Intestinal tract automatic sketching optimization method and system based on path planning prior information

Abstract

The invention discloses an automatic intestinal tract sketching optimization method and system based on path planning prior information. The method comprises the following steps that medical image data of a patient is acquired and preprocessed; generating path prior information of the intestinal tract through a path planning algorithm, wherein the path prior information comprises a starting point, an ending point and a path direction of the intestinal tract; jointly inputting the path prior information and the preprocessed medical image data into a deep learning model, and performing intestinal segmentation through the deep learning model to generate a preliminary sketching result; and performing post-processing optimization, including morphological operation, connectivity correction and smoothing processing, on the preliminary sketching result to obtain a final intestinal contour. By combining the prior information of path planning, the method can provide accurate structural guidance in the automatic delineation process, and ensures the continuity of the intestinal tract delineation path.

Description

Technical Field

[0001] The present invention relates to the field of medical information, and particularly relates to an optimized method and system for automatic intestinal contouring based on prior information of path planning. Background Art

[0002] During the process of tumor radiotherapy, the accurate contouring of normal tissues is a crucial step to ensure the treatment effect and patient safety, especially when dealing with complex anatomical structures such as the intestine. In radiotherapy, the accurate contouring of the intestine helps to ensure the accurate positioning of the target area and avoid excessive irradiation of normal tissues. However, traditional intestinal contouring methods still face many technical challenges when dealing with such complex structures. Especially in the application of automatic contouring software, problems such as incomplete contouring and discontinuous paths often occur, greatly affecting the accuracy and reliability of the contouring results. The disadvantages of traditional automatic contouring methods include:

[0003] (1) Incomplete and discontinuous automatic contouring: Modern automatic contouring software performs intestinal contouring based on anatomical models or deep learning models. Although it can speed up the contouring process, in practical applications, especially when dealing with complex anatomical structures such as the small intestine, the automatic contouring often has problems of incomplete contouring or discontinuous paths. Especially when there are complex morphological changes such as bending and torsion in the intestine, the results of automatic contouring often cannot completely cover all regions of the intestine, resulting in missing or broken drawing. This defect forces clinicians to often rely on manual repair to manually supplement the missing parts, increasing the workload and potentially causing human errors.

[0004] (2) Limitations of deep learning segmentation: Automatic segmentation methods based on deep learning have also been widely used in intestinal contouring. These methods identify intestinal structures by training neural network models. However, due to the limitations of training data and individual differences, the contouring results of deep learning models still have problems of discontinuity or blurred boundaries in some cases. Especially when the anatomical structures of patients have large differences, the model may not fully identify all intestinal paths, resulting in missing and inaccurate contouring results.

[0005] (3) Heavy workload of manual repair: Due to the above problems in the practical application of automatic contouring software, the intestinal contouring results usually require a large amount of manual correction. Doctors often need to manually check the contouring of each layer, supplement the missing parts, and adjust the inaccurate contours. Although this method ensures accuracy, it consumes a lot of time, has low efficiency, and depends on the experience of doctors, and is prone to subjective errors.

[0006] In summary, the problems of the existing intestinal path delineation methods include: the automatic delineation results cannot completely delineate the intestinal contour, especially in complex anatomical regions, and the delineated paths are often discontinuous or interrupted. The delineation results of most automatic delineation software require a large amount of manual correction. Doctors must check layer by layer and repair the missing parts, which not only increases the workload but also makes the treatment plan formulation process more cumbersome and time-consuming. The existing automated methods are difficult to fully adapt to the individual anatomical differences of patients, especially when dealing with complex anatomical structures such as the small intestine and colon with high individual differences, and the delineation effect is not satisfactory. Summary of the Invention

[0007] In view of the above-mentioned defects and deficiencies in the prior art, the present invention provides an optimized method and system for automatic intestinal delineation based on path planning prior information. By combining path planning and deep learning segmentation techniques and using the path information of the intestine as a prior constraint, this method effectively guides the delineation process to ensure the continuity and integrity of the delineated path, thus solving the incomplete and discontinuous problems in traditional methods. The path planning technique provides an accurate starting point and direction for the deep learning model, thereby helping the model to more accurately delineate the intestine and reducing errors caused by the complexity of the anatomical structure and body shape changes.

[0008] To achieve the above object, the present invention provides an optimized method for automatic intestinal delineation based on path planning prior information, which includes the following steps:

[0009] Obtain the medical image data of the patient and perform preprocessing;

[0010] Generate the path prior information of the intestine through a path planning algorithm, including the starting point, ending point and path direction of the intestine;

[0011] Input the path prior information and the preprocessed medical image data into a deep learning model together, and perform intestinal segmentation through the deep learning model to generate a preliminary delineation result;

[0012] Perform post-processing optimization on the preliminary delineation result, including morphological operations, connectivity correction and smoothing processing, to obtain the final intestinal contour.

[0013] A further improvement of the present invention is that the path planning algorithm includes the A* algorithm and the Dijkstra algorithm.

[0014] A further improvement of the present invention is that the deep learning model is a U-Net network architecture, and its input includes the fusion features of the path prior information and the medical image data, and the path prior information is combined with the image data through an additional channel or feature embedding method.

[0015] A further improvement of the present invention lies in that the U-Net network is jointly optimized using a cross-entropy loss function and a Dice loss function.

[0016] A further improvement of the present invention lies in that the generation step of the path prior information includes:

[0017] Determine the starting point and ending point of the intestine through anatomical analysis or manual annotation;

[0018] Use a path planning algorithm to calculate the optimal path of the intestine and smooth the path to eliminate noise.

[0019] A further improvement of the present invention lies in that in the post-processing optimization step, the morphological operation is used to remove isolated noise points or fill gaps, and the connectivity correction repairs path breaks through an interpolation algorithm or a region growing algorithm.

[0020] A further improvement of the present invention lies in that, according to the intestine automatic delineation optimization method described in claim 1, the medical image data includes one or more of CT, MRI, and CBCT images, and the preprocessing process includes denoising, normalization, and size standardization.

[0021] The present invention also provides an intestine automatic delineation optimization system based on path planning prior information for implementing the above-mentioned intestine automatic delineation optimization method based on path planning prior information.

[0022] The advantages of the present invention are as follows:

[0023] (1) By combining path planning prior information, the present invention can provide precise structural guidance during the automatic delineation process, ensuring the coherence of the intestine delineation path. Traditional automatic delineation methods are prone to path discontinuity or delineation errors, while path planning prior information can effectively eliminate these problems, making the delineation results more complete and accurate. Especially when facing anatomical structure changes, it can maintain the stability of the delineation.

[0024] (2) Traditional automatic delineation methods often require a large amount of manual correction, especially when dealing with complex anatomical structures such as the small intestine and colon, where errors and discontinuities are more serious. By using the path planning prior information in the present invention, the system can greatly reduce the errors in delineation, thereby reducing the workload of manual correction, improving efficiency, and reducing human errors.

[0025] (3) The present invention inputs the intestinal path information generated by the path planning technology as prior data into the deep learning model, making the model more accurate when dealing with complex anatomical structures. The prior information provided by path planning effectively constrains the output of the deep learning model, enabling the model to maintain high accuracy even under variable anatomical variations and noise interference, enhancing the robustness of the segmentation model.

[0026] (4) During radiotherapy, the change in the patient's body shape can lead to large errors in traditional contouring methods, especially in the contouring of complex structures such as the intestine. By using the spatial prior information provided by the path planning technology, the present invention can, to a certain extent, reduce the impact caused by the change in the patient's body shape, thereby improving the stability and accuracy of the contouring results.

[0027] (5) Accurate intestinal contouring is crucial for the design of radiotherapy plans, which can effectively avoid unnecessary radiation to normal tissues and ensure accurate dose distribution in the target area. The present invention can provide high-precision intestinal contouring results, provide reliable data support for radiotherapy plan optimization, improve the accuracy of treatment, reduce side effects, and ensure the treatment safety of patients.

[0028] (6) Since the path planning prior information can provide global guidance for the deep learning model, the system can adapt to the anatomical variations of different patients. Especially when the patient's body shape changes, it can still ensure the accuracy of the contouring results. Compared with traditional methods, the present invention has higher adaptability, is applicable to various types of patients, and has a broader application prospect. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a flowchart of the method for optimizing automatic intestinal contouring based on path planning prior information of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0030] The following specific examples illustrate the embodiments of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0031] For the purpose of illustration, some exemplary embodiments of the present invention are described. It should be understood that the present invention can be implemented in other ways not specifically shown in the drawings.

[0032] As Figure 1 shown, the present invention provides a method for optimizing automatic intestinal contouring based on path planning prior information. The purpose of the present invention is to solve the problems existing in the existing automatic intestinal contouring methods, such as discontinuous paths, incomplete contouring, and large workload of manual repair. By combining path planning prior information with deep learning technology, the automatic intestinal contouring process is optimized to improve the accuracy, continuity, and integrity of the contouring results. The specific objectives are as follows:

[0033] 1. Improve the accuracy and continuity of intestinal delineation: The present invention uses path planning prior information to provide constraints and guidance for automatic intestinal delineation, ensuring more accurate delineation of the intestinal contour and effectively avoiding the common problem of discontinuous paths in traditional methods. By precisely guiding the delineation of the intestinal path, it reduces delineation errors caused by anatomical variations or noise interference, ensuring that each anatomical structure can be accurately delineated.

[0034] 2. Reduce the need for manual repair: Existing automatic delineation methods often require a large amount of manual repair, increasing the workload of doctors and relying on the experience and judgment of doctors, which may lead to subjective errors. The present invention significantly reduces the need for manual correction through an optimized automatic delineation method, improves the delineation efficiency, shortens the time for formulating treatment plans, and reduces the workload of doctors.

[0035] 3. Improve the quality and safety of radiotherapy plans: By accurately delineating the contours of normal tissues such as the intestine, the present invention provides more reliable anatomical data for subsequent dose evaluation. This provides a more accurate reference basis for formulating radiotherapy plans, can effectively avoid unnecessary over-irradiation of normal tissues, reduce the occurrence of side effects, and improve the treatment effect.

[0036] 4. Adapt to individual anatomical differences: The present invention can fully consider the individual anatomical differences of patients and optimize the automatic delineation of the intestine. Especially in the case of complex and variable intestinal morphology, it can still ensure high delineation quality. Through path planning guidance, the system can adaptively adjust the delineation strategy, improving the robustness and accuracy of delineation.

[0037] The purpose of the present invention is to solve the problems of low accuracy of intestinal delineation and large workload of manual repair in the prior art through an innovative automatic delineation method combining path planning and deep learning, improve the accuracy and safety of radiotherapy plans, and promote the development of intestinal automatic delineation technology towards a more efficient and accurate direction.

[0038] The present invention first constructs a preliminary solution of the intestinal path through image data preprocessing and path planning algorithms (such as the A* algorithm). The image data is processed by denoising and standardization to ensure stable image quality. The path planning uses the A* algorithm, which can efficiently find the optimal path from the starting point to the ending point of the intestine and dynamically adjust according to the actual anatomical structure and changes of the intestine. The A* algorithm formula is:

[0039] f(n) = g(n) + h(n)

[0040] Among them, f(n) is the estimated total cost from the starting point to node n, g(n) is the actual cost, and h(n) is the heuristic estimated cost. This path information includes the starting point, ending point of the intestinal path and the direction of the path, etc., and is input as prior information of the deep learning network.

[0041] The intestinal path information provided by path planning can be input as prior information into the deep learning network. The specific method is as follows:

[0042] Take the intestinal path information (such as the coordinates of the starting and ending points, the direction of the path, etc.) generated by the path planning model as an additional input to the deep learning network. This information is input into the network in parallel with the CT image data, enabling the network to refer to these prior paths during the segmentation process to ensure the continuity and accuracy of the path. For example, the path planning result can be passed to the U-Net network by embedding the coordinates of the path points into the image features as an additional input channel (such as in the form of a mask of the CT image data). In this way, the network not only relies on image features for segmentation but also combines the anatomical path of the intestine, avoiding the situation of discontinuous paths.

[0043] `

[0044] I = f(I, P)

[0045] Where: I is the original image data, p is the prior path information provided by the path planning model, and I ` is the input data combined with the path information. When the network performs intestinal segmentation, it will consider the prior path, making the final segmentation result more in line with the true anatomical structure.

[0046] The present invention uses a convolutional neural network (CNN) based on U-Net for automatic delineation of the intestine. The U-Net architecture includes an encoder and a decoder, which extract image features through multiple layers of convolution and pooling, and then restore the image resolution through deconvolution. The output of the U-Net model is the probability that each pixel belongs to the intestine, which is optimized by the cross-entropy loss function. The formula is:

[0047]

[0048] Where, y i is the true label, and p i is the predicted probability. To improve the segmentation accuracy, the intestinal path information provided by path planning is used as an additional input and input into the network in parallel with the CT image. In this way, the network not only relies on image features for segmentation but also combines the anatomical path of the intestine to ensure the continuity and accuracy of the path. The path planning result can be passed to the U-Net network by embedding the coordinates of the path points into the image features as an additional input channel.

[0049] After the deep learning network generates the initial delineation result, morphological operations are used to further optimize the delineation result. Morphological processing includes dilation and erosion operations, which are used to remove noise or fill small gaps in the path. Path repair and connectivity optimization are based on path planning information. Interpolation methods and region growing algorithms are used to repair the segmentation result to ensure the coherence and integrity of the path.

[0050] The intestinal delineation system of the present invention consists of a data input module, a path planning module, a deep learning network, a post-processing module, and a result output module. First, the CT or MRI images of the patient are received through the image data input module and preprocessed. Then, the path planning module generates intestinal path information based on the image data. Next, the deep learning network uses the U-Net model for intestinal segmentation and uses the path information as a priori input to optimize the delineation result. Finally, the post-processing module uses morphological operations and path repair techniques to optimize the segmentation result and outputs the final intestinal delineation result for the radiotherapy planning system to perform dose evaluation and treatment optimization.

[0051] In a specific embodiment, the implementation method of the present invention includes:

[0052] First, collect the medical image data of patients containing intestinal regions such as the small intestine and colon. Commonly used image formats include CT, MRI, or CBCT images. To ensure the accuracy of the deep learning model, a large number of well-annotated data sets are required for training. The intestinal regions in these data sets should have been manually delineated by professional doctors and used as the labels of the training set.

[0053] The medical image data should meet the following requirements: CT or MRI scan data with sufficient anatomical details; well-annotated intestinal regions to ensure accurate boundaries for each intestinal structure; the image data needs to be preprocessed, with unified size, normalized, and denoised to reduce the impact of data variation.

[0054] In the present invention, path planning is provided to the deep learning model as a priori information to help improve the accuracy of the delineation result. The specific steps are as follows: Path generation: Use path planning algorithms (such as graph-based path planning methods, A* algorithm, Dijkstra algorithm, etc.) to generate possible intestinal paths in the image. The input of the path planning algorithm is the starting and ending positions of the intestine, which can be obtained through anatomical analysis or manual annotation by doctors. Path smoothing and optimization: According to the path planning result, perform path smoothing to reduce noise and irregular fluctuations in the path and ensure that the intestinal path is coherent and conforms to the anatomical structure.

[0055] During the path planning process, determine the starting and ending positions of the intestine. Based on the anatomical structure and imaging data, use graph algorithms to calculate the optimal path of the intestine. Smooth and correct the calculated path to ensure its coherence in the real anatomical structure.

[0056] During the training process of the deep learning model: The model selection uses image segmentation networks such as U-Net or 3D U-Net as the basic architecture, which is suitable for efficient segmentation tasks in medical images. The input data is preprocessed imaging data, including CT, MRI, etc.

[0057] The way of inputting prior data is as follows: Use the intestinal path generated by path planning as prior information, and through the method of feature fusion, use it as an additional input for the network to learn. Channel attention mechanisms (such as SE-ResNet) or image guidance techniques can be used to fuse the path information with the image features.

[0058] During the training process, use common medical image segmentation loss functions such as Dice loss and cross-entropy loss to optimize the model performance, and combine the prior information of path planning for guidance to reduce errors.

[0059] The dataset is divided into a training set and a validation set. The training set is used for model training, and the validation set is used to adjust the model hyperparameters. The network adopts an alternating training strategy. First, it is trained on standard segmentation tasks, and then path planning information is added for multi-stage optimization. During the training process, monitor the accuracy of the model through the validation set to ensure the accuracy and coherence of the intestinal delineation.

[0060] The process of automatic delineation using the trained model includes the following steps:

[0061] Image input: Input the new patient image into the trained deep learning model. The image data goes through the same preprocessing process as during training (such as normalization, denoising, etc.).

[0062] Prior information fusion: The model automatically receives the intestinal path information from path planning as an auxiliary input. Through the attention mechanism or feature fusion layer of the network, the network can better use the prior information to constrain the segmentation boundary.

[0063] Automatic delineation output: The model outputs the segmentation result, that is, the predicted boundary of the intestinal region. At this time, the delineation result should include a continuous and complete intestinal path, accurately depicting the contours of intestinal regions such as the small intestine and colon.

[0064] When there are details errors in the result of automatic delineation, use post-processing techniques to further optimize the delineation result; specifically, it includes:

[0065] Morphological processing: Use morphological operations such as dilation and erosion to refine the outline results and remove noise and isolated small areas.

[0066] Connectivity correction: In case of breaks in the outline, the broken path is repaired through connectivity analysis to ensure the continuity of the intestinal path.

[0067] Smoothing: Through curve smoothing technology, the jaggedness of the path is reduced and the outlining effect is further optimized.

[0068] The automatically delineated intestinal area can be used as input for radiotherapy dose assessment and provide a basis for radiotherapy plan optimization. Use a radiotherapy planning system (such as TPS) to simulate the dose distribution based on the delineated intestinal area. Determine the dose range of the intestinal tissue and assess whether it exceeds the safe range. Based on the dose assessment results, combined with the delineated intestinal area data, optimize the treatment plan to ensure a reasonable dose distribution between the target area and normal tissue.

[0069] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by a person of ordinary skill in the art without departing from the spirit and technical concept disclosed by the present invention shall still be covered by the claims of the present invention.

Claims

1. A method for automatically outlining and optimizing the intestine based on path planning prior information, characterized in that: The following steps are involved: Obtain the patient's medical imaging data and perform preprocessing; Generate intestinal path prior information through path planning algorithm, including the starting point, end point and path direction of the intestine; The path prior information and the preprocessed medical image data are input into a deep learning model, and the intestinal tract is segmented by the deep learning model to generate a preliminary delineation result; The preliminary delineation results are post-processed and optimized, including morphological operations, connectivity correction and smoothing, to obtain the final intestinal contour.

2. The method for automatic intestinal delineation optimization based on path planning prior information according to claim 1, characterized in that: The path planning algorithms include A* algorithm and Dijkstra algorithm.

3. The method for automatic intestinal delineation optimization based on path planning prior information according to claim 1, characterized in that: The deep learning model is a U-Net network architecture, and its input includes the fusion features of the path prior information and the medical image data. The path prior information is combined with the image data through an additional channel or feature embedding method.

4. The method for automatic intestinal delineation optimization based on path planning prior information according to claim 3, characterized in that: During the training process, the U-Net network uses the cross entropy loss function and the Dice loss function for joint optimization.

5. The method for automatic intestinal delineation optimization based on path planning prior information according to claim 1, characterized in that: The step of generating the path prior information comprises: Determine the start and end points of the intestine through anatomical analysis or manual annotation; The optimal path of the intestine is calculated using a path planning algorithm, and the path is smoothed to eliminate noise.

6. The method for automatic intestinal delineation optimization based on path planning prior information according to claim 1, characterized in that: In the post-processing optimization step, the morphological operation is used to remove isolated noise points or fill gaps, and the connectivity correction is used to repair path breaks through an interpolation algorithm or a region growing algorithm.

7. The method for automatic intestinal delineation optimization based on path planning prior information according to claim 1, characterized in that: The method for automatic intestinal delineation optimization according to claim 1 is characterized in that the medical imaging data includes one or more of CT, MRI, and CBCT images, and the preprocessing process includes denoising, normalization, and size standardization.

8. An automatic intestinal outline optimization system based on path planning prior information, characterized in that: Used to implement the intestinal automatic delineation optimization method based on path planning prior information as described in any one of claims 1 to 7.

Images