A method and system for identifying fracture areas in CT images

Abstract

The present invention relates to the field of image analysis technology, and discloses a method and system for identifying fracture regions of CT detection images. The method comprises: acquiring and preprocessing a CT image containing a fracture region; obtaining fusion features by multi-scale feature extraction; inputting the fusion feature map into CenterNet for detecting the fracture region; applying non-maximum suppression to the detection result to generate the final fracture region boundary box; compared with the fracture region identification method of CT detection images that relies on the Anchor mechanism in the prior art, especially under low computing resources and real-time detection conditions, the traditional method is difficult to achieve the technical problem of fracture region identification of high-precision CT detection images due to complex calculations and low efficiency. The present invention realizes efficient detection of fracture regions through a lightweight EfficientNetB0 network and a CenterNet detection framework, combined with the introduction of channel weighting, thereby avoiding the problem of low computing efficiency and improving the accuracy and real-time performance of detection.

Description

Technical Field

[0001] The present invention belongs to the technical field of image analysis, and in particular relates to a method and system for identifying a fracture area of ​​a CT detection image. Background Art

[0002] At present, with the advancement of medical imaging technology, CT images are increasingly used in fracture diagnosis. However, the existing fracture CT image recognition technology still has some shortcomings. For example, since fracture images are often accompanied by complex bone changes, cracks, lesions and other features, traditional image processing methods are often unable to accurately distinguish the location, type and severity of fractures when faced with different types of fractures and inconsistent scanning quality. This makes it difficult for existing technologies to meet the needs of fast and accurate fracture identification, especially when processing high-noise, low-contrast fracture CT images, which often fail to effectively separate the fracture area from normal tissue. Therefore, there is an urgent need for a method that can still achieve efficient and accurate identification of fracture areas under different fracture types and image quality conditions. Summary of the invention

[0003] In view of the above-mentioned technical deficiencies, the purpose of the present invention is to propose a method for identifying fracture areas in CT detection images, aiming to solve the technical problem that the conventional method for identifying fracture areas in CT detection images relying on the Anchor mechanism is difficult to achieve high-precision fracture area identification in CT detection images due to its complex calculation and low efficiency under low computing resources and real-time detection conditions.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solution: The present invention provides a method for identifying fracture areas in CT detection images.

[0005] The fracture area recognition method of the CT detection image comprises:

[0006] Step S10: acquiring CT image data including the fracture area, adjusting the window width and window level of the CT image data, linearly normalizing the pixel values ​​of the CT image data, and mapping them to the [0,1] interval to generate standardized image data X;

[0007] Step S20: using the standardized image data X as the input of the backbone network EfficientNetB0 network;

[0008] Extract the feature map from the third convolutional layer of the EfficientNetB0 network to obtain the shallow feature map X1, which contains the edge detail information and texture detail information in the image;

[0009] Extract feature maps from the 5th convolutional layer of the EfficientNetB0 network to obtain the middle-level feature map X2, including the tissue structure information around the fracture area and the continuity information of the bones;

[0010] Extract feature maps from the 7th convolutional layer of the EfficientNetB0 network to obtain deep feature maps X3, including the shape information, structural information and overall relationship information of the bones with surrounding tissues;

[0011] The shallow feature map X1, the middle feature map X2 and the deep feature map X3 are combined to obtain a multi-scale feature set Y, Y = {X1, X2, X3};

[0012] Step S30: For each feature map X in the multi-scale feature set Y i Perform global average pooling to generate a global description vector Z of the feature map i ;

[0013] The global description vector Z i Passed to the two-layer fully connected network and the channel weight vector w is generated through the Sigmoid activation function i ;

[0014] The channel weight vector w i Applied to the corresponding feature map X i The weighted feature map X is obtained from i ′ , the weighted feature map X ′ 1. X ′ 2 and X ′ 3. Add pixel by pixel to generate fusion feature map Z;

[0015] Step S40: Input the fused feature map Z obtained in step S30 into the CenterNet model to detect the fracture area. CenterNet generates a center point thermal pixel map on the fused feature map Z. Each pixel point is assigned a probability value to indicate the possibility of the pixel point being the center point of the fracture area.

[0016] Each pixel whose probability value is greater than the preset threshold is defined as a candidate pixel. For the candidate pixel, CenterNet further adjusts the corresponding size offset and scale information;

[0017] After the candidate pixels are adjusted by size offset and scale information, the CenterNet model generates a preliminary bounding box of the fracture area;

[0018] Step S50: Perform non-maximum suppression processing on the detection results output by CenterNet, set the confidence threshold to 0.5 to 0.7, remove low-confidence detection frames, and use the NMS algorithm to remove overlapping areas in the retained detection frames to generate the final fracture area boundary box.

[0019] Preferably, in step S10, the window level is set to 500HU to 700HU, and the window width is set to 1000HU to 1500HU.

[0020] Preferably, in step S30, the global description vector Z i The calculation formula is:

[0021]

[0022] Among them, H is the height of the feature map, W is the width of the feature map, h is the pixel coordinate in the vertical direction of the feature map, w is the pixel coordinate in the horizontal direction of the feature map, and X i (h,w) is the feature map X i The pixel value at the vertical pixel coordinate h and the horizontal pixel coordinate w.

[0023] Preferably, in step S30, a channel weight vector w is generated i The calculation formula is:

[0024] w i =σ(W2·ReLU(W1·Z i ))

[0025] Among them, W1 is the weight matrix of the first layer of the fully connected network, W2 is the weight matrix of the second layer of the fully connected network, σ is the Sigmoid activation function, and ReLU is the nonlinear activation function.

[0026] Preferably, in step S30, the weighted feature map X i ′ The calculation formula is: i ′ =X i ·w i .

[0027] Preferably, in step S40, the size offset is used to fine-tune the precise position of the center point, and the scale information is used to determine the size of the fracture area, thereby generating a final fracture area bounding box.

[0028] Preferably, before step S40, a CenterNet model pre-training process is included, wherein the CenterNet model adopts a focal loss function, and the formula is:

[0029]

[0030] Among them, x is the horizontal coordinate of the position, y is the vertical coordinate of the position, and N is the total number of targets in the image. is the center point probability value predicted by CenterNet at position (x, y), y xy is the true label value of the position (x, y), α and γ are adjustment hyperparameters used to control the balance of focal loss and the degree of attention paid to difficult and easy samples.

[0031] The present invention also provides a fracture region recognition system for CT detection images, comprising:

[0032] A data preprocessing module is used to obtain CT image data containing a fracture area, adjust the window width and window level of the CT image data, linearly normalize the pixel values ​​of the CT image data, and map them to the [0,1] interval to generate standardized image data X;

[0033] The feature extraction module is used to take the normalized image data X as the input of the backbone network EfficientNetB0 network;

[0034] Extract the feature map from the third convolutional layer of the EfficientNetB0 network to obtain the shallow feature map X1, which contains the edge detail information and texture detail information in the image;

[0035] Extract feature maps from the 5th convolutional layer of the EfficientNetB0 network to obtain the middle-level feature map X2, including the tissue structure information around the fracture area and the continuity information of the bones;

[0036] Extract feature maps from the 7th convolutional layer of the EfficientNetB0 network to obtain deep feature maps X3, including the shape information, structural information and overall relationship information of the bones with surrounding tissues;

[0037] The shallow feature map X1, the middle feature map X2 and the deep feature map X3 are combined to obtain a multi-scale feature set Y, Y = {X1, X2, X3};

[0038] The multi-scale feature aggregation module is used to aggregate each feature map X in the multi-scale feature set Y. i Perform global average pooling to generate a global description vector Z of the feature map i ;

[0039] The global description vector Z i Passed to the two-layer fully connected network and the channel weight vector w is generated through the Sigmoid activation function i ;

[0040] The channel weight vector w i Applied to the corresponding feature map X iThe weighted feature map X is obtained from i ′ , the weighted feature map X ′ 1. X ′ 2 and X ′ 3. Add pixel by pixel to generate fusion feature map Z;

[0041] The fracture area recognition module of the CT detection image is used to input the fusion feature map Z obtained in step S30 into the CenterNet model to detect the fracture area. CenterNet generates a center point thermal pixel map on the fusion feature map Z. Each pixel point is assigned a probability value to indicate the possibility of the pixel point being the center point of the fracture area.

[0042] Each pixel whose probability value is greater than the preset threshold is defined as a candidate pixel. For the candidate pixel, CenterNet further adjusts the corresponding size offset and scale information;

[0043] After the candidate pixels are adjusted by size offset and scale information, the CenterNet model generates a preliminary bounding box of the fracture area;

[0044] The post-processing module is used to perform non-maximum suppression on the detection results output by CenterNet, set the confidence threshold to 0.5 to 0.7, eliminate low-confidence detection frames, and use the NMS algorithm to remove overlapping areas in the retained detection frames to generate the final fracture area bounding box.

[0045] The present invention also provides a fracture area recognition device for CT detection images, comprising a memory, a processor, and a fracture area recognition program for CT detection images stored in the memory and executable on the processor. When the fracture area recognition program for CT detection images is executed by the processor, the fracture area recognition method for CT detection images is implemented.

[0046] The present invention also provides a computer program product, comprising a fracture region recognition program for CT detection images, wherein the fracture region recognition program for CT detection images implements the fracture region recognition method for CT detection images when executed by a processor.

[0047] The beneficial effect of the present invention is that compared with the fracture area recognition method of CT detection images that relies on the Anchor mechanism in the prior art, especially under low computing resources and real-time detection conditions, the traditional method is difficult to achieve the technical problem of high-precision fracture area recognition of CT detection images due to complex calculations and low efficiency. The present invention uses a lightweight EfficientNetB0 network and CenterNet detection framework, combined with the introduction of channel weighting, to achieve efficient detection of fracture areas, thereby avoiding the problem of low computing efficiency and improving the accuracy and real-time performance of detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0049] Figure 1 The figure is a flow chart of a first embodiment of a method for identifying fracture regions in CT detection images according to the present invention.

[0050] Figure 2 The present invention is a schematic diagram of a device for a method for identifying fracture areas in CT detection images. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] Embodiment 1: Figure 1 , which is a flow chart of a first embodiment of a method for identifying a fracture region in a CT detection image according to the present invention, provides a first embodiment of a method for identifying a fracture region in a CT detection image according to the present invention.

[0053] In a first embodiment, the method for identifying a fracture region of a CT detection image includes:

[0054] Step S10: acquiring CT image data including the fracture area, adjusting the window width and window level of the CT image data, linearly normalizing the pixel values ​​of the CT image data, and mapping them to the [0,1] interval to generate standardized image data X;

[0055] It should be noted that the purpose of adjusting the window width and window level of the CT image is to highlight the density characteristics of the fracture area and make the fracture area clearer in the image. Usually, the window width and window level suitable for the bone structure are selected, for example, the window level is set between 500 and 700, and the window width is set between 1000 and 1500, to ensure that the density range of the bone tissue can be clearly displayed.

[0056] It should be understood that the adjusted CT image is linearly normalized to map the pixel values ​​to the interval [0,1]. The purpose of normalization is to unify the dynamic range of the image data so that when the image is input into the neural network, the feature extraction can be more stable and robust. The standardized image data X is obtained through the above processing, which provides consistent data input for the feature extraction in the subsequent steps.

[0057] Step S20: using the standardized image data X as the input of the backbone network EfficientNetB0 network;

[0058] Extract the feature map from the third convolutional layer of the EfficientNetB0 network to obtain the shallow feature map X1, which contains the edge detail information and texture detail information in the image;

[0059] Extract feature maps from the 5th convolutional layer of the EfficientNetB0 network to obtain the middle-level feature map X2, including the tissue structure information around the fracture area and the continuity information of the bones;

[0060] Extract feature maps from the 7th convolutional layer of the EfficientNetB0 network to obtain deep feature maps X3, including the shape information, structural information and overall relationship information of the bones with surrounding tissues;

[0061] The shallow feature map X1, the middle feature map X2 and the deep feature map X3 are combined to obtain a multi-scale feature set Y, Y = {X1, X2, X3};

[0062] It should be noted that EfficientNetB0 is a lightweight convolutional neural network architecture that balances the depth, width, and resolution of the network through a composite scaling strategy, so that it has low computational overhead while maintaining high accuracy. Choosing EfficientNetB0 as the backbone network can achieve efficient feature extraction with limited computing resources, which is particularly suitable for complex feature extraction requirements in the fracture area recognition task of CT detection images.

[0063] It should be understood that the shallow features come from the 3rd convolutional layer and contain the edge and texture information of the image. This detailed information is crucial for locating tiny features such as fracture cracks. The middle-level features come from the 5th convolutional layer and contain the tissue structure and bone continuity information around the fracture area. The middle-level features provide local contextual information of the fracture area, which helps the model understand the overall shape of the fracture area. The deep features come from the 7th convolutional layer and contain the overall shape, structural information of the bone and its relationship with the surrounding tissues. These deep features provide global semantic information to help the model distinguish the overall relationship between the fracture area and the normal bone structure.

[0064] It should be understood that the multi-scale feature fusion method enables the model to capture detailed features and understand the overall structure of the bone in the identification of fracture areas in CT detection images, thereby improving the accuracy of fracture area identification. Through the above feature extraction steps, rich multi-scale information is provided for subsequent feature aggregation and detection, which helps to improve the effect of fracture area identification in CT detection images.

[0065] Step S30: For each feature map X in the multi-scale feature set Y i Perform global average pooling to generate a global description vector Z of the feature map i ;

[0066] The global description vector Z i Passed to the two-layer fully connected network and the channel weight vector w is generated through the Sigmoid activation function i ;

[0067] The channel weight vector w i Applied to the corresponding feature map X i The weighted feature map X is obtained from i ′ , the weighted feature map X ′ 1. X ′ 2 and X ′ 3. Add pixel by pixel to generate fusion feature map Z;

[0068] It should be understood that the purpose of global average pooling is to compress the spatial information of each feature map into a global description vector Z i , which can summarize the overall information of the feature map and reduce the spatial dimension of the feature map, making the subsequent channel weight generation process more efficient.

[0069] It should be understood that the two-layer fully connected network is used to i Nonlinear mapping is performed to generate weights for each channel. The Sigmoid activation function limits the output value to the interval (0, 1), thereby giving different importance to different channels in the feature map and enhancing the channel features related to the fracture area.

[0070] It should be understood that the channel weight vector w i The channel features related to the fracture area are enhanced, while irrelevant or minor channel features are suppressed. The weighted feature maps are added pixel by pixel to form a fused feature map Z, which contains both detail information and global structural information, providing a richer feature expression for subsequent fracture area detection.

[0071] Step S40: Input the fused feature map Z obtained in step S30 into the CenterNet model to detect the fracture area. CenterNet generates a center point thermal pixel map on the fused feature map Z. Each pixel point is assigned a probability value to indicate the possibility of the pixel point being the center point of the fracture area.

[0072] Each pixel whose probability value is greater than the preset threshold is defined as a candidate pixel. For the candidate pixel, CenterNet further adjusts the corresponding size offset and scale information;

[0073] After the candidate pixels are adjusted by size offset and scale information, the CenterNet model generates a preliminary bounding box of the fracture area;

[0074] It should be noted that after the fused feature map Y is input into the CenterNet model, CenterNet generates a center point heat map on the feature map. In the heat map, each pixel is assigned a probability value to indicate the possibility of the pixel being the center point of the fracture area. The purpose of generating the center point heat map is to preliminarily locate the center of the fracture area, so that the model can efficiently identify the potential location of the fracture on the feature map after multi-scale feature fusion.

[0075] It can be understood that according to the preset probability threshold, the pixels in the heat map with probability values ​​greater than the threshold are defined as candidate pixels. Candidate pixels are high-confidence locations that the model believes may belong to the center of the fracture area. Such a screening process effectively reduces the computational complexity and focuses on areas where fractures may exist.

[0076] It should be understood that for each candidate pixel, CenterNet further predicts its corresponding size offset and scale information. The size offset is used to fine-tune the position of the candidate pixel to obtain a more accurate center point position; the scale information is used to determine the size of the fracture area. Through this step, the model can more accurately determine the bounding box of the fracture area. After the candidate pixel is adjusted by the size offset and scale information, the CenterNet model generates a preliminary fracture area bounding box. This bounding box is obtained based on the candidate center point and scale prediction, and provides preliminary detection results for the subsequent non-maximum suppression (NMS) step.

[0077] Step S50: Perform non-maximum suppression processing on the detection results output by CenterNet, set the confidence threshold to 0.5 to 0.7, remove low-confidence detection frames, and use the NMS algorithm to remove overlapping areas in the retained detection frames to generate the final fracture area boundary box.

[0078] It should be noted that in the detection results output by CenterNet, each bounding box is accompanied by a confidence score, which is used to indicate the possibility of the existence of the target in the detection box. By setting the confidence threshold (usually between 0.5 and 0.7), low-confidence detection boxes can be screened out to reduce the false alarm rate. The setting of this threshold is based on the detection requirements. If higher accuracy is required, a higher threshold can be set to retain more reliable detection boxes.

[0079] It is understandable that in fracture area detection, there may be multiple detection frames overlapping on the same fracture area. Non-maximum suppression (NMS) is used to remove these redundant overlapping detection frames. The NMS algorithm retains the detection frame with the highest confidence and deletes other detection frames whose overlap exceeds a certain threshold, thereby generating a unique and accurate bounding box. Through NMS processing, the duplicate frames in the detection results can be significantly reduced, ensuring that the final output fracture area bounding box is the most realistic.

[0080] It should be understood that after confidence screening and NMS processing, the detection boxes that remain are the final fracture region bounding boxes. These bounding boxes represent the detected fracture regions and will serve as the final output results of the fracture region recognition method for CT detection images for subsequent medical diagnosis or further analysis.

[0081] In addition, the fracture region recognition system of a CT detection image provided by the present invention adopts a fracture region recognition method of a CT detection image in the above embodiment, which can solve the technical problem of fracture region recognition of a CT detection image. Compared with the prior art, the beneficial effect of the fracture region recognition system of a CT detection image provided by the present invention is the same as the beneficial effect of the fracture region recognition method of a CT detection image provided by the above embodiment, and the other technical features of the fracture region recognition system of a CT detection image are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0082] The present invention provides a fracture area recognition device for CT detection images, please refer to Figure 2A fracture region recognition device for CT detection images includes: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the fracture region recognition method for CT detection images in the above-mentioned embodiment 1. A fracture region recognition device for CT detection images in the embodiment of the present invention may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. A fracture region recognition device for CT detection images is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention. A fracture region recognition device for CT detection images may include a processing device 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. Various programs and data required for the operation of a fracture region recognition device for CT detection images are also stored in the RAM 1004. The processing device 1001, the ROM 1002, and the RAM 1004 are connected to each other via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 can allow a fracture region recognition device for CT detection images to communicate wirelessly or wired with other devices to exchange data. Although the figure shows a fracture region recognition device for CT detection images with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have alternatively.

[0083] The present invention also provides a computer program product, including a computer program, which implements the steps of the above-mentioned method for identifying a fracture area in a CT detection image when executed by a processor. The computer program product provided by the present invention can solve the technical problem of identifying a fracture area in a CT detection image. Compared with the prior art, the beneficial effects of the computer program product provided by the present invention are the same as the beneficial effects of the method for identifying a fracture area in a CT detection image provided by the above-mentioned embodiment, and will not be described in detail here.

[0084] In particular, according to the embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through a communication device, or installed from the storage device 1003, or installed from the ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present invention are executed.

[0085] It should be understood that the various parts disclosed in the present invention can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0086] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for identifying fracture areas in CT images, characterized in that: Methods include: Step S10: acquiring CT image data including the fracture area, adjusting the window width and window level of the CT image data, linearly normalizing the pixel values ​​of the CT image data, and mapping them to the [0,1] interval to generate standardized image data X; Step S20: using the standardized image data X as the input of the backbone network EfficientNetB0 network; Extract the feature map from the third convolutional layer of the EfficientNetB0 network to obtain the shallow feature map X1, which contains the edge detail information and texture detail information in the image; Extract feature maps from the 5th convolutional layer of the EfficientNetB0 network to obtain the middle-level feature map X2, including the tissue structure information around the fracture area and the continuity information of the bones; Extract feature maps from the 7th convolutional layer of the EfficientNetB0 network to obtain deep feature maps X3, including the shape information, structural information and overall relationship information of the bones with surrounding tissues; The shallow feature map X1, the middle feature map X2 and the deep feature map X3 are integrated to obtain the multi-scale feature set Y, Y = {X1, X2, X3}; Step S30: For each feature map X in the multi-scale feature set Y i Perform global average pooling to generate a global description vector Z of the feature map i ; The global description vector Z i Passed to the two-layer fully connected network and the channel weight vector w is generated through the Sigmoid activation function i ; The channel weight vector w i Applied to the corresponding feature map X i The weighted feature map X is obtained from i ′ , the weighted feature map X ′ 1. X ′ 2 and X ′ 3. Add pixel by pixel to generate fusion feature map Z; Step S40: Input the fused feature map Z obtained in step S30 into the CenterNet model to detect the fracture area. CenterNet generates a center point thermal pixel map on the fused feature map Z. Each pixel point is assigned a probability value to indicate the possibility of the pixel point being the center point of the fracture area. Each pixel whose probability value is greater than the preset threshold is defined as a candidate pixel. For the candidate pixel, CenterNet further adjusts the corresponding size offset and scale information; After the candidate pixels are adjusted by size offset and scale information, the CenterNet model generates a preliminary bounding box of the fracture area; Step S50: Perform non-maximum suppression processing on the detection results output by CenterNet, set the confidence threshold to 0.5 to 0.7, remove low-confidence detection frames, and use the NMS algorithm to remove overlapping areas in the retained detection frames to generate the final fracture area boundary box.

2. The method for identifying fracture areas in CT images according to claim 1, characterized in that: In step S10, the window level is set to 500HU to 700HU, and the window width is set to 1000HU to 1500HU.

3. The method for identifying fracture areas in CT images according to claim 1, characterized in that: In step S30, the global description vector Z i The calculation formula is: Among them, H is the height of the feature map, W is the width of the feature map, h is the pixel coordinate in the vertical direction of the feature map, w is the pixel coordinate in the horizontal direction of the feature map, and X i (h,w) is the feature map X i The pixel value at the vertical pixel coordinate h and the horizontal pixel coordinate w.

4. The method for identifying fracture areas in CT images according to claim 1, characterized in that: In step S30, a channel weight vector w is generated i The calculation formula is: In i =σ(W2 ReLU(W1 Z i )) Among them, W1 is the weight matrix of the first layer of the fully connected network, W2 is the weight matrix of the second layer of the fully connected network, σ is the Sigmoid activation function, and ReLU is the nonlinear activation function.

5. The method for identifying fracture areas in CT images according to claim 1, characterized in that: In step S30, the weighted feature map X i ′ The calculation formula is: i ′ =X i ·w i .

6. The method for identifying fracture areas in CT images according to claim 1, characterized in that: In step S40, the size offset is used to fine-tune the precise position of the center point, and the scale information is used to determine the size of the fracture area, thereby generating a final fracture area bounding box.

7. The method for identifying fracture areas in CT images according to claim 1, characterized in that: Before step S40, a CenterNet model pre-training process is included, wherein the CenterNet model adopts a focal loss function, and the formula is: Among them, x is the horizontal coordinate of the position, y is the vertical coordinate of the position, and N is the total number of targets in the image. is the center point probability value predicted by CenterNet at position (x, y), y xy is the true label value of the position (x, y), α and γ are adjustment hyperparameters used to control the balance of focal loss and the degree of attention paid to difficult and easy samples.

8. A fracture region recognition system for CT detection images, characterized in that: The fracture area recognition system of the CT detection image comprises: A data preprocessing module is used to obtain CT image data containing a fracture area, adjust the window width and window level of the CT image data, linearly normalize the pixel values ​​of the CT image data, and map them to the [0,1] interval to generate standardized image data X; The feature extraction module is used to take the normalized image data X as the input of the backbone network EfficientNetB0 network; Extract the feature map from the third convolutional layer of the EfficientNetB0 network to obtain the shallow feature map X1, which contains the edge detail information and texture detail information in the image; Extract feature maps from the 5th convolutional layer of the EfficientNetB0 network to obtain the middle-level feature map X2, including the tissue structure information around the fracture area and the continuity information of the bones; Extract feature maps from the 7th convolutional layer of the EfficientNetB0 network to obtain deep feature maps X3, including the shape information, structural information and overall relationship information of the bones with surrounding tissues; The shallow feature map X1, the middle feature map X2 and the deep feature map X3 are combined to obtain a multi-scale feature set Y, Y = {X1, X2, X3}; The multi-scale feature aggregation module is used to aggregate each feature map X in the multi-scale feature set Y. i Perform global average pooling to generate a global description vector Z of the feature map i ; The global description vector Z i Pass it to the two-layer fully connected network and generate the channel weight vector w through the Sigmoid activation function i ; The channel weight vector w i Applied to the corresponding feature map X i The weighted feature map X is obtained from i ′ , the weighted feature map X ′ 1. X ′ 2 and X ′ 3. Add pixel by pixel to generate fusion feature map Z; The fracture area recognition module of the CT detection image is used to input the fusion feature map Z obtained in step S30 into the CenterNet model to detect the fracture area. CenterNet generates a center point thermal pixel map on the fusion feature map Z. Each pixel point is assigned a probability value to indicate the possibility of the pixel point being the center point of the fracture area. Each pixel whose probability value is greater than the preset threshold is defined as a candidate pixel. For the candidate pixel, CenterNet further adjusts the corresponding size offset and scale information; After the candidate pixels are adjusted by size offset and scale information, the CenterNet model generates a preliminary bounding box of the fracture area; The post-processing module is used to perform non-maximum suppression on the detection results output by CenterNet, set the confidence threshold to 0.5 to 0.7, eliminate low-confidence detection frames, and use the NMS algorithm to remove overlapping areas in the retained detection frames to generate the final fracture area bounding box.

9. A fracture area recognition device for CT detection images, characterized in that: The fracture area recognition device of the CT detection image comprises: A memory, a processor, and a fracture area recognition program for CT detection images stored in the memory and executable on the processor, wherein the fracture area recognition program for CT detection images, when executed by the processor, implements the fracture area recognition method for CT detection images described in any one of claims 1 to 7.

10. A computer program product, characterized in that The computer program product comprises a fracture region recognition program for CT detection images, and when the fracture region recognition program for CT detection images is executed by a processor, the fracture region recognition method for CT detection images described in any one of claims 1 to 7 is implemented.

Images