Intelligent Cancer Radiation Therapy Image Processing System

Abstract

The present invention discloses an intelligent tumor radiotherapy image processing system, including an image acquisition module, an image correction module, a grayscale optimization module, a grayscale suppression module, a tumor radiotherapy image segmentation module, and a tumor radiotherapy image processing module. The present invention belongs to the field of image processing, specifically to an intelligent tumor radiotherapy image processing system. This method introduces a dynamic region-specific correction coefficient to ensure that the grayscale of normal lung tissue does not exceed the tolerance range. A tumor border gradient is introduced to enhance grayscale stretching in the border area, and non-border area protection is performed based on the dynamic correction coefficient to avoid grayscale distortion. Detail reconstruction is performed by designing an ideal tumor grayscale weight, forming a dual mechanism of enhancement and suppression to avoid detail loss in high grayscale areas. Multi-layer precision segmentation is achieved through two-dimensional modulation at the pixel and feature levels. A uniformity-constrained loss and a self-adjusting weighted Tower loss are developed to constrain the segmentation results of amplitude modulation and feature modulation, further improving the image processing effectiveness of tumor radiotherapy.

Description

[Technical Field]

[0001] The present invention relates to the field of image processing, and more particularly to an intelligent tumor radiation therapy image processing system. [Background technology]

[0002] Tumor radiotherapy image processing systems are specialized systems based on medical imaging technology, performing processes such as artifact removal, grayscale calibration, and image registration on multimodal images collected during the radiation therapy process, including CT, MRI, and PET. However, conventional tumor radiotherapy image processing systems suffer from problems such as confused boundaries between tumors and normal tissues, grayscale distortion, and an inadequate balance between detail and interference, resulting in unreliable processing results. Conventional tumor radiotherapy image processing systems also suffer from poor environment adaptability, insufficient hierarchical adaptability of the radiation therapy target region, severe loss of detail, and insufficient segmentation accuracy of critical regions, resulting in poor tumor radiotherapy image processing results. Summary of the Invention

[0003] To overcome the drawbacks of the prior art, the present invention provides an intelligent tumor radiotherapy imaging system. To address the problems inherent in conventional tumor radiotherapy imaging systems, such as the confusion between the tumor and normal tissue boundaries, grayscale distortion, and an inadequate balance between detail and interference, resulting in unreliable processing results, the present method introduces dynamic regional correction coefficients to perform nonlinear correction on tumor radiotherapy images, ensuring that tumor regions are smooth and bright while ensuring that the grayscale of normal lung tissue does not exceed an acceptable range. A tumor border gradient is introduced to enhance grayscale stretching in the border region, addressing the edge blur caused by tumor grayscale crowding. Non-border regions are protected based on dynamic correction coefficients to avoid grayscale distortion. Detail reconstruction is performed by designing ideal tumor grayscale weights, forming a dual mechanism of enhancement and suppression to avoid detail loss in high-grayscale regions and further improving the reliability of tumor radiotherapy imaging. In response to the problems of conventional tumor radiotherapy image processing systems, such as poor environmental adaptability, insufficient hierarchical adaptability of the radiotherapy target region, serious loss of details, insufficient segmentation accuracy of important regions, and poor tumor radiotherapy image processing effectiveness, this method uses two-dimensional modulation at the pixel and feature levels to adapt to the hierarchical needs of the target region and achieve multi-layer precise segmentation. Hierarchical feature optimization is used to avoid the loss of details of the tumor edge during feature transfer. A uniformity-constrained loss and a self-adjusting weighted tower loss are developed to constrain the segmentation results of amplitude modulation and feature modulation, dynamically identifying segmentation error regions. The hierarchical loss ensures the segmentation consistency of the multi-layer structure of the target region, further improving the image processing effectiveness of tumor radiotherapy.

[0004] The technical means adopted by the present invention are as follows: The present invention provides an intelligent tumor radiotherapy image processing system, including an image acquisition module, an image correction module, a grayscale optimization module, a grayscale suppression module, a tumor radiotherapy image segmentation module, and a tumor radiotherapy image processing module, the image acquisition module acquires historical tumor radiation treatment images to obtain an original radiation treatment image set; The image correction module normalizes the original radiation therapy image, dynamically generates grayscale correction coefficients by region, and performs global grayscale correction; The grayscale optimization module extracts grayscale channels from the corrected original radiation therapy image, defines an adaptive grayscale conversion function incorporating edge gradients, generates dynamic grayscale correction coefficients, and obtains a grayscale optimized image; the grayscale suppression module calculates an ideal tumor grayscale weight for the grayscale optimized image to obtain a grayscale suppressed radiation treatment image; The tumor radiotherapy image segmentation module designs a segmentation network based on U-Net++, performs pixel-level weak amplitude modulation and strong amplitude modulation on the radiotherapy image obtained by the grayscale suppression module, and performs feature modulation on the decoding layer at the characteristic level, constructs a self-adjusting weight tower loss, and segments the tumor radiotherapy image; The tumor radiotherapy image processing module provides an intelligent tumor radiotherapy image processing system that performs image processing on real-time tumor radiotherapy images.

[0005] Furthermore, the image acquisition module acquires historical tumor radiotherapy images, performs pre-processing, and then performs image annotation, finally obtaining a set of original radiotherapy images.

[0006] Furthermore, the image correction module normalizes the original radiation therapy image, introduces the grayscale average values ​​of two regions, calculates the normalized grayscale average values ​​of the two regions respectively, dynamically generates grayscale correction coefficients by region, and finally performs global grayscale correction.

[0007] Furthermore, the grayscale optimization module extracts a grayscale channel from the corrected image, calculates the probability density function of the grayscale levels, clips high-frequency interference, constructs a weighted histogram, defines an adaptive grayscale conversion function, introduces gradient information of the tumor margin, generates a dynamic grayscale correction coefficient, and obtains an optimized radiation therapy image.

[0008] Furthermore, the grayscale suppression module calculates the ideal tumor grayscale weight, constructs Laplacian pyramids for the corrected image and the optimized image respectively, generates Laplacian detail maps, fuses the detail maps of each layer according to the ideal tumor grayscale weight, and obtains the radiation therapy image after grayscale suppression through reconstruction.

[0009] Furthermore, the tumor radiation therapy image segmentation module specifically includes: Based on U-Net++, we design a split network including an input layer, an encoding layer, a decoding layer, and an output layer. The output of the grayscale suppression module is used as input data for the segmentation module. pixel-level modulation, which adds weak amplitude modulation to unannotated radiation therapy images and strong amplitude modulation to weakly modulated images; Feature level modulation, which applies feature modulation to the four decoding layers of the division network; Construct a strong amplitude modulation uniformity loss and a feature modulation uniformity loss to obtain a total uniformity loss. First, a priority weight diagram of the block region is generated, the priority of the segmentation error region is learned by a classifier, a pixel-level weight diagram is output, a self-adjusting weight tower loss is constructed based on the priority weight of the block region, and a segmentation total loss is obtained.

[0010] Furthermore, the tumor radiotherapy image processing module acquires real-time tumor radiotherapy images, and after preprocessing, the images are input sequentially to an image correction module, a grayscale optimization module, a grayscale suppression module, and a tumor radiotherapy image segmentation module to obtain tumor radiotherapy image processing results.

[0011] By using the above solution, the present invention can achieve the following effects: (1) To address the problems of conventional tumor radiotherapy image processing systems, such as the confusion of the tumor and normal tissue boundaries, grayscale distortion, and an inadequate balance between detail and interference, leading to unreliable processing results, this method introduces dynamic regional correction coefficients to perform nonlinear correction on tumor radiotherapy images, ensuring that tumor regions are smooth and bright while ensuring that the grayscale of normal lung tissue does not exceed an acceptable range. A tumor edge gradient is introduced to enhance grayscale stretching in the edge region, addressing edge blur caused by tumor grayscale congestion. Dynamic correction coefficients are used to protect non-edge regions and avoid grayscale distortion. An ideal tumor grayscale weight is designed for detail reconstruction, forming a dual mechanism of enhancement and suppression, avoiding the loss of detail in high-grayscale areas and further improving the reliability of tumor radiotherapy image processing. (2) Conventional tumor radiotherapy image processing systems have poor environmental adaptability, insufficient hierarchical adaptability for radiation therapy target regions, serious loss of detail, insufficient segmentation accuracy for important regions, and poor tumor radiotherapy image processing results. This method addresses the problems of poor environmental adaptability, insufficient hierarchical adaptability for radiation therapy target regions, serious loss of detail, insufficient segmentation accuracy for important regions, and poor tumor radiotherapy image processing results. This method uses two-dimensional modulation at the pixel and feature levels to adapt to the hierarchical needs of the target region and achieve multi-layer precision segmentation. Hierarchical feature optimization is used to prevent the loss of tumor edge details during feature transfer. Uniformity-controlled loss and self-adjusting weighted tower loss are developed to constrain the segmentation results of amplitude modulation and feature modulation, dynamically identifying mis-segmented regions. The hierarchical loss ensures the consistency of the segmentation of the multi-layer structure of the target region, further improving the effectiveness of tumor radiotherapy image processing. [Brief explanation of the drawings]

[0012] [Figure 1] 1 is a flow diagram of the intelligent tumor radiation therapy image processing system provided by the present invention; [Figure 2] FIG. 1 is a functional schematic diagram of a tumor radiation therapy image segmentation module.

[0013] The accompanying drawings are included to provide a further understanding of the invention, constitute a part of the specification, and are used to explain the invention together with the embodiments thereof, not to limit the invention. DETAILED DESCRIPTION OF THE INVENTION

[0014] Hereinafter, the technical means in the embodiments of the present invention will be described clearly and completely with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative labor are included in the protection scope of the present invention.

[0015] In the description of the present invention, terms indicating orientations or positional relationships, such as "upper," "lower," "front," "rear," "left," "right," "top," "bottom," "inside," and "outside," are based on the orientations and positional relationships shown in the drawings, are intended to facilitate and simplify the description of the present invention, and do not indicate or imply that a particular device or component must have a particular orientation or be configured or operate in a particular orientation. Therefore, they should not be understood as limiting the present invention.

[0016] [Example 1] Referring to FIG. 1, the present invention provides an intelligent tumor radiotherapy image processing system, which includes an image acquisition module, an image correction module, a grayscale optimization module, a grayscale suppression module, a tumor radiotherapy image segmentation module, and a tumor radiotherapy image processing module, the image acquisition module acquires historical tumor radiation treatment images to obtain an original radiation treatment image set; The image correction module normalizes the original radiation therapy image, dynamically generates grayscale correction coefficients by region, and performs global grayscale correction; The grayscale optimization module extracts grayscale channels from the corrected original radiation therapy image, defines an adaptive grayscale conversion function incorporating edge gradients, generates dynamic grayscale correction coefficients, and obtains a grayscale optimized image; the grayscale suppression module calculates an ideal tumor grayscale weight for the grayscale optimized image to obtain a grayscale suppressed radiation treatment image; The tumor radiotherapy image segmentation module designs a segmentation network based on U-Net++, performs pixel-level weak amplitude modulation and strong amplitude modulation on the radiotherapy image obtained by the grayscale suppression module, and performs feature modulation on the decoding layer at the characteristic level, constructs a self-adjusting weight tower loss, and segments the tumor radiotherapy image; The tumor radiotherapy image processing module provides an intelligent tumor radiotherapy image processing system that performs image processing on real-time tumor radiotherapy images.

[0017] [Example 2] Referring to Figure 1, this embodiment is based on the above-mentioned embodiment. The image acquisition module acquired historical tumor radiotherapy images, including CT images, MRI images, and PET images. Preprocessing was performed, including unified conversion of all modal formats to the NIfTI-1 format, geometric correction, and geometric correction of multimodal tumor radiotherapy images based on the Elastix registration toolkit. Image annotation was then performed, including tumor target region annotation, OAR (organ at risk) annotation, and normal tissue annotation. Target region types included GTV (gross tumor region), CTV (clinical target region), and PTV (planning target region). OARs included organs corresponding to head and neck tumors, thoracic tumors, and abdominal tumors. Normal tissues included muscle tissue, fat tissue, and bone tissue. This resulted in an original radiotherapy image set.

[0018] [Example 3] Referring to Figure 1, this embodiment is based on the above embodiment. The image correction module is adapted to correct the problem that the global grayscale balance of radiation therapy images is easily disturbed due to differences in scanning parameters (for example, the grayscale of low-dose areas of lung tumors is too dark and easily confused with normal tissue, and the grayscale of high-dose areas of bone tissue is saturated, causing the boundary of the adjacent tumor to be hidden). Therefore, the image correction module uses the global average grayscale of the image as the core basis to improve the brightness of the entire tumor area and stabilize the grayscale of normal tissue, providing a basis for globally adaptive grayscale. Specifically, the original radiation therapy image is normalized to [0,1], the grayscale average values ​​of the two regions are introduced, and the normalized grayscale average values ​​of the two regions are calculated.

number

[0019] [Example 4] Referring to Figure 1, this embodiment is based on the above embodiment. The grayscale optimization module detects local grayscale issues in the original radiation therapy image after correction, such as dense grayscale in the tumor region and sudden changes in the grayscale of normal tissue and metal artifacts. The grayscale optimization module suppresses high-frequency artifacts through histogram clipping and enhances low-frequency tumor signals through weighting. Combined with an adaptive grayscale conversion function, the grayscale of the tumor region is precisely optimized while preserving the grayscale characteristics of the OAR. Specifically, the grayscale channel is extracted from the corrected image, and the probability density function of the grayscale level is calculated.

number

number

[0020] [Example 5] Referring to Figure 1, this example is based on the above example. The grayscale suppression module recognizes that there is detail loss and OAR interference in high grayscale areas in the original radiation therapy image after grayscale optimization. By combining the ideal tumor grayscale weight and multi-scale fusion using Laplacian pyramid, more details are retained in areas close to the typical tumor grayscale and the contribution to highlights and OAR areas is suppressed, forming a dual enhancement-suppression mechanism. Specifically, the ideal tumor grayscale weight is calculated,

number

[0021] [Example 6] 1 and 2, this embodiment is based on the above embodiment. Because the tumor radiotherapy image segmentation module has modal-specific noise, scan parameter differences, and artifacts in the tumor radiotherapy image, it uses two-dimensional modulation of the pixel layer + feature layer to simulate various variations in the radiotherapy image acquisition process, including CT value drift, MRI motion artifacts, and PET tracer distribution differences, and allows the model to learn robust features that are insensitive to the variations in the radiotherapy image. Specifically, it includes the following: Segmentation network design: The segmentation network is based on U-Net++ and includes an input layer, encoding layer, decoding layer, and output layer. Parameters are updated based on the AdamW optimizer, and the learning rate uses cosine annealing scheduling. Model performance is evaluated by classification accuracy. The encoding layer includes four downsampling layers, each containing two 3x3 convolution layers + ReLU. The decoding layer includes four upsampling layers, using dense skip connections. The output layer is composed of 1x1 convolution + Softmax, and outputs the segmentation results. Input data preparation: Grayscale suppression module output

number

number

number

number

number

number

[0022] [Example 7] Referring to Figure 1, this embodiment is based on the above embodiment. The tumor radiotherapy image processing module acquires real-time tumor radiotherapy images, and after preprocessing, the images are input sequentially to an image correction module, a grayscale optimization module, a grayscale suppression module, and a tumor radiotherapy image segmentation module to obtain tumor radiotherapy image processing results.

[0023] While embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention.

[0024] Although the present invention and its embodiments have been described above, this description is not limiting, and what is shown in the accompanying drawings is merely one embodiment of the present invention, and the actual structure is not limited to this. In other words, all aspects and embodiments similar to the technical means designed by a person skilled in the art based on the suggestion thereof, without departing from the spirit of the invention, and without originality, should fall within the protection scope of the present invention.

Claims

1. An intelligent oncology radiotherapy image processing system, including: an image acquisition module, an image correction module, a grayscale optimization module, a grayscale suppression module, an oncology radiotherapy image segmentation module, and an oncology radiotherapy image processing module; the image acquisition module acquires historical tumor radiation treatment images to obtain an original radiation treatment image set; The image correction module normalizes the original radiation therapy image, dynamically generates grayscale correction coefficients by region, and performs global grayscale correction; The grayscale optimization module extracts grayscale channels from the corrected original radiation therapy image, defines an adaptive grayscale conversion function incorporating edge gradients, generates dynamic grayscale correction coefficients, and obtains a grayscale optimized image; the grayscale suppression module calculates an ideal tumor grayscale weight for the grayscale optimized image to obtain a grayscale suppressed radiation treatment image; The tumor radiotherapy image segmentation module designs a segmentation network based on U-Net++, performs pixel-level weak amplitude modulation and strong amplitude modulation and characteristic-level decoding layer feature modulation on the radiotherapy image obtained by the grayscale suppression module, and constructs a self-adjusting weighted tower loss to segment the tumor radiotherapy image; The intelligent tumor radiotherapy image processing system is characterized in that the tumor radiotherapy image processing module performs image processing on real-time tumor radiotherapy images.

2. 2. The intelligent tumor radiation therapy image processing system of claim 1, wherein the image correction module normalizes the original radiation therapy image, introduces the grayscale average values ​​of two regions, calculates the normalized grayscale average values ​​of the two regions respectively, dynamically generates grayscale correction coefficients for each region, and finally performs global grayscale correction.

3. 3. The intelligent tumor radiation therapy image processing system of claim 2, wherein the grayscale optimization module extracts a grayscale channel from the corrected image, calculates a probability density function of the grayscale levels, clips high-frequency interference, constructs a weighted histogram, defines an adaptive grayscale conversion function, introduces gradient information of the tumor margin, generates a dynamic grayscale correction coefficient, and obtains an optimized radiation therapy image.

4. The intelligent tumor radiation therapy image processing system of claim 3, characterized in that the grayscale suppression module calculates the ideal tumor grayscale weight, constructs a Laplace pyramid for each of the corrected image and the optimized image, generates a Laplace detail map, fuses the detail maps of each layer according to the ideal tumor grayscale weight, and obtains a radiation therapy image after grayscale suppression through reconstruction.

5. The tumor radiotherapy image segmentation module specifically includes: Based on U-Net++, we design a partitioned network including an input layer, an encoding layer, a decoding layer, and an output layer; The output of the grayscale suppression module is used as input data for the segmentation module. pixel-level modulation, which adds weak amplitude modulation to unannotated radiation therapy images and strong amplitude modulation to weakly modulated images; Feature level modulation, which applies feature modulation to the four decoding layers of the division network; Construct a strong amplitude modulation uniformity loss and a feature modulation uniformity loss to obtain a total uniformity loss. and constructing a division total loss.

6. The intelligent tumor radiation therapy image processing system of claim 5, characterized in that the construction of the total segmentation loss first generates a priority weight map of the block region, and then uses a discriminator to learn the priority of the segmentation error region, output a pixel-level weight map, and construct a self-adjusting weight tower loss based on the priority weight of the block region to obtain the total segmentation loss.

7. The intelligent tumor radiation therapy image processing system of claim 6, characterized in that the image acquisition module acquires historical tumor radiation therapy images, performs pre-processing, and then performs image annotation, and finally obtains a set of original radiation therapy images.

8. 10. The intelligent tumor radiotherapy image processing system of claim 7, wherein the tumor radiotherapy image processing module acquires real-time tumor radiotherapy images, preprocesses the images, and then inputs the images to an image correction module, a grayscale optimization module, a grayscale suppression module, and a tumor radiotherapy image segmentation module in order to obtain tumor radiotherapy image processing results.

Images