Surgical image acquisition method and system based on image enhancement

Abstract

The present application relates to the technical field of image processing, more particularly, the present application relates to a kind of surgical image acquisition method and system based on image enhancement, comprising: obtaining the original image obtained by CT scanning, and the original image is preprocessed to obtain surgical image image;In the local neighborhood window of any pixel point in the surgical image image, the absolute value of the difference between the gradient amplitude of each pixel point and the average value of the gradient amplitude in the window is multiplied by the exponential function of the absolute value of the difference between the gray value of each pixel point and the gray value of the center pixel point in the window.The present application constructs the jump potential by analyzing the asymmetry of pixel neighborhood gradient, and calculates the structure confidence by tracking the potential change along the edge direction, so as to accurately distinguish the real anatomical edge from noise artifact.Furthermore, the structure confidence is used to adaptively adjust the Gaussian scale, while maintaining the contrast enhancement in the flat area, the Gaussian kernel is forced to shrink at the edge, which effectively eliminates the halo and improves the detail clarity.

Description

Technical Field

[0001] This invention relates to the field of image processing technology. More specifically, this invention relates to a surgical image acquisition method and system based on image enhancement. Background Technology

[0002] CT (Computed Tomography) surgical images are medical images created by using X-ray beams to scan the area of ​​the body being examined. Detectors receive the X-rays passing through the slice and convert them into digital signals for reconstruction. In clinical surgery, CT images can display the internal anatomical structures of the human body in high-resolution three-dimensional form, serving as crucial information for surgeons in preoperative path planning, intraoperative lesion localization, and risk avoidance. However, to minimize radiation exposure, modern surgeries typically follow a low-dose principle, resulting in CT images that often suffer from high quantum noise and low contrast between soft tissues (such as blood vessels and muscles). Especially in complex interventional procedures, blurred image details can easily lead to misjudgments of tissue boundaries. Therefore, effective image enhancement processing of CT surgical images to improve visual quality and the recognizability of key features is of significant clinical importance.

[0003] Currently, the Retinex algorithm is widely used in medical image enhancement due to its ability to effectively improve image contrast. Based on color constancy theory, this algorithm estimates the illuminance component through convolution operations, thereby separating the reflectance component. However, when processing CT images with these characteristics, the traditional Retinex algorithm typically uses a globally fixed Gaussian wraparound function scale. This approach has significant drawbacks: because it cannot adaptively adjust the scale according to local image features, the algorithm easily misinterprets abrupt structural grayscale changes as slow illumination changes when processing strong edge transition areas, leading to inaccurate illuminance component estimation and consequently producing noticeable halo artifacts around the edges of the enhanced image. Furthermore, CT images often contain metallic or hardening artifacts, which also possess high gradient characteristics. Traditional algorithms struggle to effectively distinguish between real anatomical structure edges and noise spectra, easily preserving or even enhancing artifacts, severely interfering with the doctor's vision and judgment. Summary of the Invention

[0004] This invention provides a surgical image acquisition method and system based on image enhancement, aiming to solve the problem that the traditional Retinex algorithm in related technologies cannot adaptively adjust the scale according to the local features of the image. When processing strong edge transition areas, the algorithm is prone to misinterpreting structural gray-level jumps as slow illumination changes, resulting in inaccurate illuminance component estimation and thus producing obvious halo artifacts around the edges of the enhanced image.

[0005] In a first aspect, the present invention provides a surgical image acquisition method and system based on image enhancement, comprising: acquiring an original image obtained from a CT scan, and preprocessing the original image to obtain a surgical image; within a local neighborhood window of any pixel in the surgical image, multiplying the absolute value of the difference between the gradient magnitude of each pixel and the mean gradient magnitude within the window by an exponential function of the absolute value of the difference between the gray value of each pixel and the gray value of the pixel at the center of the window, and standardizing the product to obtain the gray-level transition potential energy of each pixel; and, starting from each pixel, extracting along the vertical direction of the local principal gradient of the surgical image. A tracking sequence containing multiple pixels is taken. The difference between the minimum and maximum gray-level transition potential energy of each pixel in the tracking sequence is used as the denominator, and the product of the gray-level transition potential energy of the center pixel of the sequence and the minimum value is used as the numerator to calculate the structural confidence of each pixel. Based on the structural confidence of each pixel, the difference between the maximum and minimum values ​​of the Gaussian scale parameter is adjusted to obtain the Gaussian scale parameter of each pixel. The Gaussian scale parameter and the structural confidence are negatively correlated. The Retinex algorithm based on the Gaussian scale parameter is then applied to the surgical image to obtain the enhanced surgical image. By constructing two local feature indicators, gray-level transition potential energy and structural confidence, it is possible to accurately distinguish between real anatomical structure edges and noise or artifacts. By calculating the structural confidence, the Gaussian scale parameter in the Retinex algorithm is adaptively adjusted. A large-scale parameter is used in flat areas to compress the dynamic range, and a small-scale parameter is forcibly shrunk in strong edge areas to block the halo path. This significantly enhances image contrast while effectively eliminating edge halo artifacts, improving the visual quality and detail discernibility of the surgical image.

[0006] Furthermore, the product standardization process includes dividing the product of the exponential functions by the sum of the squares of the gradient magnitudes of each pixel within the local neighborhood window. By dividing the potential energy calculation result by the sum of the squares of the gradient magnitudes within the local window, the gray-level transition potential energy is standardized.

[0007] Furthermore, constructing the structural confidence score also includes multiplying the calculated structural confidence score by the natural logarithm of the mean gray-level transition potential energy within the tracking sequence plus 1. Compared to methods that only consider the difference between minimum and maximum values, this improvement can further suppress texture or noise interference that, although possessing some structural continuity, is extremely weak, ensuring that the calculated structural confidence score can more significantly respond to high-contrast, realistic anatomical edges and improve the accuracy of adaptive parameter adjustment.

[0008] Further, performing the Retinex algorithm based on the Gaussian scale parameter on the surgical image includes: generating a Gaussian wrap function using the Gaussian scale parameter as the standard deviation; convolving the Gaussian wrap function with the surgical image to obtain the illumination component; subtracting the illumination component from the surgical image in the logarithmic domain to obtain the reflection component; and linearly stretching the reflection component to a preset grayscale range to obtain the enhanced surgical image. By generating a specific Gaussian wrap function for convolution and logarithmic domain subtraction, low-frequency components caused by illumination (or CT artifacts) are accurately estimated and removed. Compared with traditional methods, this step ensures that the reflection component (details determined by the properties of the tissue itself) retains rich texture information when stretched back to the grayscale space, while avoiding edge overshoot caused by incorrect illumination estimation.

[0009] Further, the original image obtained from the CT scan is acquired: the CT values ​​in the CT tomographic scan data are mapped to a grayscale space of 0-255 to obtain the original image. Non-local mean filtering is then used to smooth the original image, resulting in the surgical image. Preprocessing the original CT data using non-local mean filtering, compared to traditional linear filtering such as Gaussian filtering, can utilize repetitive structural information in the image to effectively remove quantum noise generated by low-dose CT scans while preserving subtle edges and texture structures to the maximum extent.

[0010] Furthermore, the tracking sequence containing multiple pixels is extracted, including: starting from the current pixel, performing a bidirectional search along the vertical direction of its local principal gradient; extracting a set number of pixels, the gray-level transition potential energy of the set number of pixels collectively constituting the tracking sequence. The bidirectional search strategy along the vertical direction of the local principal gradient (i.e., the edge extension direction) is used to extract the tracking sequence, utilizing the spatial continuity of anatomical structures. Compared to analyses based on isolated points or isotropic windows, this method can effectively identify and eliminate radially diverging metallic artifacts or isolated random noise, ensuring that only spatially continuous real tissue edges generate high confidence, thereby avoiding erroneous enhancement of artifacts.

[0011] Further, the original image is smoothed by employing nonlocal mean filtering to smooth the original image.

[0012] Furthermore, the Sobel operator is used to calculate the gradient magnitude of each pixel in the surgical image. Compared to simple difference operations, the Sobel operator combines Gaussian smoothing and differential calculation, providing noise resistance while calculating the gradient. It can more accurately capture edge direction and intensity information in the image, providing precise input data for calculating the gray-level transition potential energy.

[0013] Furthermore, the preset Gaussian scale parameter has a maximum value of 80 and a minimum value of 5. This limits the numerical range of the adaptive Gaussian scale (5 to 80), setting physical constraints for the algorithm. The upper limit of 80 ensures sufficient scale for dynamic range compression in flat regions (such as within soft tissue), while the lower limit of 5 ensures that the Gaussian kernel is sufficiently small at strong edges to avoid out-of-bounds smoothing. This explicit boundary control prevents algorithm failure or image distortion caused by overly large or small calculated scales.

[0014] In a second aspect, an image enhancement-based surgical image acquisition system is also provided, including a processor and a memory, the memory storing a computer program, the processor executing the computer program to implement the image enhancement-based surgical image acquisition method described in any of the above embodiments.

[0015] Beneficial Effects: Addressing the issue of halo artifacts and noise interference easily generated by strong edges in CT images, this method constructs abrupt potential energy by analyzing the asymmetry of pixel neighborhood gradients and tracks the potential energy changes along the edge direction to calculate structural confidence, thereby accurately distinguishing between true anatomical edges and noise artifacts. Furthermore, this confidence is used to adaptively adjust the Gaussian scale, maintaining contrast enhancement in flat areas while forcibly shrinking the Gaussian kernel at edges, effectively eliminating halos and improving detail clarity. Attached Figure Description

[0016] Figure 1 This is a schematic flowchart illustrating an image enhancement method according to an embodiment of the present invention; Figure 2 This is a schematic illustration of the original image according to an embodiment of the present invention; Figure 3 This is an illustration of an enhanced image based on an embodiment of the present invention. Detailed Implementation

[0017] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0018] like Figure 1 As shown, S101: Acquisition and digital preprocessing of surgical images.

[0019] In this embodiment, the patient's tomographic scan data is first acquired using a medical spiral CT scanner. Specifically, a high-voltage generator drives an X-ray tube to rotate and expose the lesion area around the patient, and a detector array receives the X-rays carrying attenuation information after penetrating the human body. The analog-to-digital converter in the data acquisition system (DAS) converts the analog electrical signal into a digital signal, and then the filtered back projection (FBP) algorithm or iterative reconstruction algorithm is used to reconstruct the original DICOM format tomographic image data as the original image.

[0020] After obtaining the original image (e.g.) Figure 2 The image shown is the original surgical image, which requires preprocessing. First, the window width and level are adjusted according to the surgical observation requirements (e.g., abdominal organs or bones), mapping the CT values ​​to a grayscale space of 0-255 to obtain the surgical image. ,in This represents the coordinate position of the pixel. Next, to suppress the interference of quantum noise generated by low-dose scanning on subsequent feature calculations, a nonlocal mean filtering algorithm is used to smooth the surgical image. This preprocessing step removes granular noise in flat areas while preserving the edge structure information of the image to the maximum extent, providing a high-quality data foundation for subsequent calculations of various local indices of the surgical image.

[0021] S102: Constructing grayscale transition potential energy.

[0022] In surgical images, there are significant density differences between bone, metal implants, and soft tissue, which manifest as abrupt jumps in grayscale values. Due to the partial volume effect in CT imaging, this jump is not an ideal vertical step, but rather a sloping transition of a certain width. In the traditional Retinex algorithm, these structurally strong edges are easily misinterpreted as illumination changes, leading to halo artifacts around the edges after enhancement. This unique structural jump characteristic causes extreme asymmetry in the gradient distribution within the local neighborhood: pixels on the edge slope have extremely large gradients, while those in the adjacent flat areas have extremely small gradients. To quantify this feature and distinguish between true edges and ordinary textures, it is necessary to construct a grayscale jump potential that reflects whether a pixel is in a strong edge transition zone.

[0023] Before constructing this metric, a region of size centered on each pixel in the surgical image is defined. Local neighborhood window Calculate the gradient magnitude of all pixels within the window and calculate the average gradient magnitude within the window. Based on the above analysis, construct the gray-scale transition potential energy. The formula is as follows: In the formula, Represents the coordinates of the surgical image The grayscale transition potential energy of a pixel; Represents a local neighborhood window The first in 1 pixel; Indicating the first image in the surgical imaging Gradient magnitude of each pixel; Representing the surgical image in coordinates The average gradient magnitude of all pixels within a window centered on the pixel at point 1; and These represent the first and second parts of the surgical image within the window. Each pixel and the center pixel grayscale value; To prevent constants with a denominator of zero, in this embodiment, The value of is 1.

[0024] As can be seen from the above formula, the numerator contains two core interactions: one is the gradient difference. The other item is the grayscale difference. When pixel When located in a region of strong gradient change at the edge of a bone or organ, the neighborhood window will inevitably cover both edge pixels with high gradients and flat pixels with low gradients. This asymmetrical distribution of gradients leads to gradient differences. Significantly increased. Simultaneously, the huge grayscale difference on both sides of the edge is non-linearly amplified through an exponential term, such as the grayscale of bones. With muscle grayscale This means that, for the actual edges of anatomical structures, It will output a very large value; however, for tiny fluctuations caused by noise, the gain of the exponential term is minimal, thus making... It remains at a low level. Therefore, this index can more sensitively capture structurally abrupt regions that are prone to halos than traditional gradient operators.

[0025] S103: Construct structural confidence.

[0026] Building upon the grayscale transition potential energy constructed in the aforementioned steps, further analysis of unique features is needed. CT images often contain metallic or hardening artifacts, which also exhibit high-gradient transition characteristics, leading to increased grayscale transition potential energy values. However, real anatomical structures (such as blood vessel walls and cortical bone) possess spatial continuity and closure, while artifacts often radiate outwards or are isolated. If the algorithm incorrectly protects the edges of artifact regions, it will exacerbate the artifacts' interference with the doctor's vision. Therefore, it is necessary to analyze the continuity and stability of the grayscale transition potential energy along the edge direction, constructing a structural confidence level capable of confirming whether the transition belongs to a real biological tissue structure.

[0027] Before constructing this metric, path tracing needs to be performed on the grayscale transition potential energy obtained in the above steps. Using the current pixel... Starting from the local principal gradient of the image, search and extract along the direction perpendicular to the edge extension direction. The grayscale transition potential energy values ​​of each pixel constitute a spatial tracking sequence. .

[0028] Based on this, the structural confidence score is constructed using the following formula: In the formula, Representing coordinates Structural confidence of the pixel; coordinates The grayscale transition potential energy of a pixel; Represents spatial tracking sequence The minimum value in; Represents spatial tracking sequence The maximum value in; Represents spatial tracking sequence The arithmetic mean.

[0029] The core of this formula lies in the denominator. That is, the range of the sequence. If the data features represent the true anatomical structure edges, then the potential index along the edge direction... It should remain relatively stable and at a high level. This means that the sequence fluctuates little, i.e. It tends to a smaller value, while the sequence mean The denominator becomes relatively small, while the numerator becomes relatively large, resulting in an explosive increase in the calculated structure confidence. Conversely, for isolated noise or divergent artifacts, the continuity along a specific direction is poor, and the spatial tracking sequence... Extreme minimum values ​​or drastic fluctuations can occur, causing the denominator to increase or the numerator to decrease, thus significantly reducing the structure confidence score. Through this step, the system can accurately identify the true edges that genuinely require special anti-halo treatment.

[0030] S104: Construct adaptive Gaussian scaling parameters.

[0031] Based on the steps described above, strong edge regions requiring special processing were accurately identified using grayscale transition potential energy and structural confidence. In the Retinex algorithm, halos are caused by an excessively large scale of the wraparound function when estimating the illumination component, leading to energy leakage from bright areas to dark areas. For CT surgical images, a larger scale is needed to compress the dynamic range in flat areas (such as liver parenchyma), while at bone / tissue boundaries, the scale of the Gaussian kernel must be forcibly contracted, limiting its effective range to one side of the edge, thereby cutting off the physical path of halo generation. Therefore, it is necessary to construct an adaptive Gaussian scale parameter that can automatically adjust the Gaussian kernel size based on structural confidence.

[0032] The preprocessing steps before constructing this index are: setting an upper limit on the Gaussian kernel size allowed by the algorithm. Gaussian kernel size limit For use in flat areas, such as The value is 80, and the lower limit of the Gaussian kernel size allowed by the algorithm is set. Used for strong edge regions, such as The value is 5; and the structure confidence score is calculated for the entire graph. maximum value Based on this, an adaptive Gaussian scaling parameter is constructed. The formula is as follows: In the formula, Representing coordinates The adaptive Gaussian scaling parameters for each pixel; and These are the preset maximum and minimum Gaussian scale constants, respectively; coordinates Structural confidence of the pixel; This represents the maximum structural confidence level for the entire image.

[0033] In the above formula, the more obvious the true anatomical edge features in the data, the higher the corresponding structural confidence. The larger the value, the better. In the formula, It is located in the added terms of the denominator. With As the value increases, the denominator increases rapidly, leading to changes in the calculated adaptive Gaussian scaling parameters. The value decreases nonlinearly and approaches the lower limit. Conversely, in flat areas, As the denominator approaches 0, the denominator approaches 0. This makes the result approach the upper limit. This parameter, compared to the globally fixed scale setting in existing technologies, achieves pixel-level scale adaptive control, and physically blocks the smoothing effect across edges.

[0034] S105: Retinex enhancement based on adaptive scale.

[0035] Based on the adaptive Gaussian scaling parameters obtained from the above steps, this embodiment improves the Retinex algorithm. For each pixel in the surgical image, a specific Gaussian wrapping function is generated using its corresponding adaptive Gaussian scaling parameters. This variable-scale Gaussian function is then convolved with the original image to obtain the accurate local illumination components. Due to the strong edges Automatic shrinkage, illuminance component It can closely follow changes at the image edges without going out of bounds, thus ensuring the accuracy of illumination estimation.

[0036] Obtain the illuminance component Then, perform the subtraction operation in the logarithmic field: The reflection component is obtained. , Represents the center pixel within a window in a surgical image. The grayscale value. Due to the accurate estimation of the illuminance component at the edges, the reflected component is calculated. At the junction of bone and soft tissue, the overshoot or undershoot phenomena common in traditional algorithms are eliminated, thus removing halos. Finally, the reflection component... Linear stretching and mapping back to the 0-255 grayscale space yields the final enhanced surgical image (e.g., Figure 3 As shown, the area within the red box is the region with the most significant enhancement (more detailed parts are visible compared to the unenhanced image). This image retains the contrast enhancement effect in flat areas while maintaining the sharpness of tissue edges, making it directly usable for subsequent surgical navigation and diagnosis.

[0037] The present invention also provides a surgical image acquisition system based on image enhancement. The system includes a processor and a memory, the memory storing computer program instructions. When the processor executes the computer program instructions, it implements the surgical image acquisition method based on image enhancement according to the first aspect of the present invention.

[0038] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and therefore will not be described in detail here.

[0039] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented using computer-readable / executable instructions stored or otherwise maintained on such a computer-readable medium.

[0040] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A surgical image acquisition method based on image enhancement, characterized in that, include: The raw images obtained from CT scans are acquired, and the raw images are preprocessed to obtain surgical images. In the local neighborhood window of any pixel in the surgical image, the absolute value of the difference between the gradient magnitude of each pixel and the mean gradient magnitude within the window is multiplied by an exponential function of the absolute value of the difference between the gray value of each pixel and the gray value of the center pixel of the window. The gray-level transition potential energy of each pixel is obtained by standardizing the product. Starting from each pixel, a tracking sequence containing multiple pixels is extracted along the vertical direction of the local principal gradient of the surgical image. The difference between the minimum and maximum gray-level jump potential energy of each pixel in the tracking sequence is used as the denominator, and the product of the gray-level jump potential energy of the center pixel of the sequence and the minimum value is used as the numerator. The structural confidence of each pixel is then calculated. The difference between the maximum and minimum values ​​of the Gaussian scale parameter is adjusted based on the structural confidence of each pixel to obtain the Gaussian scale parameter of each pixel. The Gaussian scale parameter and the structural confidence are negatively correlated. The Retinex algorithm based on the Gaussian scale parameter is then applied to the surgical image to obtain the enhanced surgical image.

2. The surgical image acquisition method based on image enhancement according to claim 1, characterized in that, The product standardization process includes: The product of the exponential functions is divided by the sum of the squares of the gradient magnitudes of each pixel within the local neighborhood window.

3. The surgical image acquisition method based on image enhancement according to claim 1, characterized in that, The construction of structural confidence also includes: the structural confidence is positively correlated with the mean gray-level jump potential energy within the tracking sequence.

4. The surgical image acquisition method based on image enhancement according to claim 1, characterized in that, Performing the Retinex algorithm based on the Gaussian scale parameter on the surgical image includes: The Gaussian scale parameter is used as the standard deviation to generate a Gaussian wrap function. The Gaussian wrap function is convolved with the surgical image to obtain the illumination component. The illumination component is subtracted from the surgical image in the logarithmic domain to obtain the reflection component. The reflection component is linearly stretched to a preset grayscale range to obtain the enhanced surgical image.

5. The surgical image acquisition method based on image enhancement according to claim 1, characterized in that, Obtain the raw images obtained from the CT scan, including: The CT values ​​in the CT tomographic scan data are mapped to a grayscale space of 0-255 to obtain the original image. The original image is then smoothed to obtain the surgical image.

6. The surgical image acquisition method based on image enhancement according to claim 1, characterized in that, Extract the tracking sequence containing multiple pixels, including: Starting from the current pixel, a bidirectional search is performed along the vertical direction of its local principal gradient; a set number of pixels are extracted, and the gray-level transition potential energy of the set number of pixels together constitutes the tracking sequence.

7. The surgical image acquisition method based on image enhancement according to claim 5, characterized in that, Smoothing the original image includes: smoothing the original image using non-local mean filtering.

8. The surgical image acquisition method based on image enhancement according to claim 1, characterized in that, include: The gradient magnitude of each pixel in the surgical image was calculated using the Sobel operator.

9. The surgical image acquisition method based on image enhancement according to claim 1, characterized in that, The preset Gaussian scale parameter has a maximum value of 80 and a minimum value of 5.

10. A surgical image acquisition system based on image enhancement, comprising a processor and a memory, characterized in that, The memory stores a computer program, and the processor executes the computer program to implement the surgical image acquisition method based on image enhancement as described in any one of claims 1-9.

Images