manpower and resources. Furthermore, the accuracy and consistency of the labeled data cannot be guaranteed. Because deep learning model structures are often black boxes, image feature extraction and enhancement capabilities are poorly interpretable, making them unconvincing for medical diagnosis and reducing the reliability of deep learning-based medical image diagnosis. However, filtering images with highly interpretable filters, such as Gabor filters or edge detection filters, can achieve image feature extraction and enhancement. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and application for enhancing weak signals in medical images in order to solve the above problems.

[0005] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0006] The present invention comprises the following steps:

[0007] S1: Input a medical image and preprocessing parameters. Let f(x,y) represent the pixel value at any coordinate (x,y) in the image. The input preprocessing parameters include: Flip, which can be either Flip = 0 or Flip = 1, with Flip = 1 being the default; and Drift, which can be Drift ∈ [0, 3] and has a default of Drift = 1. The value of Flip depends on whether the target pixel to be enhanced is higher than the background. If so, Flip = 0; if lower, Flip = 1.

[0008] S2: Complete the image preprocessing operation for mean drift suppression according to the preprocessing parameters;

[0009] S3: Construct a wavelet filter group modulated by Z-waves at multiple scales, perform filtering operations on the preprocessed image, and obtain an image processing result of multi-scale feature fusion;

[0010] S4: Output processing results.

[0011] Furthermore, the specific process of image preprocessing is as follows:

[0012] S1.1: For the input medical image f(x, y), determine whether to flip it according to the flip parameter: if Flip=0, then do not process the input image, that is, let f flip (x,y)=f(x,y); if Flip=1, then traverse all pixels in the image and make the pixel value of each pixel equal to 255 minus the original pixel value, that is, let f flip (x,y)=255-f(x,y). The pixel value flipping operation is completed.

[0013] S1.2: image f processed in step S1.1 flip (x, y), get the grayscale frequency histogram of the image, use the mixed Gauss function to fit the histogram, the default is the bimodal Gauss function, and get the mean parameter {μ n}、Standard deviation parameter {σ n} and amplitude parameters {A n}. The default value is the maximum amplitude parameter A max , you can also take the amplitude parameter under other rules, set its index to j, and let the drift value Movement = μ j +Drift·σ j Traverse all the pixels in the image processed in step i) and set the pixel value of each pixel equal to the original pixel value minus the drift value. If the pixel value of a pixel is less than 0, then set the pixel value to 0, that is, set f move (x,y)=max(f flip (x,y)-Movement,0). f move (x,y) is the preprocessing result.

[0014] Furthermore, the steps of constructing a multi-scale Z-shaped modulation wavelet filter bank for filtering are as follows:

[0015] S3.1: The default mother wavelet is the Gauss function g σ,γ (x, y), where the default parameter γ = 1. The mother wavelet can also be other functions such as the multi-order derivatives of the Gauss function, Cauchy function, Shanno wavelet function and other probability distribution functions or wavelet functions as the mother wavelet. Its scale parameter is The default is a single scale feature, that is, For a certain scale σ, the wavelength parameter defaults to λ=2σ. According to the wavelength parameter, a periodic even function T1(x) is constructed, which satisfies: T1(x)=T1(x+λ), T1(x)=T1(-x), Construct periodic odd function based on T1(x) Satisfying T2(x)=T2(x+λ), and T2(x)=-T2(-x); the periodic function T1(x) can be a continuous or discontinuous waveform function that satisfies the conditions, and can be directly taken as a trigonometric function, that is, It can also be square wave, sawtooth wave, etc. The default is square wave function:

[0016]

[0017]

[0018] The Fourier transform is denoted as F(f), where f is the input function. The periodic functions T1(x) and T2(x) are transformed by Fourier transform to obtain F(T1(x)) and F(T2(x)). The two periodic functions are approximated by trigonometric series using the inverse Fourier transform. The default parameter K = 30, and the actual K value can be any positive integer:

[0019]

[0020]

[0021] Then order:

[0022]

[0023]

[0024] Get the trigonometric series approximate function Λ of T1(x) and T2(x) (1) , Λ (2) .

[0025] Step S3.2: Construct a Z-shaped modulated wavelet filter using the following Z-shaped wave modulated wavelet filtering formula:

[0026] G cos (x,y)=g σ,γ (x,y)Λ1(x')

[0027] G sin (x,y)=g σ,γ (x,y)Λ2(x')

[0028]

[0029] x'=xcosθ+ysinθ

[0030] y'=-xsinθ+ycosθ

[0031] The wavelet basis defined by the even periodic function T1(x) is called the even wavelet basis, and the wavelet basis defined by the odd function T2(x) is called the odd wavelet basis. The angle parameter is defined as (x, y) is the parameter coordinate in the two-dimensional convolution kernel. According to the odd wavelet basis formula, a total of n odd wavelet filters of size odd_size×odd_size are generated. According to the even wavelet basis formula, a total of n even wavelet filters of size even_size×even_size are generated. Thus, 2n Z-shaped modulated wavelet filters under a single scale σ are generated; all filters are normalized, and for the i-th filter, the parameters greater than 0 in the filter are summed to obtain S i , let all parameters in the filter be divided by S i , complete the normalization operation, when setting the filter parameters under a single scale σ, the scale parameter σ can be an odd number greater than 1, the default σ=5; parameter The default value is n=3; the filter size parameter satisfies even_size∈[σ,2σ+1], odd_size∈[σ,2σ+1], and the filter size parameter is an odd number. Different filter sizes and different scale parameters can be set for the dual wavelet filter and the odd wavelet filter according to the task objectives; during the normalization operation, a scaling function s(x,y) can be taken according to the specific task to perform linear or nonlinear scaling on the filter;

[0032] S3.3: Call the 2n Z-shaped modulated wavelet filters generated in step S3.2 to process the pre-processed image f obtained after step S2. move (x, y) is filtered and the 2n images are averaged to obtain the processed image f Z-Wavelet (x,y); obtained f Z-Wavelet (x,y) is the method in a single scale The final output result is obtained. Multiple scales can be used according to specific tasks. The image processing operations proposed in this patent are performed in sequence, and the image processing operations of different scales σ are performed. i The output of Perform weighted summation as the final output, that is, where ω i is the weighted value. The weighting is not limited to linear weighting, and any nonlinear weighting method can be used depending on the situation.

[0033] By analyzing the statistical characteristics of an image and combining it with a highly interpretable wavelet filter, the Z-shaped modulation proposed by the present invention utilizes the wavelet basis in the Fourier transform modulation wavelet filter to maximize the preservation of both edge information of weak signals in the image and smooth features within the edges. This allows for effective feature extraction and enhancement of medical images without extensive training. The medical image processing method is suitable for processing single-channel grayscale medical images with relatively few objects or single-channel images within multi-channel medical images, such as segmented liver CT plain scan images.

[0034] The beneficial effects of the present invention are:

[0035] The present invention is a method and application for enhancing weak signals in medical images. Compared with existing technologies, the present invention can effectively enhance abnormal points, tiny or extremely tiny lesions, and the like in images. For example, the contrast of lesions or blood vessels in a segmented liver image can be directly used to assist in clinical diagnosis or, after appropriate modifications, embedded in the shallow structure of a deep learning model to enhance the model's robustness and interpretability and reduce the model's training cost. The present invention can process the input image without any prior data annotation, and then use it as a target guide for the deep learning model, greatly reducing the deep learning model's reliance on large amounts of labeled data. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a schematic diagram of the medical image processing system modules provided by an embodiment of the present invention.

[0037] Figure 2 This is a flow chart of a medical image processing method provided by an embodiment of the present invention.

[0038] Figure 3 This is the lesion feature enhancement effect of the liver plain scan image provided by an embodiment of the present invention, wherein the first row is the original image, and the second row is the lesion feature enhancement effect image.

[0039] Figure 4 This is the enhancement effect of lesion features and vascular features on liver enhanced images provided by an embodiment of the present invention, wherein the first row is the original image, the second row is the lesion feature enhancement effect image, and the third row is the vascular feature enhancement effect image. DETAILED DESCRIPTION

[0040] The present invention will be further described below in conjunction with the accompanying drawings:

[0041] like Figure 1 As shown: the present invention comprises the following steps:

[0042] S1: Input a medical image and preprocessing parameters. Let f(x,y) represent the pixel value at any coordinate (x,y) in the image. The input preprocessing parameters include: Flip, which can be either Flip = 0 or Flip = 1, with Flip = 1 being the default; and Drift, which can be Drift ∈ [0, 3] and has a default of Drift = 1. The value of Flip depends on whether the target pixel to be enhanced is higher than the background. If so, Flip = 0; if lower, Flip = 1.

[0043] S2: Image preprocessing operation completed according to preprocessing parameters. The specific steps are as follows:

[0044] S2.1: For the input medical image f(x, y), determine whether to flip it according to the flip parameter: if Flip=0, then do not process the input image, that is, let f flip (x,y)=f(x,y); if Flip=1, then traverse all pixels in the image and make the pixel value equal to 255 minus the original pixel value, that is, f flip (x,y)=255-f(x,y). The pixel value flipping operation is completed.

[0045] S2.2: image f processed in step S2.1 flip (x, y), get the grayscale frequency histogram of the image, use the mixed Gauss function to fit the histogram, the default is the bimodal Gauss function, and get the mean parameter {μ n}、Standard deviation parameter {σ n} and amplitude parameters {A n}. The default value is the maximum amplitude parameter A max , you can also take the amplitude parameter under other rules, set its index to j, and let the drift value Movement = μ j +Drift·σ j Traverse all the pixels in the image processed in step S1.1, and set the pixel value of each pixel equal to the original pixel value minus the drift value. If the pixel value of a pixel is less than 0, then set the pixel value to 0, that is, set f move (x,y)=max(f flip (x,y)-Movement,0). f move (x,y) is the preprocessing result.

[0046] S3: Construct a multi-scale Z-shaped modulated wavelet filter group to filter the pre-processed image. The specific steps are as follows:

[0047] Step S3.1 takes the mother wavelet as Gauss function g by default. σ,γ (x, y), where the default parameter γ = 1. The mother wavelet can also be other functions such as the multi-order derivatives of the Gauss function, Cauchy function, Shanno wavelet function and other probability distribution functions or wavelet functions as the mother wavelet. Its scale parameter is The default is a single scale feature, that is, For a certain scale σ, the wavelength parameter defaults to λ=2σ. According to the wavelength parameter, a periodic even function T1(x) is constructed, which satisfies: T1(x)=T1(x+λ), T1(x)=T1(-x), Construct periodic odd function based on T1(x) Satisfying T2(x)=T2(x+λ), and T2(x)=-T2(-x); the periodic function T1(x) can be a continuous or discontinuous waveform function that satisfies the conditions, and can be directly taken as a trigonometric function, that is, It can also be square wave, sawtooth wave, etc. The default is square wave function:

[0048]

[0049]

[0050] The Fourier transform is denoted as F(f), where f is the input function. The periodic functions T1(x) and T2(x) are transformed by Fourier transform to obtain F(T1(x)) and F(T2(x)). The two periodic functions are approximated by trigonometric series using the inverse Fourier transform. The default parameter K = 30, and the actual K value can be any positive integer:

[0051]

[0052]

[0053] Then order:

[0054]

[0055]

[0056] Get the trigonometric series approximate function Λ of T1(x) and T2(x) (1) , Λ (2) .

[0057] Step S3.2: Construct a Z-shaped modulated wavelet filter using the following Z-shaped wave modulated wavelet filtering formula:

[0058] G cos (x,y)=g σ,γ (x,y)Λ1(x')

[0059] G sin (x,y)=g σ,γ (x,y)Λ2(x')

[0060]

[0061] x'=xcosθ+ysinθ

[0062] y'=-xsinθ+ycosθ

[0063] The wavelet basis defined by the even periodic function T1(x) is called the even wavelet basis, and the wavelet basis defined by the odd function T2(x) is called the odd wavelet basis. The angle parameter is defined as (x, y) is the parameter coordinate in the two-dimensional convolution kernel. According to the odd wavelet basis formula, a total of n odd wavelet filters of size odd_size×odd_size are generated. According to the even wavelet basis formula, a total of n even wavelet filters of size even_size×even_size are generated. Thus, 2n Z-shaped modulated wavelet filters under a single scale σ are generated; all filters are normalized, and for the i-th filter, the parameters greater than 0 in the filter are summed to obtain S i , let all parameters in the filter be divided by S i , complete the normalization operation, when setting the filter parameters under a single scale σ, the scale parameter σ can be an odd number greater than 1, the default σ=5; parameter The default value is n=3; the filter size parameter satisfies even_size∈[σ,2σ+1], odd_size∈[σ,2σ+1], and the filter size parameter is an odd number. Different filter sizes and different scale parameters can be set for the dual wavelet filter and the odd wavelet filter according to the task objectives; during the normalization operation, a scaling function s(x,y) can be taken according to the specific task to perform linear or nonlinear scaling on the filter;

[0064] Step S3.3: Call the 2n Z-shaped modulated wavelet filters generated in step S3.2 to process the pre-processed image f obtained after step S2. move (x, y) is filtered and the 2n images are averaged to obtain the processed image f Z-Wavelet (x,y); obtained f Z-Wavelet (x,y) is the method in a single scale The final output result is obtained. Multiple scales can be used according to specific tasks. The image processing operations proposed in this patent are performed in sequence, and the image processing operations of different scales σ are performed. i The output of Perform weighted summation as the final output, that is, where ω i is the weighted value. The weighting is not limited to linear weighting, and any nonlinear weighting method can be used depending on the situation.

[0065] The wavelet can be any form of wavelet, such as Gaussian, Laplacian, Cauchy, etc.

[0066] like Figure 2 The flow chart of the medical image lesion processing method provided by the present invention is shown in the figure.

[0067] In order to demonstrate the feature enhancement effect of the present invention on the lesion in the liver plain scan image, further combined with Figure 3 The following three examples are used to illustrate this. Figure 3 As shown in the figure, the first row shows the liver segmentation images, and the second row shows the feature enhancement effect of the present invention on the lesions in the liver segmentation images. It can be seen that the medical image processing method proposed in the present invention can effectively enhance the features of small lesions in liver plain scan images based on liver segmentation.

[0068] In order to demonstrate the feature enhancement effect of the present invention on lesions and blood vessels in liver enhanced images, further combined with Figure 4 The following three examples are used to illustrate this. Figure 4 As shown in the figure, the first row shows an enhanced liver image, the second row shows the effect of the present invention on enhancing the features of lesions in the enhanced liver image, and the third row shows the effect of the present invention on enhancing the features of blood vessels in the enhanced liver image. It can be seen that the medical image processing method proposed in the present invention can effectively enhance the features of lesions and blood vessels in the enhanced liver image.

[0069] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware portion can be implemented using dedicated logic; the software portion can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those skilled in the art will appreciate that the above-mentioned devices and methods can be implemented using computer-executable instructions and / or contained in processor control code, for example, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, or can be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0070] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for enhancing weak signals in medical images, characterized in that: The following steps are involved: S1: Input medical images and preprocessing parameters; S2: Image mean drift suppression preprocessing; S2.1: For the medical image f(x,y) input in step S1, i.e., the image pixel value with coordinates (x,y), determine whether to flip it according to the flip parameter: if Flip=0, then the input image is not processed, i.e., let f flip (x,y)=f(x,y); if Flip=1, then traverse all pixels in the image and make the pixel value of each pixel equal to 255 minus the original pixel value, that is, let f flip (x,y)=255-f(x,y), complete the pixel value flip operation; S2.2: image f processed in step S2.1 flip (x, y), get the grayscale frequency histogram of the image, use the mixed Gauss function to fit the histogram, the default is the bimodal Gauss function, and get the mean parameter {μ n }、Standard deviation parameter {σ n } and amplitude parameters {A n }; The default value is the maximum amplitude parameter A max , you can also take the amplitude parameter under other rules, set its index to j, and let the drift value Movement = μ j +Drift·σ j ; Then traverse all the pixels in the image processed in step S2.1, and set the pixel value of each pixel equal to the original pixel value minus the drift value. If the pixel value of a pixel is less than 0, then set the pixel value to 0, that is, set f move (x,y)=max(f flip (x,y)-Movement,0); S3: Construct a wavelet filter group modulated by Z-waves at multiple scales, perform filtering operations on the preprocessed image, and obtain an image processing result of multi-scale feature fusion; S4: Output processing results.

2. The method for enhancing weak signals in medical images according to claim 1, wherein: The preprocessing parameters input in step S1 include: a flip parameter Flip, which takes the value Flip=0 or Flip=1, and the default value is Flip=1; a drift parameter Drift, which takes the value Drift∈[0,3], and the default value is Drift=1; wherein the value of the flip parameter Flip depends on whether the target pixel value to be enhanced is higher than the background part. If it is higher than the background part, then Flip=0 is taken, and if it is lower than the background part, then Flip=1 is taken.

3. The method for enhancing weak signals in medical images according to claim 1, wherein: In step S3, the steps of constructing a Z-wave modulated wavelet filter bank for filtering are as follows: Step S3.1: The default mother wavelet is the Gauss function g σ,γ (x, y), where the default parameter γ = 1, the mother wavelet can also be other functions such as the multi-order derivatives of the Gauss function, Cauchy function, Shanno wavelet function, other probability distribution functions or wavelet functions as the mother wavelet; Its scale parameter is The default is a single scale feature, that is, For a certain scale σ, the wavelength parameter defaults to λ=2σ. According to the wavelength parameter, a periodic even function T1(x) is constructed, which satisfies: T1(x)=T1(x+λ), T1(x)=T1(-x), Construct periodic odd function based on T1(x) Satisfying T2(x)=T2(x+λ), and T2(x)=-T2(-x); the periodic function T1(x) can be a continuous or discontinuous waveform function that satisfies the conditions, and can be directly taken as a trigonometric function, that is, It can also be a square wave or sawtooth wave; the default is a square wave function: The Fourier transform is denoted as F(f), where f is the input function. The periodic functions T1(x) and T2(x) are transformed by Fourier transform to obtain F(T1(x)) and F(T2(x)). The trigonometric series of the periodic function output by the inverse Fourier transform is approximately expressed in the form of K = 30 by default, and the actual K value is any positive integer: Again Get the trigonometric series approximate function Λ of T1(x) and T2(x) (1) , Λ (2) ; Step S3.2: Construct a Z-shaped modulated wavelet filter using the following Z-shaped wave modulated wavelet filtering formula: G even (x,y)=g σ,γ (x,y)Λ1(x') G odd (x,y)=g σ,γ (x,y)Λ2(x') x'=xcosθ+ysinθ y'=-xsinθ+ycosθ Among them, the wavelet basis defined by the even periodic function T1(x) is called the even wavelet basis, and the wavelet basis defined by the odd function T2(x) is called the odd wavelet basis; the angle parameter is defined as (x, y) is the parameter coordinate in the two-dimensional convolution kernel. According to the odd wavelet basis formula, a total of n odd wavelet filters of size odd_size×odd_size are generated. According to the even wavelet basis formula, a total of n even wavelet filters of size even_size×even_size are generated. Thus, 2n Z-shaped modulated wavelet filters under a single scale σ are generated; all filters are normalized, and for the i-th filter, the parameters greater than 0 in the filter are summed to obtain S i , let all parameters in the filter be divided by S i , complete the normalization operation, when setting the filter parameters under a single scale σ, the scale parameter σ can be an odd number greater than 1, the default σ=5; parameter The default value is n=3; the filter size parameter satisfies even_size∈[σ,2σ+1], odd_size∈[σ,2σ+1], and the filter size parameter is an odd number. Different filter sizes and different scale parameters can be set for the dual wavelet filter and the odd wavelet filter according to the task objectives; during the normalization operation, a scaling function s(x,y) can be taken according to the specific task to perform linear or nonlinear scaling on the filter; Step S3.1: Call the 2n Z-shaped modulated wavelet filters generated in step S3.2 to process the pre-processed image f obtained after step S2. move (x, y) is filtered and the 2n images are averaged to obtain the processed image f Z-Wavelet (x,y); obtained f Z-Wavelet (x,y) is the value in a single scale The final output result can be obtained by taking multiple scales according to the specific task. Perform image processing operations in sequence, and perform image processing on different scales σ i The output of Perform weighted summation as the final output, that is, where ω i is the weighted value. The weighting is not limited to linear weighting, and any nonlinear weighting method can be used depending on the situation.

4. An application of the medical image weak signal enhancement processing method according to claim 1, characterized in that: Used to process single-channel grayscale medical images with fewer objects in the image or single-channel images in multi-channel medical images.

Images