Method for constructing postpartum hemorrhage amount recognition database

Abstract

The invention provides a postpartum hemorrhage amount identification system database construction method, which comprises the following steps: collecting a postpartum hemorrhage blood sample according to a preset standardized operation process; preparing simulated bleeding dressings with different bleeding amount gradients based on the blood sample; and acquiring images of all the simulated bleeding dressings, and establishing a simulated dressing image database containing different bleeding amount gradients. And later data processing can be further carried out based on the database. The database constructed by the method can be used for realizing the construction of a standardized and processized postpartum hemorrhage amount recognition algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of medical monitoring and evaluation, and in particular to a method for constructing a postpartum hemorrhage volume identification database, and in particular to postpartum hemorrhage volume evaluation and management based on multi-dimensional dynamic monitoring. Background Art

[0002] Postpartum hemorrhage (PPH) is one of the important causes of maternal mortality and complications, especially in the short period after delivery. The identification and management of postpartum hemorrhage is crucial to reducing maternal and infant mortality. The key to preventing and controlling PPH lies in early identification and diagnosis. The most intuitive data is the amount of bleeding, which is also an important indicator of the severity of the disease. Therefore, the amount of postpartum bleeding should be measured and estimated in a timely and accurate manner. Underestimating the amount of bleeding will lose the opportunity for rescue, while overestimating the amount of bleeding will cause the loss of blood product resources and the occurrence of transfusion-related complications. Using a more accurate method to dynamically evaluate the cumulative amount of bleeding in parturients since the delivery of the fetus will help to respond in time and avoid the occurrence of adverse outcomes.

[0003] At present, the methods for evaluating the amount of postpartum blood loss in clinical practice can be roughly divided into two categories: subjective estimation (Estimated Blood Loss, EBL) and quantitative calculation (Quantitative Blood Loss, QBL). EBL mainly refers to the visual estimation of blood loss, while QBL includes objective measurement methods (weighing measurement and volume measurement), area measurement, laboratory index method, vital sign assessment method, shock index method, ultrasound measurement of inferior vena cava diameter, uterine three-dimensional ultrasound measurement, pulse carbon monoxide-blood oxygen saturation monitoring method, etc. However, the above traditional methods for evaluating blood loss have certain limitations, resulting in inaccurate and untimely evaluation of blood loss. Therefore, how to accurately, conveniently and in real time evaluate the amount of postpartum blood loss has always been a clinical problem that needs to be solved urgently in midwifery work.

[0004] Artificial Intelligence (AI) can now perform complex tasks such as object recognition and prediction. ML is a branch of AI technology. It is based on the cross-application of multiple disciplines (such as probability theory, statistics, etc.) and specializes in the study of computer simulation of human learning ability. At present, it has achieved fruitful results in medical imaging, neurosurgery, dentistry, ophthalmology, orthopedics, etc. There are studies that apply ML technology to the field of PPH, but focus on risk assessment rather than quantitative calculation of bleeding volume. In the field of bleeding volume assessment, some studies and commercial applications have begun to use image processing technology to automatically identify bleeding volume during surgical operations. However, since 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, gauze, etc.), more empirical research is needed to detect the applicability of such models in the obstetric field, so as to truly realize the implementation of the technology.

[0005] In summary, early identification and treatment of PPH is the top priority of clinical midwifery work, and accurate identification of bleeding volume is the key to early warning diagnosis. Therefore, it is urgent to develop a scientific and standardized process to build a postpartum hemorrhage dressing identification database to achieve standardized and process-based identification of postpartum hemorrhage volume. Summary of the invention

[0006] In view of the above-mentioned deficiencies in the prior art, an object of the present invention is to provide a method for constructing a database of a postpartum hemorrhage volume identification system, which is based on the simulated bleeding dressing image database and can be specifically used for the identification and evaluation of postpartum hemorrhage volume.

[0007] The present invention provides a method for constructing a postpartum hemorrhage volume identification database, comprising:

[0008] Blood specimens for postpartum hemorrhage were collected according to the preset standardized operating procedures;

[0009] preparing simulated bleeding dressings with different bleeding volume gradients based on the blood specimen;

[0010] Images of all the simulated bleeding dressings are acquired, and a simulated dressing image database containing different bleeding volume gradients is established.

[0011] Optionally, the collection of blood from postpartum hemorrhage refers to the collection of placental blood from the third stage of labor.

[0012] Optionally, collecting a blood sample for postpartum hemorrhage includes:

[0013] After the placenta is delivered in the third stage of labor and the examination shows no abnormalities, blood is drawn from the umbilical vein at a set volume of one unit until no more blood can be obtained.

[0014] Optionally, before collecting the blood sample of postpartum hemorrhage, the method further comprises:

[0015] Determine standardized operating procedures for simulated bleeding dressing images;

[0016] In the standardized operating procedure, the operating procedures and requirements for collecting blood samples for postpartum hemorrhage and preparing simulated bleeding dressings with different bleeding volume gradients are stipulated.

[0017] Optionally, acquiring images of all the simulated bleeding dressings to establish a simulated dressing image database containing different bleeding gradients includes:

[0018] Using an image acquisition device and following the standardized operating procedure to acquire an image of the simulated bleeding dressing;

[0019] renaming the image of the simulated bleeding dressing according to the blood volume corresponding to each bleeding volume gradient, and saving the named image of the simulated bleeding dressing;

[0020] All stored images of the simulated bleeding dressings containing different bleeding volume gradients constitute a preliminary postpartum bleeding volume recognition database.

[0021] Optionally, the acquiring the image of the simulated bleeding dressing using an image acquisition device and following the standardized operating procedure includes:

[0022] Using the image acquisition device, obtaining an image of the simulated bleeding dressing;

[0023] The image of the simulated bleeding dressing is checked, and if it does not meet the set requirements, it is re-photographed.

[0024] Optionally, the failure to meet the set requirements includes one or more of the following situations:

[0025] There is blood residue in the syringe;

[0026] Blood sprays outside the dressing;

[0027] The background color and lighting of the shooting are inconsistent;

[0028] The four corners of the dressing are not all shown in the picture;

[0029] The auxiliary materials are wrinkled;

[0030] The picture is blurry or fuzzy.

[0031] Optionally, the method for constructing a postpartum hemorrhage volume identification database further includes:

[0032] Data enhancement operation is performed on the image of the simulated bleeding dressing to expand the sample size in the database, thereby obtaining a final postpartum hemorrhage volume recognition database for use in training a postpartum hemorrhage volume recognition model.

[0033] Optionally, the data enhancement operation includes:

[0034] manually segmenting and labeling irregular bleeding areas in the image of the simulated bleeding dressing;

[0035] After randomly shuffling the order of the labeled images, they are randomly divided into training set and test set databases according to the set ratio;

[0036] The flipping operation in the geometric transformation method is used to augment the images of the training set and the test set database.

[0037] Optionally, manually segmenting and labeling the irregular bleeding area in the image of the simulated bleeding dressing includes:

[0038] Using the polygon segmentation method of Labelme, the irregular bleeding areas in all the images of the simulated bleeding dressings are segmented, and different color areas are marked accurately to the pixel point;

[0039] The segmentation and annotation are completed in the Anaconda environment, and the annotated images are output in JSON format.

[0040] Compared with the prior art, the present invention has at least one of the following beneficial effects:

[0041] The method for constructing a database for identifying the amount of postpartum bleeding provided by the present invention collects placental blood in the third stage of labor and prepares simulated bleeding dressings with different bleeding volume gradients. The ultimately formed postpartum bleeding volume identification system can truly serve the clinic so as to subsequently realize the construction of a standardized and process-based postpartum bleeding volume identification algorithm.

[0042] The above-mentioned postpartum bleeding volume identification database construction method of the present invention pre-establishes standardized operating procedures, and all operations are performed in accordance with the standardized operating procedures to ensure rigor, standardization and operability. The established simulated dressing image database containing different bleeding volume gradients provides support for the subsequent construction of a postpartum bleeding volume identification algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0044] Figure 1 A schematic diagram of a flow chart of an image database construction method according to an embodiment of the present invention;

[0045] Figure 2 A schematic diagram of a flow chart of an image database construction method according to a preferred embodiment of the present invention;

[0046] Figure 3 A flow chart of the standardized operation of blood sample collection provided by an embodiment of the present invention;

[0047] Figure 4 Establishing a standardized operation flow chart for image acquisition and image database provided by the embodiment of the present invention;

[0048] Figure 5 This is an example of annotated image provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The specific embodiments of the present invention will be described below in conjunction with the accompanying drawings. It should be noted that in the specific description of these embodiments, for the sake of concise description, this specification cannot provide a detailed description of all the features of the actual embodiments. Without departing from the spirit and scope of the present invention, those skilled in the art may modify and replace the embodiments of the present invention, and the resulting embodiments are also within the scope of protection of the present invention.

[0050] Reference Figure 1 As shown, in one embodiment of the present invention, a method for constructing a postpartum hemorrhage volume identification database is provided, comprising:

[0051] S100, blood samples for postpartum hemorrhage were collected according to the preset standardized operating procedures;

[0052] In this step, collecting a blood sample for postpartum hemorrhage refers to collecting placental blood in the third stage of labor.

[0053] The third stage of labor refers to the period from birth to the delivery of the placenta and fetal membranes. After the fetus is delivered, the uterus suddenly becomes smaller, and the placenta cannot shrink accordingly, so it is dislocated from the uterine wall and detached, forming retroplacental hematoma. The uterus continues to contract, increasing the detachment area, so that the placenta is completely detached and delivered. This process usually lasts 5 to 15 minutes, and no longer than 30 minutes.

[0054] The third stage of labor is one of the main periods of postpartum hemorrhage. After the placenta is separated, the uterus needs to contract to compress the blood vessels and block the lumen to stop bleeding. If the uterus is weak in contraction, it will cause immediate and massive blood loss, endangering the life of the mother. Therefore, in the third stage of labor, medical staff need to closely observe the mother's uterine contraction and postpartum bleeding, and take hemostatic measures in time.

[0055] In the embodiment of the present invention, based on the above delivery, after the placenta is delivered in the third stage of labor (the placenta is separated from the human body), the doctor checks that there is no abnormality in the placenta and then draws blood from the placenta through the umbilical vein.

[0056] In this step, the preset standardized operating procedure is a standardized operating procedure produced after literature search and expert consultation, which is rigorous, standardized and operational.

[0057] S200, simulated bleeding dressings with different bleeding gradients prepared based on blood specimens;

[0058] In this step, because different bleeding volumes need to be identified, the database must have corresponding simulated bleeding dressings with different bleeding volumes when it is established. Therefore, different bleeding volume gradients can be set first according to the actual situation, and then different bleeding volume gradients can be set. The specific bleeding volume gradient can be set according to the accuracy / requirements of subsequent bleeding volume identification, such as a gradient of 20ml or a gradient of 30ml.

[0059] Specifically, the setting of different bleeding volume gradients may be related to the maximum amount of blood that the dressing can absorb, for example, the dressing may be divided into 10 equal parts according to the maximum absorption capacity, and set from small to large. Of course, in other embodiments, other bleeding volume gradients may also be set, and are not limited to this example.

[0060] Similarly, dressing preparation is also required to follow standardized operating procedures to avoid adverse effects caused by non-compliant operations.

[0061] In this embodiment, the above-mentioned dressings generally refer to medical sanitary materials and dressings, including cloths that can absorb blood, such as gauze, sterile towels, middle sheets, maternity pads, sick clothes, etc. When establishing a database, one dressing needs to establish one database, and different dressings need to establish different databases.

[0062] S300, acquiring images of all simulated bleeding dressings, and establishing a simulated dressing image database containing different bleeding volume gradients.

[0063] In this step, the images of all simulated bleeding dressings can be obtained by taking photos with an image acquisition device to obtain simulated dressing images with different blood gradients. The images are exported to a memory (such as a hard disk), renamed according to the blood volume (corresponding to different bleeding gradients), and saved in a folder with the corresponding name.

[0064] Of course, in some embodiments, in order to provide conditions for the accuracy of subsequent bleeding volume identification, the image acquisition device is preferably a high-resolution camera device, and the image acquisition process follows a standardized operating procedure to ensure image consistency and high quality. The image quality needs to be checked immediately after shooting. If it is blurred or does not meet the requirements, it needs to be re-shot immediately. For example, if there is blood residue in the syringe, blood is sprayed outside the dressing, the shooting background color and lighting are inconsistent, the four corners of the dressing are not fully presented in the picture, the dressing has wrinkles, the image is blurred, etc., all these situations need to be re-shot.

[0065] In addition, in order to better identify the amount of postpartum hemorrhage, the database construction method in the above embodiment should further include post-processing. Figure 2 As shown, the post-processing may be: performing data augmentation operations on images of simulated bleeding dressings to expand the sample size in the database for training a postpartum hemorrhage volume recognition model.

[0066] In some preferred embodiments, the data enhancement operation includes: manually segmenting and labeling the irregular bleeding areas in the image of the simulated bleeding dressing; randomly shuffling the labeled images, and randomly dividing them into training set and test set databases according to a set ratio; using the flip operation in the geometric change method to amplify the image data of the training set and test set databases. Through this data enhancement operation, the image data in the database can be greatly expanded, and more training sets and test sets can be obtained, which is convenient for the accuracy of the subsequent postpartum hemorrhage recognition model. For example, in a specific embodiment, after randomly shuffling the images, they can be randomly divided into training set and test set databases according to a ratio of 9:1. Of course, in other embodiments, other ratios can also be used.

[0067] In addition, in some other embodiments, the irregular bleeding areas in the image of the simulated bleeding dressing are manually segmented and labeled, including: using the polygon segmentation method of Labelme to segment the irregular bleeding areas in all the images of the simulated bleeding dressing, and labeling the different color areas accurately to the pixel point. Through this segmentation and labeling method, the subsequent training and recognition accuracy and precision can be higher. The specific examples of the labeled images can be referred to Figure 5 shown.

[0068] In a specific embodiment, the above segmentation and annotation are completed in the Anaconda environment, and the annotated image is output in JSON format, which is easy for humans to read and write, and also easy for machines to parse and generate.

[0069] In order to ensure that the above-mentioned blood collection, image production and other operations are carried out in accordance with standardized operating procedures, and to ensure rigor, standardization and operability, the preset standardized operating procedures are also an important part. The specific standardized operating procedures can be formulated after literature retrieval and expert consultation and discussion, and they can meet the requirements of medical operations and database establishment.

[0070] In a specific embodiment, referring to Figure 3 The figure shows the standardized operation flow chart for blood specimen collection, which specifies the specific materials and operation steps for sampling. The materials are prepared according to the general situation, including but not limited to: surgical isolation gowns, disposable sterile surgical gloves, 20mL syringes, placenta sealing bags, label paper, and signature pens.

[0071] After preparing the above materials, wait for the placenta to be delivered, place it on the delivery table, and start collecting blood samples. For details, please refer to the following:

[0072] (1) Check the integrity of the placenta and fetal membranes to determine whether there are any abnormalities. If there are any abnormalities, the research object, i.e. the abnormal placenta, will be eliminated. If there are no abnormalities, proceed to the following steps;

[0073] (2) Preparation: Wear surgical isolation gown and disposable sterile gloves;

[0074] (3) Positioning: Find the thick and thin-walled umbilical vein at the end of the umbilical cord;

[0075] (4) Blood collection: Draw blood from the umbilical vein using a 20ml syringe. When the blood level is flush with the 20ml mark, replace the syringe. Repeat this process several times until no more blood can be drawn, then replace the needle cap.

[0076] (5) Arrangement: Place the syringe containing the blood in a placenta sealing bag, label it, and record the collection time and number of syringes.

[0077] The above process is used to complete the collection of placental blood samples in the third stage of labor. Of course, during the above collection process, some standardized operations should be paid attention to, such as: standardizing and carefully putting the needle cap of the syringe back on to prevent needlestick injuries; blood can only be collected after the midwife has fully checked the placenta, fetal membranes and umbilical cord to find no abnormalities; the operator who collects the blood sample is familiar with the mechanism of normal delivery, the normal anatomical structure of placental tissue, and the blood collection process and requirements.

[0078] In a specific embodiment, referring to Figure 4The following is a standardized operation flow chart for obtaining images of all simulated bleeding dressings. First, prepare materials, including but not limited to: gauze (9cmx10cm), camera phone, 2mL syringe, green non-woven fabric (50cmx50cm), square wooden board (50cmx50cm), surgical shadowless lamp, ruler (50cm), disposable medical examination gloves, and hard disk.

[0079] To simulate the bleeding dressing image, you can follow the following process:

[0080] (1) Arrange the background

[0081] Turn on the shadowless light, place the green hardwood board set as the background in the center under the shadowless light, and fix the ruler vertically on one side of the hardwood board.

[0082] (2) Calibration blood

[0083] Wearing disposable medical examination gloves, take out the 20mL syringe and remove the needle, and use the needle of the 2mL syringe to draw 1mL of blood from the mastoid of the 20mL syringe.

[0084] (3) Preparation of gauze

[0085] Remove the needle of the 2mL syringe, slowly push the piston of the 2mL syringe, and stain the calibrated 1mL of blood on the gauze through the mastoid part. The following requirements apply: ① no blood should be attached to the wall of the syringe mastoid part; ② no blood should be splashed outside the gauze;

[0086] Fully unfold the prepared blood-stained gauze on the laid green background board and expose it under a shadowless lamp. The requirements are: the gauze should be flat, without wrinkles or overlaps.

[0087] (4) Image preparation

[0088] Use a camera phone with the flash off, fix the camera at 25±1cm as indicated by the ruler, and photograph the blood-stained gauze perpendicular to the plane of the gauze. The requirement is that all four corners of the gauze must be within the shooting screen.

[0089] (5) Review image

[0090] Check the image quality immediately after taking it, and retake it if it is blurry or does not meet the requirements for preparing the image.

[0091] (6) Rename

[0092] After verifying that the image is correct, use the renaming function of the camera phone to rename the blood volume calibrated with the gauze in the image. The path is: More-Rename-(X)mL-Confirm.

[0093] (7) End

[0094] The filming is completed after all the blood in the prepared 20mL syringe has been used up.

[0095] (8) Archive

[0096] The images obtained in this shooting were uploaded to the computer hard disk through the sharing function of the mobile phone, a new "1-10mL" folder was created in the hard disk, and the corresponding gauze images were classified and saved in the corresponding folder.

[0097] The above standardized operation process of blood sample collection and image acquisition is a detailed description of a specific embodiment. In other embodiments, other rulers, photographing equipment and other dressings may also be used, and are not limited to the above specific embodiments.

[0098] In order to better illustrate the technical solution of the present invention, a specific operation example is described in detail below, but it should be understood that the following example is only for a better understanding of the technology of the present invention and is not used to limit the scope of the present invention. In order to comply with relevant medical regulations, all the following operations in this embodiment, such as blood sampling, are performed after complying with the regulations.

[0099] Specifically, this embodiment provides a method for constructing a database of a postpartum bleeding volume identification system. Before constructing the database, a standardized operating procedure is first formulated. This standardized operating procedure can cover subsequent blood collection, dressing preparation for different bleeding volumes, image acquisition, etc.

[0100] Step 1: Collection of blood samples

[0101] (1) Collection of placental blood in the third stage of labor. Specifically, after the placenta is delivered in the third stage of labor, if the midwife checks that there is no abnormality, blood is drawn from the umbilical vein in units of 20 mL until no more blood can be obtained.

[0102] (2) After blood collection, put the needle cap on and put the syringe (with needle cap) into the placenta sealing bag and label it, recording the collection time and number of syringes.

[0103] Step 2: Preparation of the simulated bleeding dressing image

[0104] (1) The blood volume is calibrated according to the set gradient, and the simulated bleeding dressing is prepared according to the standardized operation process. In this embodiment, the blood gradient is set to 1-10 mL.

[0105] (2) Arrange the background. Turn on the shadowless light, place the green hardwood board as the background in the center under the shadowless light, and fix the ruler vertically on one side of the hardwood board.

[0106] (3) Wear disposable medical examination gloves and calibrate the blood volume. In this embodiment, a 20 mL syringe is taken out and the needle is removed, and 1 mL of blood is drawn from the mastoid of the 20 mL syringe using the needle of a 2 mL syringe.

[0107] (4) Spray all the blood on the dressing. In this embodiment, remove the needle of the 2 ml syringe, slowly push the piston of the 2 ml syringe to stain the calibrated 1 ml of blood on the gauze through the mastoid part, with the following requirements: ① no blood hanging on the wall of the syringe mastoid part; ② no blood splashing outside the gauze.

[0108] (5) Spread out the prepared blood-stained dressing completely on a green background board and expose it to a shadowless lamp. The requirements are: the gauze should be flat, wrinkle-free and without overlap.

[0109] (6) Use a camera, turn off the flash, fix the camera, and shoot perpendicular to the dressing surface. The requirement is that all four corners of the dressing should be within the shooting frame.

[0110] (7) Check the image quality immediately after shooting. If it is blurred or does not meet the requirements for image preparation, reshoot it immediately.

[0111] (8) The filming is terminated after all the blood in the prepared syringes has been used up.

[0112] Step 3: Archiving images

[0113] (1) Export the images to the hard disk, rename the images according to the blood volume, and save them in folders with corresponding names.

[0114] Step 4: Data preprocessing and database establishment

[0115] (1) Image annotation: In this embodiment, the pose key points in MSCOCO format are used to build the Anaconda environment, and the irregular bleeding area is manually segmented and annotated using the polygon segmentation method of Labelme.

[0116] In this embodiment, the specific operation process is as follows: Open folder (OpenDir) > Select image > Edit (Edit) > Create Polygons > Segment gauze border area > Define as 0 > Segment light area > Define as 1 > Segment dark area > Define as 2 > Segment area with blood coagulation > Define as 3 > Segment blank area in blood-covered area > Define as 4 > Annotation completed > Save (JSON format). The original image (JPG format) and the annotated image (JSON format) are saved in the folder named by the corresponding blood content.

[0117] (2) Dataset allocation: The samples were arranged in ascending order according to the blood gradient of the calibrated bleeding volume, randomly shuffled and divided into 10 equal subsets, and the dataset was randomly divided into a training set database and a test set database in a ratio of 9:1.

[0118] (3) Data enhancement: This embodiment uses the flip operation in the geometric transformation method to augment the data, including horizontal flip and vertical flip.

[0119] Implementation effect:

[0120] In order to verify the effectiveness of the method for constructing a database for a postpartum hemorrhage volume recognition system provided by the present invention, the database constructed by the above embodiment of the present invention was used to construct a postpartum hemorrhage volume recognition algorithm. The mean square error (MSE) and the coefficient of determination (R 2 ) is used to evaluate the prediction accuracy of the model for the amount of bleeding. The experimental results show that this method has shown high efficiency and accuracy under various test conditions. In addition, by performing detailed segmentation and pixel area mapping of bloodstain areas at different depths, the present invention can significantly improve the estimation accuracy of the amount of bleeding.

[0121] The construction method of the simulated bleeding dressing image database in the above-mentioned preferred embodiment of the present invention can be clear by consulting experts in the obstetric field and the field of computer vision, and the problems existing in the image production process (such as blood residue in the syringe, blood spraying outside the dressing, shooting background color and illumination inconsistency, dressing four corners not all present in the picture, auxiliary materials have wrinkles, image blur, etc.) are clarified through pre-experiments, and the standardized operation process of blood collection and image production is formulated respectively, and a simulated dressing image database containing different bleeding volume gradients is established. At the same time, by taking data enhancement operation to the original image, the sample size of model training is expanded, and a guarantee is provided for the optimization of model performance. Labelme software is used to mark all images in different color areas, which is accurate to pixel points, and the depth and breadth of the data set are measured. The image database constructed by the present invention is suitable for establishing a reliable and stable machine learning bleeding volume recognition system.

[0122] The above describes a specific embodiment of the present invention. It should be understood that although specific implementation methods are provided, those skilled in the art can make various modifications and improvements without departing from the core technical concept of the present invention, and these modifications and improvements still fall within the scope of protection of the present invention.

Claims

1. A method for constructing a postpartum hemorrhage volume identification database, characterized in that: include: Blood specimens for postpartum hemorrhage were collected according to the preset standardized operating procedures; preparing simulated bleeding dressings with different bleeding volume gradients based on the blood specimen; Images of all the simulated bleeding dressings are acquired, and a simulated dressing image database containing different bleeding volume gradients is established.

2. The method for constructing a postpartum bleeding volume identification database according to claim 1, characterized in that: The collecting of blood from postpartum hemorrhage refers to collecting placental blood from the third stage of labor.

3. The method for constructing a postpartum bleeding volume identification database according to claim 2, characterized in that: The blood sample collected for postpartum hemorrhage includes: After the placenta is delivered in the third stage of labor and no abnormalities are found during examination, blood is drawn from the umbilical vein at a set volume per unit until no more blood can be obtained.

4. The method for constructing a postpartum bleeding volume identification database according to claim 1, characterized in that: Prior to collecting a blood sample for postpartum hemorrhage, the method further comprises: Determine standardized operating procedures for simulated bleeding dressing images; The standardized operating procedure is used to specify the operating procedure and requirements for collecting blood samples for postpartum hemorrhage and preparing simulated bleeding dressings with different bleeding volume gradients.

5. The method for constructing a postpartum bleeding volume identification database according to claim 4, characterized in that: The step of acquiring images of all the simulated bleeding dressings and establishing a simulated dressing image database containing different bleeding gradients includes: Using an image acquisition device and following the standardized operating procedure to acquire an image of the simulated bleeding dressing; renaming the image of the simulated bleeding dressing according to the blood volume corresponding to each bleeding volume gradient, and saving the named image of the simulated bleeding dressing; All stored images of the simulated bleeding dressings containing different bleeding volume gradients constitute a preliminary postpartum bleeding volume recognition database.

6. The method for constructing a postpartum bleeding volume identification database according to claim 5, characterized in that: The method of using an image acquisition device and following the standardized operation procedure to acquire the image of the simulated bleeding dressing includes: Using the image acquisition device, obtaining an image of the simulated bleeding dressing; The image of the simulated bleeding dressing is checked, and if it does not meet the set requirements, it is re-photographed.

7. The method for constructing a postpartum bleeding volume identification database according to claim 6, characterized in that: The failure to meet the set requirements includes one or more of the following situations: There is blood residue in the syringe; Blood sprays outside the dressing; The background color and lighting of the shooting are inconsistent; The four corners of the dressing are not all shown in the picture; Dressing has wrinkles; The picture is blurry or fuzzy.

8. The method for constructing a postpartum bleeding volume identification database according to claim 5, characterized in that: Also includes: Data enhancement operation is performed on the image of the simulated bleeding dressing to expand the sample size in the database, thereby obtaining a final postpartum hemorrhage volume recognition database for use in training a postpartum hemorrhage volume recognition model.

9. The method for constructing a postpartum bleeding volume identification database according to any one of claims 1 to 8, characterized in that: The data enhancement operation includes: manually segmenting and labeling irregular bleeding areas in the image of the simulated bleeding dressing; After randomly shuffling the order of the labeled images, they are randomly divided into training set and test set databases according to the set ratio; The flipping operation in the geometric transformation method is used to augment the images of the training set and the test set database.

10. The method for constructing a postpartum bleeding volume identification database according to claim 9, characterized in that: The step of manually segmenting and labeling the irregular bleeding area in the image of the simulated bleeding dressing comprises: Using the polygon segmentation method of Labelme, the irregular bleeding areas in all the images of the simulated bleeding dressings are segmented, and different color areas are marked accurately to the pixel point; The segmentation and annotation are completed in the Anaconda environment, and the annotated images are output in JSON format.

Images