Postpartum hemorrhage amount identification method and system based on deep neural network

Abstract

The invention provides a postpartum hemorrhage amount identification method and system based on a deep neural network, and the method comprises the steps: obtaining an image containing a plurality of blood gradients; preprocessing the image containing multiple blood gradients, and determining the preprocessed image; inputting the preprocessed image into a pre-trained deep neural network model, and determining the predicted bleeding amount of the image containing multiple blood gradients; post-processing the predicted bleeding amount of the image containing the multiple blood gradients according to the global information of the image containing the multiple blood gradients, and outputting the total predicted bleeding amount corresponding to the image containing the multiple blood gradients and the blood pixel distribution of the bleeding area. Through the method and the device, feature extraction and analysis are performed on the dressing image of postpartum hemorrhage based on the deep neural network model, the postpartum hemorrhage amount is dynamically and accurately evaluated in real time, auxiliary decision support is provided for clinical midwives, and the response speed and accuracy of postpartum hemorrhage treatment are improved.

Description

Technical Field

[0001] The present disclosure relates to the technical field of medical image processing, and in particular, to a method and system for identifying postpartum hemorrhage based on a deep neural network. Background Art

[0002] With the rapid development of medical technology and artificial intelligence, automated medical diagnostic tools based on machine learning and deep learning are gradually becoming a research hotspot in the medical field. In recent years, advances in image processing technology have provided new solutions for medical image analysis, especially in automated diagnosis and pathological analysis. Postpartum hemorrhage is a common obstetric emergency that can threaten the life of the mother in severe cases. Therefore, it is particularly important to quickly and accurately assess the amount of postpartum hemorrhage.

[0003] Traditional methods for assessing the amount of postpartum blood loss usually rely on visual inspection and weighing by medical staff. Visual inspection relies on the experience of medical staff, is highly subjective, and is prone to large errors. Especially when faced with a large or rapidly increasing amount of blood loss, the accuracy of the visual inspection method decreases significantly, resulting in the failure to effectively treat postpartum hemorrhage in a timely manner, increasing the health risks of the mother. The weighing method can improve the accuracy of blood loss measurement to a certain extent, but its operation is cumbersome, time-consuming, and requires specialized equipment, which is not conducive to rapid application in emergency settings.

[0004] At the same time, with the widespread application of deep learning technology, especially the excellent performance of convolutional neural networks (CNN) in image segmentation, object recognition and classification tasks, medical image processing technology has made significant progress. Studies have shown that image processing models based on deep learning can effectively improve the segmentation and classification accuracy of medical images. For example, the UNet model, as a classic convolutional neural network structure, has achieved widespread success in medical image segmentation tasks. It can capture local and global information of images at the same time through the skip connection mechanism, which is particularly suitable for medical image processing under complex backgrounds.

[0005] In the field of bleeding volume assessment, some research and commercial applications have begun to use image processing technology to automatically identify bleeding volume. For example, Gauss Surgical, an American company, has developed a bleeding volume assessment system based on image recognition technology, which uses a smartphone camera to photograph surgical dressings and calculates the amount of bleeding on the dressings through image processing technology. Although this system has achieved good results in surgical scenarios, its application in postpartum bleeding volume identification is not mature enough, and existing models are mostly concentrated in surgical environments. There is still a large technical gap in bleeding volume identification in obstetric scenarios.

[0006] In addition, the assessment of postpartum hemorrhage not only needs to deal with complex clinical scenarios, but also needs to process mixed information of multimodal data (such as blood, body fluids, etc.). This places higher demands on traditional image processing methods, requiring more intelligent algorithms to accurately identify bleeding areas and estimate the amount of bleeding. Traditional image processing methods often show problems of insufficient robustness and low accuracy when faced with complex situations such as different lighting conditions and mixing of blood with other body fluids. Therefore, the development of an automated postpartum hemorrhage volume recognition system based on deep learning can greatly improve the accuracy and real-time performance of the assessment, help medical staff respond quickly in emergencies, and thus reduce the risk to the lives of mothers.

[0007] At present, there is still a relative lack of specialized automated identification systems for postpartum hemorrhage, and existing studies mostly focus on the assessment of intraoperative blood loss, lacking application in obstetric emergencies.

[0008] In summary, although some progress has been made in the field of automated identification of bleeding volume, the existing technology still has many shortcomings, such as low evaluation accuracy and inability to quickly adapt to complex clinical environments. Therefore, it is urgent to develop an automated postpartum bleeding volume identification method and system that can meet clinical needs, with high accuracy and robustness, to solve the current clinical needs for accurate identification and rapid treatment of postpartum hemorrhage. Summary of the invention

[0009] In view of the defects in the prior art, the purpose of the present invention is to provide a method and system for identifying postpartum hemorrhage based on deep neural network.

[0010] To achieve the above object, according to one aspect of the present disclosure, a method for identifying postpartum hemorrhage based on a deep neural network is provided, comprising:

[0011] Acquire images containing multiple blood gradients;

[0012] Preprocessing the image containing the multiple blood gradients to determine a preprocessed image;

[0013] Inputting the preprocessed image into a pretrained deep neural network model to determine the predicted bleeding volume of the image containing multiple blood gradients;

[0014] The predicted bleeding volume of the image containing multiple blood gradients is post-processed according to the global information of the image containing multiple blood gradients, and the total predicted bleeding volume corresponding to the image containing multiple blood gradients and the blood pixel distribution in the bleeding area are output, wherein the global information of the image containing multiple blood gradients represents the proportion of each blood gradient.

[0015] Optionally, the pre-trained deep neural network model includes a preset bleeding volume weight prediction model.

[0016] Optionally, inputting the preprocessed image into a pretrained deep neural network model to determine the predicted bleeding volume of the image containing multiple blood gradients includes:

[0017] Performing image segmentation processing on the preprocessed image to determine a bleeding area containing multiple blood gradients and bleeding area features, wherein the bleeding area features include the area, color depth, and morphology of the bleeding area;

[0018] Determining a bleeding volume characteristic of the bleeding area corresponding to each blood gradient according to the bleeding area including the multiple blood gradients and the bleeding area characteristics;

[0019] The preset bleeding volume weight prediction model is used to perform weighted sum processing on the bleeding volume characteristics of the bleeding area corresponding to each blood gradient to determine the predicted bleeding volume of the image containing multiple blood gradients.

[0020] Optionally, determining the bleeding volume feature of the bleeding area corresponding to each blood gradient according to the bleeding area containing multiple blood gradients and the bleeding area feature includes:

[0021] Performing pixel area mapping on the bleeding area containing multiple blood gradients to determine the pixel area ratio of the bleeding area corresponding to each blood gradient;

[0022] The bleeding volume feature of the bleeding area corresponding to each blood gradient is determined according to the pixel area ratio of the bleeding area corresponding to each blood gradient and the bleeding area feature of the bleeding area corresponding to each blood gradient.

[0023] Optionally, performing pixel area mapping on the bleeding area containing multiple blood gradients to determine the pixel area ratio of the bleeding area corresponding to each blood gradient includes:

[0024] Performing color segmentation on the bleeding area containing multiple blood gradients to determine the number of pixels in the bleeding area corresponding to each blood gradient;

[0025] According to the number of pixels in the bleeding area corresponding to each blood gradient, the pixel area ratio of the bleeding area corresponding to each blood gradient is determined.

[0026] Optionally, the using the preset bleeding volume weight prediction model to perform weighted sum processing on the bleeding volume features of the bleeding area corresponding to each blood gradient to determine the predicted bleeding volume of the image containing multiple blood gradients includes:

[0027] Inputting the pixel area ratio of the bleeding area corresponding to each blood gradient into the preset bleeding volume weight prediction model to determine the weight of the bleeding area corresponding to each blood gradient;

[0028] The bleeding volume feature of the bleeding area corresponding to each blood gradient and the weight of the bleeding area corresponding to each blood gradient are weighted summed to determine the predicted bleeding volume of the image containing multiple blood gradients.

[0029] Optionally, the acquiring of images containing multiple blood gradients comprises:

[0030] An image acquisition device is used to acquire images containing multiple blood gradients from a dressing containing multiple blood gradients.

[0031] Optionally, preprocessing the image containing multiple blood gradients to determine the preprocessed image includes:

[0032] The image containing the multiple blood gradients is subjected to image resolution adjustment processing, image normalization operation, cropping operation, noise reduction operation and edge detection operation to determine the pre-processed image.

[0033] According to a second aspect of the present disclosure, a postpartum hemorrhage volume identification system based on a deep neural network is provided, comprising:

[0034] An image acquisition module, used for acquiring images containing multiple blood gradients;

[0035] An image preprocessing module, used for preprocessing the image containing multiple blood gradients to determine the preprocessed image;

[0036] A bleeding volume prediction module, used for inputting the preprocessed image into a pretrained deep neural network model to determine the predicted bleeding volume of the image containing multiple blood gradients;

[0037] The total bleeding volume determination module is used to post-process the predicted bleeding volume of the image containing multiple blood gradients according to the global information of the image containing multiple blood gradients, and output the total predicted bleeding volume and blood pixel distribution in the bleeding area corresponding to the image containing multiple blood gradients, wherein the global information of the image containing multiple blood gradients represents the proportion of each blood gradient.

[0038] According to a third aspect of the present disclosure, a non-temporary computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the steps of any one of the methods provided in the first aspect of the present disclosure are implemented.

[0039] According to a fourth aspect of the present disclosure, there is provided an electronic device, characterized in that it includes:

[0040] a memory having a computer program stored thereon;

[0041] A processor is used to execute the computer program in the memory to implement the steps of any one of the methods provided in the first aspect of the present disclosure.

[0042] Compared with the prior art, the embodiments of the present disclosure have at least one of the following beneficial effects:

[0043] Through the above technical scheme, a pre-trained deep neural network model is introduced to extract and analyze the features of the dressing image, effectively segment the bleeding areas of the various blood gradient images of the dressing image, and predict the amount of bleeding in the bleeding areas with different blood gradients. Combined with the global information of the proportion of each blood gradient of the image containing multiple blood gradients, the total postpartum bleeding volume and bleeding area can be evaluated in real time, dynamically and accurately, which is convenient for real-time bleeding volume evaluation in the clinical environment, improves the accuracy of postpartum bleeding volume identification, provides auxiliary decision support for clinical midwives, and improves the response speed and accuracy of postpartum hemorrhage treatment.

[0044] The embodiments of the present disclosure use a pre-trained deep neural network model to cope with a variety of complex situations in clinical scenarios, such as changes in lighting and mixing of blood with other body fluids, thereby improving the accuracy of postpartum hemorrhage assessment and having wide applicability and strong robustness.

[0045] The embodiments of the present disclosure provide real-time bleeding volume assessment in a clinical environment based on a standardized image acquisition process and pre-trained deep neural network model processing, facilitating medical staff to make quick decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Other features, objects and advantages of the present disclosure will become more apparent by reading the detailed description of non-limiting embodiments with reference to the following drawings:

[0047] Figure 1 It is a flowchart of a method for identifying postpartum hemorrhage based on a deep neural network according to an exemplary embodiment.

[0048] Figure 2 It is a schematic diagram of the structure of a pre-trained deep neural network model according to an exemplary embodiment.

[0049] Figure 3 It is a schematic diagram showing pixel area mapping of the bleeding regions corresponding to the multiple blood gradients according to an exemplary embodiment.

[0050] Figure 4The figure is a schematic diagram of a confusion matrix of a postpartum hemorrhage volume recognition result according to an exemplary embodiment.

[0051] Figure 5 It is a block diagram of a postpartum hemorrhage volume identification system based on a deep neural network according to an exemplary embodiment. DETAILED DESCRIPTION

[0052] The present disclosure is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present disclosure, but are not intended to limit the present disclosure in any form. It should be noted that, for those of ordinary skill in the art, several variations and improvements may be made without departing from the concept of the present disclosure. These all fall within the scope of protection of the present disclosure.

[0053] Figure 1 It is a flowchart of a method for identifying postpartum hemorrhage based on a deep neural network according to an exemplary embodiment.

[0054] like Figure 1 As shown, the present disclosure provides a method for identifying postpartum hemorrhage based on a deep neural network, including S11 to S14.

[0055] S11, acquiring an image containing multiple blood gradients.

[0056] Among them, an image acquisition device can be used to photograph a dressing containing multiple blood gradients to obtain an image containing multiple blood gradients. The image containing multiple blood gradients is an image of a dressing containing different blood concentrations, such as an image of gauze containing different blood concentrations.

[0057] The image acquisition device uses a high-resolution camera device, and the image acquisition process follows a standardized operating procedure, wherein the standardized operating procedure is a blood sample collection procedure prescribed by medicine, which will not be repeated here, in order to ensure the clarity and consistency of the acquired images.

[0058] S12, preprocessing the image containing multiple blood gradients to determine the preprocessed image.

[0059] Among them, preprocessing can include image resolution processing, normalization operation processing and noise reduction processing. Image resolution processing can be achieved through cropping operations and edge detection operations to enhance images containing multiple blood gradients and adapt to the input requirements of subsequent pre-trained deep neural network models.

[0060] S13, inputting the preprocessed image into a pre-trained deep neural network model to determine the predicted bleeding volume of the image containing multiple blood gradients.

[0061] S14, post-processing the predicted bleeding volume of the image containing multiple blood gradients according to the global information of the image containing multiple blood gradients, and outputting the total predicted bleeding volume corresponding to the image containing multiple blood gradients and the blood pixel distribution in the bleeding area.

[0062] The global information of the image containing multiple blood gradients represents the proportion of each blood gradient.

[0063] The post-processing step is implemented using a U-Net residual neural network, which combines the global information and local features of images containing multiple blood gradients to comprehensively analyze the predicted bleeding volume in the bleeding area of ​​each blood gradient to ensure the robustness of the bleeding volume prediction results.

[0064] Through the above technical solution, a pre-trained deep neural network model is introduced to extract and analyze the features of the dressing image, effectively segment the bleeding areas of the dressing image with multiple blood gradient images, and predict the amount of bleeding in the bleeding areas with different blood gradients. Combined with the global information of the image containing multiple blood gradients, the total postpartum bleeding volume can be evaluated in real time, dynamically and accurately, which is convenient for real-time bleeding volume evaluation in the clinical environment, improves the accuracy of postpartum bleeding volume identification, provides auxiliary decision support for clinical midwives, and improves the response speed and accuracy of postpartum hemorrhage treatment.

[0065] Figure 2 It is a schematic diagram of the structure of a pre-trained deep neural network model according to an exemplary embodiment.

[0066] like Figure 2 FIG. 1 shows a pre-trained deep neural network model, wherein the left side of the dotted line represents an image feature extractor of a U-Net structure, which is used to extract information such as images of gauze and blood stains of different color depths. The refinement and improvement stage on the right side represents the calculation of the bleeding volume value using an improved network structure based on the information extracted by the image feature extractor of the U-Net structure, and the refined and improved network structure includes a convolutional network and a fully connected network.

[0067] Specifically, Figure 2 As shown, the pre-trained deep neural network model of the present disclosure can adopt a pre-trained encoder-decoder model based on a U-net structure, that is, a pre-trained U-net model. The model adopts an Encoder / Decoder structure based on U-Net, including an encoder, an intermediate feature layer, and a decoder.

[0068] Among them, Encoder: The encoder consists of multiple convolutional layers and pooling layers, which are used to gradually reduce the spatial resolution of the input image. A nonlinear activation function (such as ReLU) is used after each convolutional layer to introduce nonlinear properties. Each pooling layer reduces the size of the image and extracts the feature information of the image at the same time, with the goal of capturing the low-level features of the input image.

[0069] Intermediate feature layer: An intermediate feature layer is set between the encoder and decoder to retain high-level semantic information. The intermediate feature layer consists of multiple convolutional layers to extract high-level feature representations of the image.

[0070] Decoder: The decoder part is set symmetrically with the encoder. The decoder contains upsampling operation layers and deconvolution layers, which are used to gradually restore high-level feature representations to segmentation maps with the same resolution as the input image. At the same time, a jump connection is set in the U-Net structure to connect a certain layer of the encoder with the corresponding layer of the decoder to help transfer low-level and high-level feature information. These connections help improve the accuracy of the segmentation model, enabling the model to capture local features and global semantic information at the same time. Finally, a convolution layer is set at the last layer of the decoder, and the number of channels in this layer is equal to the number of segmentation categories. This layer is used to generate the segmentation mask, in which each pixel is classified into the corresponding category, namely the dressing area and the bloodstain area at different depths, so as to determine the bleeding area and bleeding area features.

[0071] Then, an improved network structure with a convolutional network and a fully connected network is used to calculate the predicted bleeding amount, i.e., the bleeding value, of an image containing multiple blood gradients.

[0072] The input of the pre-trained deep neural network model disclosed in the present invention is a pre-processed image, and the output is the total predicted bleeding volume and the blood pixel distribution in the bleeding area corresponding to the image containing multiple blood gradients. The pre-trained deep neural network model is used to accurately segment the bloodstain area in complex image scenes and accurately predict the total bleeding volume.

[0073] In a possible embodiment, S11, acquiring an image containing multiple blood gradients, may include:

[0074] An image acquisition device is used to acquire images containing multiple blood gradients from a dressing containing multiple blood gradients.

[0075] Among them, multiple blood gradients represent different blood concentrations, and the same dressing can contain areas with multiple blood concentrations.

[0076] As an example, high-resolution imaging equipment is used to follow standardized operating procedures to photograph gauze containing various blood gradients and obtain gauze images containing various blood concentrations to ensure image clarity and consistency.

[0077] In a possible embodiment, S12, preprocessing the image containing multiple blood gradients to determine the preprocessed image may include:

[0078] The image containing multiple blood gradients is subjected to image resolution adjustment operation, image normalization operation, cropping operation, noise reduction operation and edge detection operation to determine a preprocessed image.

[0079] Among them, the image normalization operation is used to implement image normalization processing, the image resolution adjustment operation, the cropping operation, and the edge detection operation are used to implement image resolution processing, and the noise reduction operation is used to implement noise reduction processing.

[0080] In the present disclosure, the resolution of the pre-processed image is 224×224 pixels.

[0081] By preprocessing images containing multiple blood gradients, the images are enhanced to improve the recognition accuracy of the subsequent pre-trained deep neural network model for the bleeding area in the image.

[0082] In a possible embodiment, a pre-trained deep neural network model is obtained:

[0083] The dressing images were collected through standardized operation procedures, and the total amount of blood was marked on the dressing images to build a dressing image database containing different bleeding volume gradients;

[0084] A dressing image database containing different bleeding volume gradients and a preset deep neural network model, such as a preset U-net model, are used to perform model training, a pre-trained deep neural network model, such as a pre-trained U-net model.

[0085] During the model training process, the parameters of the deep neural network model are optimized through multiple iterations, and then the pre-trained deep neural network is fine-tuned using actual clinical data to improve the prediction accuracy and generalization ability of the pre-trained deep neural network.

[0086] The deep neural network model disclosed in the present invention may also adopt other types of deep neural network models.

[0087] In a possible embodiment, S13 may include S21 to S23.

[0088] S21, performing image segmentation processing on the pre-processed image to determine the bleeding area containing multiple blood gradients and the bleeding area features.

[0089] The bleeding area characteristics include the area, color depth and shape of the bleeding area.

[0090] As an example, a pre-trained Unet network may be used to perform image segmentation processing on the pre-processed image, extract bleeding areas with different blood gradients, and obtain bleeding area features corresponding to each bleeding area.

[0091] The pre-trained UNet network has been pre-trained on a dressing image database containing different bleeding gradients. It can automatically learn the bleeding area features in the pre-processed images and classify the bleeding areas based on the area, color depth and morphology of the bleeding areas.

[0092] S22, determining a bleeding volume characteristic of a bleeding area corresponding to each blood gradient according to the bleeding areas and bleeding area characteristics including a plurality of blood gradients.

[0093] Among them, a pre-trained deep neural network model can be used to perform pixel area mapping on the bleeding area containing multiple blood gradients, determine the pixel area ratio of the bleeding area corresponding to each blood gradient, and then determine the bleeding volume characteristics of the bleeding area corresponding to each blood gradient based on the color depth information of the bleeding area corresponding to each blood gradient.

[0094] S23, using a preset bleeding volume weight prediction model to perform weighted sum processing on the bleeding volume features of the bleeding area corresponding to each blood gradient, to determine the predicted bleeding volume of the image containing multiple blood gradients.

[0095] In the present disclosure, the pre-trained deep neural network model includes a preset bleeding volume weight prediction model. During the pre-training process of the pre-trained deep neural network model, the preset bleeding volume weight prediction model learns the weight prediction of the bleeding area and the bleeding volume prediction ability.

[0096] The preset bleeding volume weight prediction model is used to predict the weight of the bleeding area corresponding to each blood gradient and predict the predicted bleeding volume of an image containing multiple blood gradients. The input of the preset bleeding volume weight prediction model is the pixel area ratio of the bleeding area corresponding to each blood gradient, and the output is the predicted bleeding volume of the image containing multiple blood gradients.

[0097] In a possible embodiment, S22 may include S201 to S202.

[0098] S201, performing pixel area mapping on the bleeding area including multiple blood gradients, and determining the pixel area ratio of the bleeding area corresponding to each blood gradient.

[0099] The bleeding volume pixel area mapping model may be used to perform pixel area mapping on the bleeding area containing multiple blood gradients.

[0100] Figure 3It is a schematic diagram showing pixel area mapping of bleeding regions corresponding to multiple blood gradients according to an exemplary embodiment.

[0101] like Figure 3 As shown in the figure, the four pictures above are randomly selected from the test set and are displayed through manually marked label images, that is, the areas of 4 bloodstain depths, from left to right, are level 1 to level 4. The four pictures below are the bloodstain areas of different depths predicted by the bleeding pixel area mapping model.

[0102] Specifically, color segmentation is performed on the bleeding area containing multiple blood gradients to determine the number of pixels in the bleeding area corresponding to each blood gradient; and according to the number of pixels in the bleeding area corresponding to each blood gradient, the pixel area ratio of the bleeding area corresponding to each blood gradient is determined.

[0103] Among them, color segmentation can be performed based on the color depth information of the bleeding area of ​​multiple blood gradients, and the number of pixels in the bleeding area corresponding to each blood gradient after segmentation can be counted to determine the pixel area ratio of the bleeding area corresponding to each blood gradient. By counting the pixel area ratio of the bleeding area corresponding to each blood gradient, the bleeding volume ratio of the bleeding area of ​​blood gradients of different depths can be determined.

[0104] S202, determining the bleeding volume characteristics of the bleeding area corresponding to each blood gradient according to the pixel area ratio of the bleeding area corresponding to each blood gradient and the bleeding area characteristics of the bleeding area corresponding to each blood gradient.

[0105] The bleeding area feature of the bleeding area corresponding to each blood gradient adopts color depth information. The bleeding volume feature of the bleeding area corresponding to each blood gradient represents the blood content of the bleeding area corresponding to each blood gradient.

[0106] The bleeding volume characteristics are determined based on the pixel area ratio of the bleeding area corresponding to each blood gradient and the bleeding area characteristics of the bleeding area corresponding to each blood gradient. A refined and improved network is used to accurately quantify the bleeding volume of bloodstains at different depths by using a convolutional network and a fully connected network. The mask images of each blood gradient and gauze area are input, and the bleeding volume value is output.

[0107] In a possible embodiment, S23 may include S231 to S232.

[0108] S231, inputting the pixel area ratio of the bleeding area corresponding to each blood gradient into a preset bleeding volume weight prediction model to determine the weight of the bleeding area corresponding to each blood gradient.

[0109] The preset bleeding volume weight prediction module may be used to assign a weight to the bleeding area corresponding to each blood gradient according to the pixel area ratio of the bleeding area corresponding to each blood gradient.

[0110] S232, performing weighted sum processing on the bleeding volume feature of the bleeding area corresponding to each blood gradient and the weight of the bleeding area corresponding to each blood gradient, to determine the predicted bleeding volume of the image containing multiple blood gradients.

[0111] In the present disclosure, a preset bleeding volume weight prediction model in a pre-trained deep neural network model is used to perform weighted summation on the bleeding volume features of bleeding areas with different blood gradients, summarize the bleeding volume features of each bleeding area on an image containing multiple blood gradients, determine the predicted bleeding volume of the image containing multiple blood gradients, and ensure the accuracy of bleeding volume prediction by weighting the feature distribution and actual clinical data.

[0112] Based on the above steps S201 to S202 and steps S231 to S232, specifically, the amount of bleeding of bloodstains of different colors and depths on the gauze can be accurately quantified mainly through the following three steps:

[0113] ①Pixel area mapping: used to predict the proportion of bloodstains of various color depths to the gauze area, providing the main information for subsequent bleeding volume prediction.

[0114] First, the number of pixels in the gauze area and the bloodstain area of ​​each color depth is calculated respectively, and the proportion of the pixel area of ​​the bloodstain area of ​​each color depth to the entire gauze area is obtained. Secondly, a multi-layer perceptron is designed to input the pixel area proportion information of the bloodstain area of ​​each color depth into the neural network to obtain the bloodstain proportion information of different color depths.

[0115] ② Bleeding volume weight prediction: It is used to predict the actual bleeding volume of bloodstains in various areas on the gauze, and generate bleeding volume weights that ultimately act on the bleeding volume information.

[0116] A blood volume weight prediction model is designed, which includes a bloodstain feature extraction network. The main structure of the bloodstain feature extraction network consists of 4 residual blocks, each of which contains multiple convolutional layers. These residual blocks include different depths. Convolutional layer 1: includes 3×3 convolution operations, and the number of convolution kernels varies according to the depth of the residual block; Convolutional layer 2: includes 3×3 convolution operations, and the number of convolution kernels is the same as that of convolutional layer 1; Residual connection: The output of convolutional layer 2 is added to the input of the residual block to form a residual connection.

[0117] ③ Bleeding volume prediction: It is used to integrate the outputs of the above two steps and predict the bleeding volume based on the pixel area ratio information of the bloodstains in the bleeding area and the bleeding volume weight information of the bleeding area.

[0118] First, the weight information of the amount of bleeding is input into the convolutional network and the weight features are extracted through the Encoder structure, and the one-dimensional weight features are obtained through the fully connected network. Similarly, the pixel area ratio information of the bloodstain is input into the fully connected network to obtain the one-dimensional features. Finally, the one-dimensional weight features and the one-dimensional features are spliced ​​and input into the bleeding amount prediction model set in the fully connected network to determine the bleeding amount prediction result of the gauze in the image.

[0119] In a possible embodiment, after the pre-trained deep neural network model outputs the total predicted bleeding volume corresponding to the image containing multiple blood gradients and the blood pixel distribution in the bleeding area, a diagnostic report can be generated based on the total predicted bleeding volume corresponding to the image containing multiple blood gradients and the blood pixel distribution in the bleeding area for reference by medical staff.

[0120] In a possible embodiment, a simulated gauze image dataset containing multiple blood gradients is used to test a method for identifying the amount of postpartum hemorrhage based on a deep neural network provided by the present disclosure, so as to test the effectiveness of the method for identifying the amount of postpartum hemorrhage based on a deep neural network provided by the present disclosure.

[0121] Among them, the simulated gauze image dataset containing various blood gradients includes simulations of bleeding volume in various real clinical scenarios, and each image has been manually annotated with the amount of bleeding as a standard reference.

[0122] Figure 4 The figure is a schematic diagram of a confusion matrix of a postpartum hemorrhage volume recognition result according to an exemplary embodiment.

[0123] Figure 4 The data on the diagonal of represents the ratio of the correct number of predictions to the actual number of predictions.

[0124] like Figure 4 As shown, a postpartum hemorrhage volume recognition method based on a deep neural network provided by the present invention can accurately identify bleeding areas of different color depths through a pre-trained deep neural network model, and calculate the precise bleeding volume of each bleeding area based on pixel area mapping. The experimental results show that this method exhibits high efficiency and accuracy under various test conditions. Especially when the amount of bleeding is large, the error of the U-Net neural network structure model is significantly reduced, and the accuracy meets the requirements of clinical applications.

[0125] During the test experiment, the pre-trained deep neural network model uses the U-Net neural network structure, which has a good performance in medical image segmentation tasks. The U-Net neural network structure is set with experimental conditions including multiple image resolutions and multiple network layers, and the accuracy of the pixel area mapping method is optimized. The performance of the U-Net neural network structure in complex scenarios such as different lighting conditions and blood concentration is tested, and the mean square error (MSE) and determination coefficient (R 2 ) is used as an evaluation indicator for the bleeding volume prediction accuracy of the U-Net neural network structure.

[0126] As shown in Table 1, the experiment also shows that the method provided by the present invention has the highest prediction accuracy, the smallest MSE value, and R 2 The value is closer to 1, indicating that the pre-trained deep neural network model of the method provided in the present disclosure can effectively fit the actual bleeding volume.

[0127] MSE <![CDATA[R 2 ]]> Convolutional infrastructure 0.641 0.418 YOLO Infrastructure 0.302 0.729 UNet Infrastructure 0.264 0.840 The method provided by the present invention 0.187 0.896

[0128] Table 1

[0129] Moreover, compared with the traditional manual measurement method and the identification method based on simple image processing algorithm, the postpartum hemorrhage amount identification method based on deep neural network provided by the present disclosure has significant advantages. Traditional methods often show instability when facing complex clinical environments (such as changes in lighting, mixing of blood and body fluids, etc.), while the pre-trained deep neural network model provided by the present disclosure has strong robustness and wide adaptability. The experimental results also show that the estimation accuracy of the amount of bleeding can be significantly improved by performing detailed segmentation and pixel area mapping of bloodstain areas of different color depths.

[0130] The present invention provides a method for identifying the amount of postpartum bleeding based on a deep neural network, which uses a pre-trained deep neural network to accurately segment the bleeding area in the dressing image, thereby improving the accuracy of the assessment of the amount of bleeding, especially in the case of a large amount of bleeding, reducing the error and improving the accuracy; the method can also complete large-scale data processing in a relatively short time, is suitable for real-time diagnosis needs in clinical scenarios, and has real-time performance; the pre-trained deep neural network model provided by the method can adapt to image processing under different lighting conditions and complex scenarios, maintain high accuracy and stability, and has a certain degree of robustness.

[0131] Through the above experimental verification and data analysis, the postpartum hemorrhage identification method based on deep neural network provided by the present disclosure has shown extremely high clinical application potential. Medical staff can use this method to quickly and accurately judge the postpartum hemorrhage situation, take timely intervention measures, and effectively reduce the health risks of parturient women.

[0132] Figure 5 It is a block diagram of a postpartum hemorrhage volume identification system based on a deep neural network according to an exemplary embodiment.

[0133] Based on the same concept, the present disclosure also provides a postpartum hemorrhage volume recognition system 100 based on a deep neural network, such as Figure 5 As shown, it includes: an image acquisition module 110, an image preprocessing module 120, a bleeding volume prediction module 130, and a total bleeding volume determination module 140.

[0134] An image acquisition module 110, for acquiring images containing multiple blood gradients;

[0135] An image preprocessing module 120, configured to preprocess the image containing the multiple blood gradients and determine the preprocessed image;

[0136] A bleeding volume prediction module 130, configured to input the preprocessed image into a pretrained deep neural network model to determine a predicted bleeding volume of the image containing multiple blood gradients;

[0137] The total bleeding volume determination module 140 is used to post-process the predicted bleeding volume of the image containing multiple blood gradients according to the global information of the image containing multiple blood gradients, and output the total predicted bleeding volume and blood pixel distribution in the bleeding area corresponding to the image containing multiple blood gradients, wherein the global information of the image containing multiple blood gradients represents the proportion of each blood gradient.

[0138] Through the above technical solution, a pre-trained deep neural network model is introduced to extract and analyze the features of the dressing image, effectively segment the bleeding areas of the dressing image in various blood gradient images, and predict the bleeding amount in the bleeding areas with different blood gradients. Combined with the global information of the image containing multiple blood gradients, the total postpartum bleeding volume can be evaluated in real time, dynamically and accurately, which is convenient for providing real-time bleeding volume evaluation in the clinical environment, improving the accuracy of postpartum bleeding volume identification, providing auxiliary decision support for clinical midwives, and improving the response speed and accuracy of postpartum hemorrhage treatment.

[0139] Regarding the embodiment of the above system, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0140] Based on the same concept as above, in another embodiment of the present disclosure, there is also provided an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, and the processor is used to execute a method for identifying postpartum bleeding volume based on a deep neural network when executing the program.

[0141] Optionally, the memory is used to store programs; the memory may include volatile memory (English: volatile memory), such as random-access memory (English: random-access memory, abbreviated: RAM), such as static random-access memory (English: static random-access memory, abbreviated: SRAM), double data rate synchronous dynamic random access memory (English: Double Data Rate Synchronous Dynamic Random Access Memory, abbreviated: DDR SDRAM), etc.; the memory may also include non-volatile memory (English: non-volatile memory), such as flash memory (English: flash memory). The memory is used to store computer programs (such as applications, functional modules, etc. that implement the above method), computer instructions, etc., and the above computer programs, computer instructions, etc. can be partitioned and stored in one or more memories. And the above computer programs, computer instructions, data, etc. can be called by the processor.

[0142] The above-mentioned computer programs, computer instructions, etc. may be stored in one or more memories in partitions, and the above-mentioned computer programs, computer instructions, data, etc. may be called by a processor.

[0143] The processor is used to execute the computer program stored in the memory to implement the various steps in the method involved in the above embodiment. For details, please refer to the relevant description in the above method embodiment.

[0144] The processor and the memory may be independent structures or integrated structures. When the processor and the memory are independent structures, the memory and the processor may be coupled and connected via a bus.

[0145] In an embodiment of the present disclosure, a non-temporary computer-readable storage medium is also provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of a method for identifying postpartum bleeding volume based on a deep neural network in any of the above embodiments are implemented.

[0146] Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, systems, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0147] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0148] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0149] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0150] Although the preferred embodiments of the present disclosure have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present disclosure.

[0151] Obviously, those skilled in the art can make various changes and modifications to the present disclosure without departing from the spirit and scope of the present disclosure. Thus, if these modifications and variations of the present disclosure fall within the scope of the claims of the present disclosure and their equivalents, the present disclosure is also intended to include these modifications and variations.

Claims

1. A method and system for identifying postpartum hemorrhage based on deep neural network, characterized in that: include: Acquire images containing multiple blood gradients; Preprocessing the image containing the multiple blood gradients to determine a preprocessed image; Inputting the preprocessed image into a pretrained deep neural network model to determine the predicted bleeding volume of the image containing multiple blood gradients; The predicted bleeding volume of the image containing multiple blood gradients is post-processed according to the global information of the image containing multiple blood gradients, and the total predicted bleeding volume corresponding to the image containing multiple blood gradients and the blood pixel distribution in the bleeding area are output, wherein the global information of the image containing multiple blood gradients represents the proportion of each blood gradient.

2. The method according to claim 1, characterized in that The pre-trained deep neural network model includes a preset bleeding volume weight prediction model; Inputting the preprocessed image into a pretrained deep neural network model to determine the predicted bleeding volume of the image containing multiple blood gradients includes: Performing image segmentation processing on the pre-processed image to determine a bleeding area containing multiple blood gradients and bleeding area features, wherein the bleeding area features include the area, color depth, and morphology of the bleeding area; Determining a bleeding volume characteristic of the bleeding area corresponding to each blood gradient according to the bleeding area including the multiple blood gradients and the bleeding area characteristics; The preset bleeding volume weight prediction model is used to perform weighted sum processing on the bleeding volume characteristics of the bleeding area corresponding to each blood gradient to determine the predicted bleeding volume of the image containing multiple blood gradients.

3. The method according to claim 2, characterized in that The step of determining the bleeding volume feature of the bleeding area corresponding to each blood gradient according to the bleeding area including the multiple blood gradients and the bleeding area feature comprises: Performing pixel area mapping on the bleeding area containing multiple blood gradients to determine the pixel area ratio of the bleeding area corresponding to each blood gradient; The bleeding volume feature of the bleeding area corresponding to each blood gradient is determined according to the pixel area ratio of the bleeding area corresponding to each blood gradient and the bleeding area feature of the bleeding area corresponding to each blood gradient.

4. The method according to claim 3, characterized in that: The performing pixel area mapping on the bleeding area including the multiple blood gradients to determine the pixel area ratio of the bleeding area corresponding to each blood gradient includes: Performing color segmentation on the bleeding area containing multiple blood gradients to determine the number of pixels in the bleeding area corresponding to each blood gradient; According to the number of pixels in the bleeding area corresponding to each blood gradient, the pixel area ratio of the bleeding area corresponding to each blood gradient is determined.

5. The method according to claim 3, characterized in that: The method of using the preset bleeding volume weight prediction model to perform weighted sum processing on the bleeding volume characteristics of the bleeding area corresponding to each blood gradient to determine the predicted bleeding volume of the image containing multiple blood gradients includes: Inputting the pixel area ratio of the bleeding area corresponding to each blood gradient into the preset bleeding volume weight prediction model to determine the weight of the bleeding area corresponding to each blood gradient; The bleeding volume feature of the bleeding area corresponding to each blood gradient and the weight of the bleeding area corresponding to each blood gradient are weighted summed to determine the predicted bleeding volume of the image containing multiple blood gradients.

6. The method according to claim 1, characterized in that The acquisition of images containing multiple blood gradients includes: An image acquisition device is used to acquire images containing multiple blood gradients from a dressing containing multiple blood gradients.

7. The method according to claim 1, characterized in that The preprocessing of the image containing the multiple blood gradients to determine the preprocessed image includes: The image containing the multiple blood gradients is subjected to image resolution adjustment, image normalization, cropping, noise reduction and edge detection to determine the preprocessed image.

8. A postpartum hemorrhage volume recognition system based on deep neural network, characterized in that: include: An image acquisition module, used for acquiring images containing multiple blood gradients; An image preprocessing module, used for preprocessing the image containing multiple blood gradients to determine the preprocessed image; A bleeding volume prediction module, used for inputting the preprocessed image into a pretrained deep neural network model to determine the predicted bleeding volume of the image containing multiple blood gradients; The total bleeding volume determination module is used to post-process the predicted bleeding volume of the image containing multiple blood gradients according to the global information of the image containing multiple blood gradients, and output the total predicted bleeding volume and blood pixel distribution in the bleeding area corresponding to the image containing multiple blood gradients, wherein the global information of the image containing multiple blood gradients represents the proportion of each blood gradient.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 7 are implemented.

10. An electronic device, characterized in that: include: a memory having a computer program stored thereon; A processor, configured to execute the computer program in the memory to implement the steps of the method according to any one of claims 1 to 7.

Images