Wound negative pressure nursing treatment condition monitoring method and system

Abstract

The invention provides a wound negative pressure nursing treatment condition monitoring method and system, and relates to the technical field of health monitoring, and the method comprises the steps: shooting a stoma image and an ostomy bag image, respectively inputting a first convolutional neural network model and a second convolutional neural network model, and obtaining first feature information and second feature information; obtaining first hidden state information through the first recurrent neural network model, and obtaining second hidden state information through the second recurrent neural network model; and splicing the first feature information and the second feature information to obtain state feature information of each date, and inputting the state feature information into a third recurrent neural network model to obtain third hidden state information, thereby determining stoma nursing treatment condition information and obtaining alarm information. According to the method and the device, the health condition of the stoma can be determined through the stoma image and the stoma bag image, exudate of skin around the stoma does not need to be extracted, the pain of a patient is reduced, and the accuracy of stoma nursing treatment condition information is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of health monitoring, and in particular to a method and system for monitoring the status of negative pressure wound care treatment. Background Art

[0002] In related technologies, the health status of a stoma can usually be determined by detecting whether the skin around the stoma is infected. However, due to the contamination of excrement, the possibility of infection of the skin around the stoma is high, and the monitoring process may require the extraction of exudate from the skin around the stoma, causing increased pain for the patient. The frequency of changing the ostomy bag is high, which reduces the convenience of the patient.

[0003] The information disclosed in the background technology section of this application is only intended to deepen the understanding of the general background technology of this application, and should not be regarded as an admission or any form of suggestion that the information constitutes the prior art already known to those skilled in the art. Summary of the Invention

[0004] The present invention provides a method and system for monitoring the status of wound negative pressure nursing treatment, which can solve the technical problem in the related art that monitoring the health status of the stoma causes increased pain to the patient.

[0005] According to a first aspect of the present invention, a method for monitoring the status of wound negative pressure nursing treatment is provided, comprising:

[0006] After a fungus-shaped tube was implanted at the stoma for negative pressure therapy in patients with ileostomy, images of the stoma and stoma bag were taken at the first preset moment on each day;

[0007] Inputting the stoma image into a first convolutional neural network model to obtain first feature information of the stoma image;

[0008] Inputting the ostomy bag image into a second convolutional neural network model to obtain second feature information of the ostomy bag image;

[0009] The current date and the date before it The first feature information of each date is input into the first recurrent neural network model in date order to obtain the first hidden state information corresponding to multiple dates, and the current date and the dates before it are combined into a single feature information. The first feature information of the dates is input into the second recurrent neural network model in reverse order of the dates to obtain the second hidden state information corresponding to the multiple dates, wherein, is the first preset quantity;

[0010] Concatenate the first characteristic information and the second characteristic information to obtain status characteristic information of each date;

[0011] The current date and the date before it The state feature information is input into the third recurrent neural network model to obtain the third latent state information corresponding to the current date;

[0012] Obtaining stoma care treatment status information corresponding to the current date according to the first latent state information, the second latent state information, and the third latent state information;

[0013] Continuously obtain daily stoma care treatment status information, and generate alarm information when the stoma care treatment status information is abnormal.

[0014] According to the present invention, obtaining stoma care treatment status information corresponding to the current date based on the first latent state information, the second latent state information, and the third latent state information includes:

[0015] Obtain the state similarity between the first latent state information and the second latent state information of the same date;

[0016] The state similarities corresponding to multiple dates are combined into a state similarity vector;

[0017] The first hidden state information corresponding to the current date and the The second latent state information corresponding to the dates is spliced ​​to obtain the stoma disease trend state information, where the current date is the jth date;

[0018] The state similarity vector and the stoma disease trend state information are spliced ​​to obtain the stoma disease trend global state information;

[0019] The global state information of stoma disease trend and the third latent state information corresponding to the current date are combined to obtain characteristic information of stoma care status;

[0020] The stoma care status feature information is input into the first fully connected layer to obtain the stoma care treatment status information.

[0021] According to the present invention, the method further comprises:

[0022] acquiring first sample stoma images and first sample ostomy bag images on multiple dates for multiple first sample patients implanted with a fungiform tube for negative pressure therapy;

[0023] acquiring second sample stoma images and second sample ostomy bag images on multiple dates for multiple second sample patients who do not have a fungiform tube implanted for negative pressure therapy;

[0024] Obtaining, by a first convolutional neural network model, first stoma sample feature information of the first sample stoma image and second stoma sample feature information of the second sample stoma image;

[0025] Obtaining, by a second convolutional neural network model, first ostomy bag sample feature information of the first sample ostomy bag image and second ostomy bag sample feature information of the second sample ostomy bag image;

[0026] Processing the first stoma sample feature information through a first recurrent neural network model to obtain first sample latent state information, and processing the first stoma sample feature information through a second recurrent neural network model to obtain second sample latent state information;

[0027] Processing the second stoma sample feature information through the first recurrent neural network model to obtain third sample hidden state information, and processing the second stoma sample feature information through the second recurrent neural network model to obtain fourth sample hidden state information;

[0028] Obtaining a stoma state loss function according to the first sample hidden state information, the second sample hidden state information, the third sample hidden state information, and the fourth sample hidden state information;

[0029] Obtaining an ostomy bag state loss function according to the first ostomy bag sample feature information and the second ostomy sample feature information;

[0030] Obtaining a stoma health state loss function according to the first stoma sample feature information, the first stoma bag sample feature information, the second stoma sample feature information, and the second stoma bag sample feature information, and a third recurrent neural network model;

[0031] Obtaining a stoma care status loss function according to the first sample latent state information, the second sample latent state information, the third sample latent state information, the fourth sample latent state information, the first stoma sample feature information, the first stoma bag sample feature information, the second stoma sample feature information, and the second stoma bag sample feature information, and a third recurrent neural network model;

[0032] Obtain a comprehensive loss function based on the stoma state loss function, the stoma bag state loss function, the stoma health state loss function, and the stoma care state loss function;

[0033] Through the comprehensive loss function, the first convolutional neural network model, the second convolutional neural network model, the first recurrent neural network model, the second recurrent neural network model and the third recurrent neural network model are trained.

[0034] According to the present invention, a stoma state loss function is obtained based on the first sample hidden state information, the second sample hidden state information, the third sample hidden state information, and the fourth sample hidden state information, including:

[0035] Input the first sample hidden state information corresponding to the last date into the second fully connected layer to obtain the first stoma disease sample probability value of the first sample patient, and input the third sample hidden state information corresponding to the last date into the second fully connected layer to obtain the second stoma disease sample probability value of the second sample patient;

[0036] Obtaining the first sample state similarity between the first sample hidden state information and the second sample hidden state information corresponding to the same date, and obtaining the second sample state similarity between the third sample hidden state information and the fourth sample hidden state information corresponding to the same date;

[0037] Obtaining global state information of the stoma disease trend of the first sample according to the first sample state similarity, the first sample hidden state information, and the second sample hidden state information;

[0038] Obtaining global state information of the stoma disease trend of the second sample according to the second sample state similarity, the third sample hidden state information, and the fourth sample hidden state information;

[0039] Inputting the global state information of the stoma disease trend of the first sample into the third fully connected layer to obtain a third stoma disease sample probability value, and inputting the global state information of the stoma disease trend of the second sample into the third fully connected layer to obtain a fourth stoma disease sample probability value;

[0040] A stoma state loss function is obtained according to the first stoma diseased sample probability value, the second stoma diseased sample probability value, the third stoma diseased sample probability value, and the fourth stoma diseased sample probability value.

[0041] According to the present invention, a stoma state loss function is obtained according to the first stoma disease sample probability value, the second stoma disease sample probability value, the third stoma disease sample probability value, and the fourth stoma disease sample probability value, including:

[0042] According to the formula , Obtain stoma state loss function ,in, The probability of stoma disease in the first sample patient is marked, The probability of stoma disease in the second sample patients is marked. is the probability value of the first stoma disease sample, is the probability value of the third stoma disease sample, is the probability value of the second stoma disease sample, is the probability value of the fourth stoma disease sample, and is the preset weight, and max is the maximum value function.

[0043] According to the present invention, the ostomy bag state loss function is obtained according to the first ostomy bag sample characteristic information and the second ostomy sample characteristic information, including:

[0044] Input the first ostomy bag sample feature information of each date into the fourth fully connected layer to obtain the first usage ratio of ostomy bags for the first sample patients on each date, and input the second ostomy bag sample feature information of each date into the fourth fully connected layer to obtain the second usage ratio of ostomy bags for the second sample patients on each date;

[0045] According to the formula , Obtaining the pouch state loss function ,in, is the first usage ratio of ostomy bag for the first sample patient on the i-th day, The first usage ratio of ostomy bag for the first sample patients on date 1, is the first usage ratio of ostomy bag for the first sample patient on the i-1th date, The labeling information for the proportion of stoma bags used by the first sample patients on the first day, The labeling information of the usage ratio of ostomy bags for the first sample patient on the i-th day, is the second usage ratio of ostomy bag for the second sample patients on the i-th day, is the second usage ratio of ostomy bags for the second sample patients on the i-1th date, The labeling information for the usage ratio of ostomy bags for the second sample patients on the i-th date, The second usage ratio of ostomy bag for the second sample patients on date 1, The labeling information for the proportion of stoma bags used by the second sample patients on the first day, is the first preset quantity, and is the preset weight, is the preset parameter, and if is the conditional function.

[0046] According to the present invention, a stoma health state loss function is obtained based on the first stoma sample feature information, the first stoma bag sample feature information, the second stoma sample feature information, and the second stoma bag sample feature information, and the third recurrent neural network model, including:

[0047] The first stoma sample characteristic information and the first stoma bag sample characteristic information of the same date are spliced ​​together to obtain the first state characteristic information of each date, and the second stoma sample characteristic information and the second stoma bag sample characteristic information of the same date are spliced ​​together to obtain the second state characteristic information of each date;

[0048] The first state feature information of each date is processed by the third recurrent neural network model to obtain the fifth sample hidden state information of the last date, and the second state feature information of each date is processed to obtain the sixth sample hidden state information of the last date;

[0049] Inputting the fifth sample hidden state information into the fifth fully connected layer to obtain the first sample stoma disease probability value, and inputting the sixth sample hidden state information into the fifth fully connected layer to obtain the second sample stoma disease probability value;

[0050] A stoma health state loss function is obtained according to the stoma disease probability value of the first sample, the stoma disease probability value of the second sample, the probability labeling of the stoma disease of the first sample patient, and the probability labeling of the stoma disease of the second sample patient.

[0051] According to the present invention, based on the first sample hidden state information, the second sample hidden state information, the third sample hidden state information, the fourth sample hidden state information, the first stoma sample feature information, the first ostomy bag sample feature information, the second stoma sample feature information and the second ostomy bag sample feature information, and the third recurrent neural network model, a stoma care status loss function is obtained, including:

[0052] Obtaining global state information of the stoma disease trend of the first sample according to the first sample hidden state information, the second sample hidden state information, the first stoma sample feature information, the first stoma bag sample feature information, and the third recurrent neural network model;

[0053] Obtaining global state information of the stoma disease trend of the second sample according to the third sample hidden state information, the fourth sample hidden state information, the second stoma sample feature information, the second stoma bag sample feature information, and the third recurrent neural network model;

[0054] Inputting the first sample stoma disease trend global state information into the first fully connected layer to obtain the first sample stoma nursing treatment status probability value, and inputting the second sample stoma disease trend global state information into the first fully connected layer to obtain the second sample stoma nursing treatment status probability value;

[0055] Inputting the first sample stoma disease trend global state information into the sixth fully connected layer to obtain a first probability value that the first sample patient is a patient with a mushroom tube implanted, and inputting the second sample stoma disease trend global state information into the sixth fully connected layer to obtain a second probability value that the second sample patient is a patient with a mushroom tube implanted;

[0056] A stoma care condition loss function is obtained according to the first sample stoma care treatment condition probability value, the second sample stoma care treatment condition probability value, the first probability value, and the second probability value.

[0057] According to the present invention, a stoma care treatment condition loss function is obtained according to the first sample stoma care treatment condition probability value, the second sample stoma care treatment condition probability value, the first probability value, and the second probability value, including:

[0058] According to the formula , Obtain stoma care status loss function ,in, is the first probability value, is the second probability value, The probability of stoma disease in the first sample patient is marked, The probability of stoma disease in the second sample patients is marked. is the probability value of the stoma care treatment status of the first sample, is the probability value of stoma care treatment status of the second sample, and is the preset weight.

[0059] According to a second aspect of the present invention, a wound negative pressure care treatment status monitoring system is provided, comprising:

[0060] an implantation module, configured to capture an image of the stoma and an image of the stoma bag at the first preset moment on each day after a fungus-shaped tube is implanted at the stoma of an ileostomy patient for negative pressure therapy;

[0061] A first feature information module, configured to input the stoma image into a first convolutional neural network model to obtain first feature information of the stoma image;

[0062] A second feature information module is used to input the ostomy bag image into the second convolutional neural network model to obtain second feature information of the ostomy bag image;

[0063] Hidden state information module, used to store the current date and the previous date The first feature information of each date is input into the first recurrent neural network model in date order to obtain the first hidden state information corresponding to multiple dates, and the current date and the dates before it are combined into a single feature information. The first feature information of the dates is input into the second recurrent neural network model in reverse order of the dates to obtain the second hidden state information corresponding to the multiple dates, wherein, is the first preset quantity;

[0064] A status characteristic information module, configured to concatenate the first characteristic information and the second characteristic information to obtain status characteristic information for each date;

[0065] The third hidden state information module is used to store the current date and the date before it. The state feature information is input into the third recurrent neural network model to obtain the third latent state information corresponding to the current date;

[0066] The stoma nursing treatment status information module is used to obtain the stoma nursing treatment status information corresponding to the current date based on the first latent state information, the second latent state information, and the third latent state information;

[0067] The alarm information module is used to continuously obtain daily stoma care and treatment status information and generate alarm information when the stoma care and treatment status information is abnormal.

[0068] By adopting the above technical solution, the present invention can achieve the following technical effects:

[0069] According to the present invention, using a mushroom-shaped tube for negative pressure therapy not only reduces excreta flowing into the ostomy bag, thus reducing the frequency of bag changes, but also reduces contamination of the skin around the stoma by excreta, thereby reducing the probability of peristomal skin infection. Furthermore, by photographing the stoma and pouch, the stoma's health status can be determined by comparing multiple images of the stoma and pouch taken on different dates. This eliminates the need to extract exudate from the peristomal skin, thus reducing patient suffering. Based on the stoma and pouch images, the frequency of bag changes and other information, such as the stoma's status, can be comprehensively determined, thereby improving the accuracy of information on stoma care and treatment status. Furthermore, the fungiform tube can divert intestinal fluid and excreta from the pouch, thereby reducing the possibility of local skin irritation caused by excreta leakage and reducing the frequency of bag changes. Furthermore, the stoma and pouch's status can be monitored daily by taking images of the stoma and pouch and using multiple models. Leveraging the characteristics of recurrent neural networks, latent state information can be used to express the likelihood of a stoma's health status, improving monitoring accuracy and providing accurate alerts to doctors and caregivers. When determining the stoma status loss function, a maximum function can be selected. This allows for direct reduction of the error between the first sample's latent state information and the second sample's latent state information, as well as for indirect reinforcement of model training by reducing the error in information such as the first sample's state similarity, thereby reducing the error between the first sample's latent state information and the second sample's latent state information. This increases training intensity and improves training efficiency and model performance. When determining the ostomy bag status loss function, a conditional function can be used to select the loss function value based on the theoretically faster ostomy bag usage speed of the second sample patient. This increases training intensity and efficiency when the ostomy bag usage speed prediction is incorrect, reduces the error between the first usage ratio and the second usage ratio, and improves model accuracy, thereby more accurately representing the impact of the fungus tube on the frequency of ostomy bag replacement. Furthermore, when constructing the stoma care status loss function, a loss function for predicting whether a fungus tube has been implanted can be used for auxiliary training, thereby increasing the training intensity of the model before the first fully connected layer. By multiplying, this prevents the first fully connected layer from being unable to train, improving training intensity and efficiency, and overall speeding up model accuracy.

[0070] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and not limiting of the present invention. Other features and aspects of the present invention will become more apparent from the following detailed description of exemplary embodiments with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can derive other embodiments based on these drawings without inventive efforts.

[0072] Figure 1 A schematic diagram exemplarily illustrates a flow chart of a method for monitoring wound negative pressure nursing treatment conditions according to an embodiment of the present invention;

[0073] Figure 2 A schematic diagram exemplarily illustrates obtaining stoma care treatment status information according to an embodiment of the present invention;

[0074] Figure 3 The following is a block diagram of a system for monitoring the status of negative pressure wound care treatment according to an embodiment of the present invention. DETAILED DESCRIPTION

[0075] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0076] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0077] Figure 1 The following is a flow chart showing a method for monitoring the status of wound negative pressure nursing treatment according to an embodiment of the present invention, wherein the method includes:

[0078] Step S1, after a fungus-shaped tube is implanted at the stoma of an ileostomy patient for negative pressure therapy, an image of the stoma and an image of the stoma bag are captured at the first preset time on each date;

[0079] Step S2, inputting the stoma image into a first convolutional neural network model to obtain first feature information of the stoma image;

[0080] Step S3, inputting the ostomy bag image into a second convolutional neural network model to obtain second feature information of the ostomy bag image;

[0081] Step S4: The current date and the date before it The first feature information of each date is input into the first recurrent neural network model in date order to obtain the first hidden state information corresponding to multiple dates, and the current date and the dates before it are combined into a single feature information. The first feature information of the dates is input into the second recurrent neural network model in reverse order of the dates to obtain the second hidden state information corresponding to the multiple dates, wherein, is the first preset quantity;

[0082] Step S5, concatenating the first characteristic information and the second characteristic information to obtain status characteristic information of each date;

[0083] Step S6: The current date and the date before it The state feature information is input into the third recurrent neural network model to obtain the third latent state information corresponding to the current date;

[0084] Step S7, obtaining stoma care treatment status information corresponding to the current date based on the first hidden state information, the second hidden state information, and the third hidden state information;

[0085] Step S8: continuously obtain daily stoma care and treatment status information, and generate alarm information when the stoma care and treatment status information is abnormal.

[0086] According to an embodiment of the present invention, the wound negative pressure nursing treatment status monitoring method utilizes a mushroom tube for negative pressure therapy, which not only reduces excreta flowing into the ostomy bag, thus reducing the frequency of bag changes, but also reduces contamination of the skin around the stoma by excreta, thereby lowering the probability of peristomal skin infection. Furthermore, by photographing the stoma and ostomy bag, the stoma's health status can be determined by comparing multiple dates of stoma and ostomy bag images to determine changes in the stoma's status. This eliminates the need to extract exudate from the peristomal skin, thus reducing patient pain. Based on the stoma and ostomy bag images, the stoma bag change frequency and stoma status can be comprehensively determined, thereby improving the accuracy of stoma nursing treatment status information.

[0087] According to one embodiment of the present invention, in step S1, the excreta of ileostomy patients contains alkaline digestive enzymes. If the excreta in the ostomy bag leaks out, it will soak the skin, causing damage to the skin around the stoma, increasing the patient's pain and the difficulty of stoma care. Therefore, a mushroom-shaped tube can be used to insert an intestinal tube through the stoma to continuously collect intestinal fluid and excreta, reduce the amount of intestinal fluid and excreta flowing into the ostomy bag, and reduce the possibility of local skin irritation caused by excreta leakage. It can also reduce the speed of ostomy bag use and replacement frequency, reduce the probability of ostomy bag chassis leakage, and reduce the probability of stoma complications, saving care time and costs and improving patient comfort. In the example, the size of the stoma can be measured, and a mushroom tube of appropriate size can be selected, and the mushroom tube can be implanted into the intestinal tube. For example, the head of the mushroom tube can be implanted 7 to 10 cm into the intestinal tube, and a small hole can be made above the stoma bag to facilitate the tail end of the mushroom tube to pass through. The gap where the mushroom tube passes through the small hole of the stoma bag is sealed with a dressing. The tail end of the mushroom tube can be connected to a low-negative pressure suction device to drain intestinal fluid and some excrement (for example, liquid excrement), thereby reducing the probability of intestinal fluid and some excrement contaminating the skin and causing infection, and also reducing the frequency of changing the stoma bag.

[0088] According to one embodiment of the present invention, after performing the above-described negative pressure therapy treatment using the fungiform tube, an image of the stoma and the stoma bag can be captured at the first preset time of each day. For example, the stoma image can include the stoma location, allowing observation of the stoma itself and the surrounding skin. The stoma bag image can also include the stoma bag, allowing observation of information such as the amount of fecal matter within the bag, as well as its color and condition.

[0089] Figure 2 A schematic diagram of obtaining stoma care treatment status information according to an embodiment of the present invention is exemplarily shown.

[0090] According to one embodiment of the present invention, in step S2, the first convolutional neural network model can be a VGGNet, ResNet, or other model, which processes the input image to obtain feature information in vector form. That is, the stoma image can be processed to output first feature information in vector form. Furthermore, the stoma image for each date can be processed in the same manner to obtain the first feature information for each date.

[0091] According to one embodiment of the present invention, in step S3, similar to the first convolutional neural network model, the second convolutional neural network model can be a VGGNet, ResNet, or other model, which processes the input image to obtain feature information in vector form. That is, the ostomy bag image can be processed to output second feature information in vector form. Furthermore, the ostomy bag image for each date can be processed in the same manner to obtain the second feature information for each date.

[0092] According to one embodiment of the present invention, in step S4, the first feature information can be used to describe the state of the stoma, for example, whether the stoma is red and swollen, whether there is exudate, etc. Based on the first feature information of multiple dates, it can be determined whether the state of the stoma has changed on multiple dates. In the example, the first preset number can be 3, 5 or 7, and the present invention does not limit the specific value of the first preset number. The first feature information of the current date and the first preset number of dates before can be input into the first recurrent neural network model in date order. For example, if the current date is May 20 and the first preset number is 3, the first feature information of May 18, May 19 and May 20 can be input into the first recurrent neural network model in date order, and the process can be repeated in daily processing. For example, on May 21, the first feature information of May 19, May 20 and May 21 can be input into the first recurrent neural network model in date order. The first recurrent neural network model can be a long short-term memory network model (LSTM). The first recurrent neural network model has a first preset number of input ports, each used to receive first feature information of a first preset number of dates, and has a first preset number of processing blocks. The blocks are connected in sequence. The block can receive the first feature information and output first hidden state information after processing. The first hidden state information can be input to the next block, processed together with the first feature information received by the next block, and the first hidden state information is output again. Therefore, the first recurrent neural network model can output a first preset number of first hidden state information, and each first hidden state information corresponds to each date. Since the first hidden state information output by each block is passed to the next block, the last first hidden state information can carry information of multiple dates, which is used to reflect the status of the stoma on multiple dates and the changes in the stoma on multiple dates. The structure of the second recurrent neural network model is the same as that of the first recurrent neural network model (the number of blocks and input ports is also the first preset number), but its specific model parameters may be different. When inputting the second recurrent neural network model, the input is performed in reverse order of the date. For example, if the date is May 20 and the first preset number is 3, the corresponding first feature information can be input into the second recurrent neural network model in the order of May 20, May 19 and May 18, and the hidden state information of multiple dates can also be obtained, that is, the second hidden state information, and the second hidden state information output by the last block (for example, the block corresponding to May 18) can carry information of multiple dates, which is used to reflect the status of the stoma on multiple dates and the changes in the stoma on multiple dates.

[0093] According to one embodiment of the present invention, the first recurrent neural network model and the second recurrent neural network model can constitute a bidirectional recurrent neural network model. If the stoma is stable, with no redness, swelling, or exudate, then the difference between the hidden state information output by the first and second recurrent neural network models for the corresponding dates is theoretically small. The difference between the first hidden state information output by the last block of the first recurrent neural network model, which carries information on multiple dates, and the second hidden state information output by the first block of the second recurrent neural network model, which carries information on multiple dates, is also theoretically small. If the stoma is in poor condition, with redness, swelling, or exudate, the severity of the redness, swelling, or exudate increases in order of date. Since each block of the recurrent neural network model can access information from the previous block for processing, the hidden state information output by the last block retains some information from the previous block. Therefore, in the first recurrent neural network model, each block retains some information from the previous block, resulting in the recorded severity of the redness, swelling, or exudate being lower than that of the current date. Conversely, in the second recurrent neural network model, since the input order of the first feature information is reversed, when retaining the information of the previous block, information about conditions such as redness, swelling, and exudate with a higher severity than the date corresponding to the current block will be retained. The second hidden state information corresponding to the earliest date will reflect the partial retention of information about conditions such as redness, swelling, and exudate with a higher severity than that date, while the second hidden state information corresponding to the current date will not retain information about conditions such as redness, swelling, and exudate with a lower severity than the current date. Therefore, when the stoma experiences redness, swelling, exudate, etc., the difference between the hidden state information output by the first and second recurrent neural network models for the corresponding date blocks is theoretically large. Based on this, the difference between the hidden state information output by the first and second recurrent neural network models for the corresponding date blocks can be used as one of the bases for judging the health status of the stoma.

[0094] According to one embodiment of the present invention, in step S5, the first characteristic information and the second characteristic information can be spliced ​​to obtain status characteristic information, wherein the first characteristic information can be used to describe the health status of the stoma, and the second characteristic information can be used to describe the condition of the stoma bag. Splicing the two can be used to describe the overall condition of the day. The splicing is to connect the two vectors end to end. For example, if the first characteristic information and the second characteristic information are both 64-dimensional vectors, 128-dimensional status characteristic information can be obtained after splicing.

[0095] According to one embodiment of the present invention, in step S6, the third recurrent neural network model is similar to the first recurrent neural network model and may be an LSTM model. The number of its blocks is also the first predetermined number. The state feature information for each date may be input into the third recurrent neural network model in chronological order to obtain the third latent state information for each date. As described above, the subsequent block of the recurrent neural network model may be calculated in conjunction with the latent state information output by the previous block. Therefore, the third latent state information output by the last block (i.e., the block corresponding to the current date) may retain some information from the previous date. The third latent state information corresponding to the current date may be used as information capable of describing the overall state changes for the first predetermined number of dates.

[0096] According to one embodiment of the present invention, in step S7, the above information describing the daily conditions of the stoma and stoma bag and the information describing the overall status changes are obtained. The above information can be combined to determine the stoma care status, that is, the stoma care treatment status information, which is used to indicate whether the stoma care treatment is reasonable and whether there are any abnormalities in the stoma.

[0097] According to one embodiment of the present invention, in step S7, the stoma care treatment status information corresponding to the current date is obtained based on the first hidden state information, the second hidden state information, and the third hidden state information, including: obtaining the state similarity of the first hidden state information and the second hidden state information of the same date; forming a state similarity vector from the state similarities corresponding to multiple dates; and combining the first hidden state information corresponding to the current date and the second hidden state information. The second hidden state information corresponding to the jth date is spliced ​​to obtain the stoma disease trend state information, where the current date is the jth date; the state similarity vector and the stoma disease trend state information are spliced ​​to obtain the stoma disease trend global state information; the stoma disease trend global state information and the third hidden state information corresponding to the current date are spliced ​​to obtain the stoma nursing status feature information; the stoma nursing status feature information is input into the first fully connected layer to obtain the stoma nursing treatment status information.

[0098] According to one embodiment of the present invention, as described above, the difference between the first latent state information and the second latent state information on the same date can be used as one basis for determining the stoma health status. The state similarity (e.g., cosine similarity) between the first latent state information and the second latent state information on the same date can be calculated to describe the stoma health status. The state similarities of multiple dates are combined to form a state similarity vector, which can be used to describe the overall stoma health status across the multiple dates.

[0099] According to one embodiment of the present invention, the hidden state information output by the last block of the first recurrent neural network model and the second recurrent neural network model can retain part of the hidden state information output by the previous block. Therefore, the first hidden state information corresponding to the current date can be used to describe the overall stoma health change status of the current and past dates. Similarly, the first hidden state information corresponding to the current date can be used to describe the overall stoma health change status of the current and past dates. The second hidden state information corresponding to the date is the hidden state information output by the last block of the second recurrent neural network model, and some hidden state information output by the previous block can be retained. Therefore, The second latent state information corresponding to the date can be used to describe the The overall stoma health changes on a date and multiple dates thereafter can be spliced ​​together (i.e., connected end to end) to obtain stoma disease trend status information, which can describe the overall stoma health changes from both positive and reverse directions.

[0100] According to one embodiment of the present invention, the stoma disease trend state information and the state similarity vector are concatenated (i.e., connected end to end) to obtain global stoma disease trend state information, which can be used to describe whether the stoma is abnormal or whether there is a trend of pathological changes. Furthermore, as described above, the third latent state information corresponding to the current date can describe the overall state changes for a first preset number of dates, while the global stoma disease trend state information can be used to describe the stoma state changes for the first preset number of dates. Concatenating the two can be used to describe whether the stoma and ostomy bag have abnormalities or whether there is a trend of abnormalities over the first preset number of dates. In other words, it can indicate whether the stoma treatment and care over the first preset number of dates are reasonable. When the stoma care status feature information is input into the first fully connected layer for calculation, the fully connected operation can also be used to strengthen the relationship between the stoma status and the stoma bag status. For example, abnormal color of fecal matter in the stoma bag and redness and swelling of the stoma can further indicate abnormal stoma health and inappropriate care and treatment. The fully connected operation can strengthen this relationship. After passing through the first fully connected layer, stoma care treatment status information can be obtained through an activation layer (for example, through an activation operation performed by a softmax function), which can represent the probability that the care and treatment are reasonable.

[0101] According to one embodiment of the present invention, in step S8, daily stoma care and treatment status information can be continuously acquired. For example, on May 20, stoma care and treatment status information can be determined based on images captured from May 18 to 20, and on May 21, stoma care and treatment status information can be determined based on images captured from May 19 to 21. Stoma care and treatment status information can be continuously determined, and if the stoma care and treatment status information falls below a preset threshold (e.g., 0.5), it can be determined that the stoma or ostomy bag has an abnormality, or a trend of abnormality, indicating inappropriate care or treatment. For example, if the drainage capacity of the fungus duct is weakened, causing more intestinal fluid and fecal matter to enter the ostomy bag, which can easily overflow and infect the skin, indicating possible fungus duct obstruction, an alarm message can be sent to alert the doctor or caregiver to the specific condition of the fungus duct and to disinfect the patient's stoma.

[0102] In this way, the fungiform tube can be used to divert intestinal fluid and excrement flowing into the ostomy bag, thereby reducing the possibility of local skin irritation caused by excrement leakage. The frequency of ostomy bag replacement can also be reduced. The condition of the stoma and ostomy bag can be monitored by taking images of the stoma and ostomy bag every day and using multiple models. By utilizing the characteristics of the recurrent neural network model, the possibility of a health condition of the stoma can be expressed using latent state information, thereby improving monitoring accuracy and providing accurate alarm information to doctors and caregivers.

[0103] According to one embodiment of the present invention, the above multiple neural network models may be trained before use, and the method further includes:

[0104] Step S101, acquiring first sample stoma images and first sample ostomy bag images of multiple first sample patients implanted with fungus-forming tubes for negative pressure therapy on multiple dates;

[0105] Step S102, obtaining second sample stoma images and second sample ostomy bag images on multiple dates of multiple second sample patients who have not had mushroom tubes implanted for negative pressure therapy;

[0106] Step S103, obtaining first stoma sample feature information of the first sample stoma image and second stoma sample feature information of the second sample stoma image through the first convolutional neural network model;

[0107] Step S104, obtaining first ostomy bag sample feature information of the first sample ostomy bag image and second ostomy bag sample feature information of the second sample ostomy bag image through a second convolutional neural network model;

[0108] Step S105, processing the first stoma sample feature information through the first recurrent neural network model to obtain first sample latent state information, and processing the first stoma sample feature information through the second recurrent neural network model to obtain second sample latent state information;

[0109] Step S106, processing the second stoma sample feature information through the first recurrent neural network model to obtain third sample latent state information, and processing the second stoma sample feature information through the second recurrent neural network model to obtain fourth sample latent state information;

[0110] Step S107, obtaining a stoma state loss function according to the first sample latent state information, the second sample latent state information, the third sample latent state information, and the fourth sample latent state information;

[0111] Step S108, obtaining an ostomy bag state loss function according to the first ostomy bag sample characteristic information and the second ostomy bag sample characteristic information;

[0112] Step S109, obtaining a stoma health status loss function according to the first stoma sample characteristic information, the first stoma bag sample characteristic information, the second stoma sample characteristic information, the second stoma bag sample characteristic information, and the third recurrent neural network model;

[0113] Step S110, obtaining a stoma care status loss function based on the first sample latent state information, the second sample latent state information, the third sample latent state information, the fourth sample latent state information, the first stoma sample feature information, the first ostomy bag sample feature information, the second stoma sample feature information, the second ostomy bag sample feature information, and the third recurrent neural network model;

[0114] Step S111, obtaining a comprehensive loss function according to the stoma state loss function, the stoma bag state loss function, the stoma health state loss function and the stoma care condition loss function;

[0115] Step S112: training the first convolutional neural network model, the second convolutional neural network model, the first recurrent neural network model, the second recurrent neural network model, and the third recurrent neural network model through a comprehensive loss function.

[0116] According to one embodiment of the present invention, in steps S101 and S102, multiple first sample patients who have had fungiform tubes implanted for negative pressure therapy may be used as positive samples, and multiple second sample patients who have not had fungiform tubes implanted for negative pressure therapy may be used as negative samples. The multiple positive samples may be combined into a positive sample set, and the multiple negative samples may be combined into a negative sample set. First sample stoma images and first sample ostomy bag images may be acquired for multiple dates for the first sample patients, and second sample stoma images and second sample ostomy bag images may be acquired.

[0117] According to one embodiment of the present invention, the processing methods of step S103 and step S104 are similar to the processing methods of the above-mentioned step S2 and step S3, and the processing methods of step S105 and step S106 are similar to the above-mentioned step S4, which will not be repeated here.

[0118] According to one embodiment of the present invention, in step S107, a stoma state loss function is obtained based on the first sample hidden state information, the second sample hidden state information, the third sample hidden state information, and the fourth sample hidden state information, including: inputting the first sample hidden state information corresponding to the last date into the second fully connected layer to obtain a first stoma disease sample probability value of the first sample patient, and inputting the third sample hidden state information corresponding to the last date into the second fully connected layer to obtain a second stoma disease sample probability value of the second sample patient; obtaining the first sample state similarity between the first sample hidden state information and the second sample hidden state information corresponding to the same date, and obtaining the second sample state similarity between the third sample hidden state information and the fourth sample hidden state information corresponding to the same date. State similarity; according to the first sample state similarity, the first sample hidden state information and the second sample hidden state information, obtain the global state information of the stoma disease trend of the first sample; according to the second sample state similarity, the third sample hidden state information and the fourth sample hidden state information, obtain the global state information of the stoma disease trend of the second sample; input the global state information of the stoma disease trend of the first sample into the third fully connected layer to obtain the third stoma disease sample probability value, and input the global state information of the stoma disease trend of the second sample into the third fully connected layer to obtain the fourth stoma disease sample probability value; according to the first stoma disease sample probability value, the second stoma disease sample probability value, the third stoma disease sample probability value and the fourth stoma disease sample probability value, obtain the stoma state loss function.

[0119] According to one embodiment of the present invention, the first sample hidden state information corresponding to the last date and the third sample hidden state information corresponding to the last date are similar to the first hidden state information of the current date and can be used to represent the stoma status on multiple dates. The first sample hidden state information can be input into the second fully connected layer and then processed by the activation layer to obtain the first stoma disease sample probability value for the first sample patient. Similarly, the second stoma disease sample probability value can be obtained.

[0120] According to one embodiment of the present invention, in accordance with the state similarity of the first hidden state information and the second hidden state information obtained for the same date, the first sample state similarity of the first sample hidden state information and the second sample hidden state information corresponding to the same date, as well as the second sample state similarity of the third sample hidden state information and the fourth sample hidden state information corresponding to the same date can be obtained. Similar to the above method of obtaining the global state information of the stoma disease trend, the first sample state similarities can be combined into a first sample state similarity vector, and the first sample hidden state information corresponding to the last date and the second sample hidden state information corresponding to the last date are concatenated, and then concatenated with the first sample state similarity vector to obtain the first sample stoma disease trend global state information. Similarly, the second sample stoma disease trend global state information can be obtained. The first sample stoma disease trend global state information and the second sample stoma disease trend global state information can be used to describe the state change of the stoma, that is, the trend of whether the stoma has lesions or not. The first sample stoma disease trend global state information can be processed through a third fully connected layer and then through an activation layer to obtain a third stoma disease sample probability value. Similarly, a fourth stoma disease sample probability value can be obtained.

[0121] According to one embodiment of the present invention, a stoma state loss function is obtained according to the first stoma disease sample probability value, the second stoma disease sample probability value, the third stoma disease sample probability value, and the fourth stoma disease sample probability value, including: obtaining the stoma state loss function according to formula (1): , (1), in, The probability of stoma disease in the first sample patient is marked, The probability of stoma disease in the second sample patients is marked. is the probability value of the first stoma disease sample, is the probability value of the third stoma disease sample, is the probability value of the second stoma disease sample, is the probability value of the fourth stoma disease sample, and is the preset weight, and max is the maximum value function.

[0122] According to one embodiment of the present invention, in formula (1), is the cross entropy loss function of the probability value of the first stoma disease sample and the probability label of the first sample patient's stoma disease (the probability label of the first sample patient's stoma disease and the probability label of the second sample patient's stoma disease are both 0 or 1, 1 means diseased, 0 means no disease), The cross entropy loss function is the probability value of the third stoma disease sample and the probability of the first sample patient having a stoma disease, In order to take the maximum value between the two as part of the stoma state loss function for training, the first convolutional neural network model, the first recurrent neural network model and the second recurrent neural network model were used when obtaining the first stoma diseased sample probability value and the third stoma diseased sample probability value. When obtaining the third stoma diseased sample probability value, information such as the first sample state similarity was also combined. In theory, the cross-entropy loss function corresponding to the third stoma diseased sample probability value is smaller and the prediction is more accurate. However, there may be errors in the parameters of the first recurrent neural network model and the second recurrent neural network model, which leads to errors in the first sample state similarity and the possibility of introducing more errors when determining the third stoma diseased sample probability value. Therefore, the above two cross-entropy loss functions can be selected. The maximum value of is used as part of the stoma state loss function for training, and the maximum value function is selected to reduce the error of information such as the first sample state similarity, thereby reducing the error of the first sample hidden state information and the second sample hidden state information. That is, when the first cross entropy loss function is high, the first cross entropy loss function is selected for direct training, and the error of the first sample hidden state information and the second sample hidden state information is directly reduced. When the second cross entropy loss function is high, it can better reflect that the model error is large, and the error between the first sample hidden state information and the second sample hidden state information is large. Therefore, the model training can be strengthened by reducing the error of information such as the first sample state similarity to reduce the error of the first sample hidden state information and the second sample hidden state information. The meaning of is similar to the above and will not be repeated here. The weighted sum of the above two items can be used to obtain the stoma state loss function.

[0123] In this way, selection can be made by taking the maximum value function, so that the method of directly reducing the error between the first sample hidden state information and the second sample hidden state information, and the method of reducing the error of information such as the first sample state similarity to indirectly strengthen model training and thereby reduce the error between the first sample hidden state information and the second sample hidden state information can be used to improve training intensity, thereby improving training efficiency and model performance.

[0124] According to one embodiment of the present invention, in step S108, the ostomy bag state loss function is obtained based on the first ostomy bag sample feature information and the second ostomy bag sample feature information, including: inputting the first ostomy bag sample feature information of each date into the fourth fully connected layer to obtain the first usage ratio of the ostomy bag of the first sample patient on each date, and inputting the second ostomy bag sample feature information of each date into the fourth fully connected layer to obtain the second usage ratio of the ostomy bag of the second sample patient on each date; and obtaining the ostomy bag state loss function according to formula (2) , (2), in, is the first usage ratio of ostomy bag for the first sample patient on the i-th day, The first usage ratio of ostomy bag for the first sample patients on date 1, is the first usage ratio of ostomy bag for the first sample patient on the i-1th date, The labeling information for the proportion of stoma bags used by the first sample patients on the first day, The labeling information of the usage ratio of ostomy bags for the first sample patient on the i-th day, is the second usage ratio of ostomy bag for the second sample patients on the i-th day, is the second usage ratio of ostomy bags for the second sample patients on the i-1th date, The labeling information for the usage ratio of ostomy bags for the second sample patients on the i-th date, The second usage ratio of ostomy bag for the second sample patients on date 1, The labeling information for the proportion of stoma bags used by the second sample patients on the first day, is the first preset quantity, and is the preset weight, is the preset parameter, and if is the conditional function.

[0125] According to one embodiment of the present invention, the replacement frequency of the ostomy bag can be determined by the first ostomy bag sample characteristic information and the second ostomy sample characteristic information. Therefore, the first ostomy bag sample characteristic information of each date can be processed by the fourth fully connected layer and then by the activation layer to obtain the first usage ratio of the ostomy bag of the first sample patient on each date. Similarly, the second usage ratio of the ostomy bag of the second sample patient on each date can be obtained.

[0126] According to one embodiment of the present invention, in formula (2), theoretically, the frequency of changing the ostomy bag of a patient with a fungus-shaped tube implanted is lower because the fungus-shaped tube will drain a portion of the excrement and prevent it from entering the ostomy bag. Indicates In the case of , otherwise . is the change in the first usage ratio between the i-th date and the i-1-th date, which can represent the usage speed of the ostomy bag of the first sample patient. is the change in the second usage ratio between the i-th date and the i-1-th date, which can represent the usage speed of the ostomy bag of the second sample patients. The usage speed of the ostomy bag of the second sample patients is theoretically faster. Therefore, if , indicating that the prediction of the usage speed is correct, but there may be a certain error. Therefore, the error between the first usage ratio and the annotation information can be used As the conditional function value, otherwise, it indicates that the prediction error of the usage speed. In this case, it means that both the first usage ratio and the second usage ratio may have large errors. Therefore, As the coefficient ( is a parameter greater than 0, for example, 0.01, which is used to prevent the denominator from being equal to 0 and making the loss function meaningless), to amplify the error between the first usage ratio and the annotation information, and As the conditional function value, to improve the training intensity and reduce the error. The conditional function values ​​of the dates are summed as part of the ostomy bag status loss function. When i=1, it is impossible to calculate the change in the first usage ratio between the i-th date and the i-1-th date, so we directly use As part of the pouch state loss function. On the other hand, the conditional function Indicates In the case of , otherwise , has a similar meaning to the above and will not be repeated here. The weighted sum of the above two items can be used to obtain the pouch state loss function.

[0127] In this way, based on the fact that the second sample patient's ostomy bag usage speed is theoretically faster, the loss function value can be selected through the conditional function, so as to increase the training intensity and efficiency when the ostomy bag usage speed prediction is wrong, reduce the error between the first usage ratio and the second usage ratio, and improve the model accuracy, so as to more accurately represent the impact of the fungus tube on the frequency of ostomy bag replacement.

[0128] According to one embodiment of the present invention, in step S109, a stoma health state loss function is obtained based on the first stoma sample feature information, the first stoma bag sample feature information, the second stoma sample feature information, and the second stoma bag sample feature information, and the third recurrent neural network model, including: splicing the first stoma sample feature information and the first stoma bag sample feature information of the same date to obtain the first state feature information of each date, and splicing the second stoma sample feature information and the second stoma bag sample feature information of the same date to obtain the second state feature information of each date; processing the first state feature information of each date through the third recurrent neural network model to obtain the fifth sample hidden state information of the last date, and processing the second state feature information of each date to obtain the sixth sample hidden state information of the last date; inputting the fifth sample hidden state information into the fifth fully connected layer to obtain the first sample stoma disease probability value, and inputting the sixth sample hidden state information into the fifth fully connected layer to obtain the second sample stoma disease probability value; obtaining the stoma health state loss function based on the first sample stoma disease probability value, the second sample stoma disease probability value, the probability labeling of the first sample patient's stoma disease, and the probability labeling of the second sample patient's stoma disease.

[0129] According to one embodiment of the present invention, the method for obtaining the first state feature information and the second state feature information is similar to the method for obtaining the above state feature information, and the method for obtaining the fifth sample hidden state information and the sixth sample hidden state information is similar to the method for obtaining the above third hidden state information, which will not be repeated here.

[0130] According to one embodiment of the present invention, the fifth sample hidden state information and the sixth sample hidden state information can be used to represent the overall state changes of the stoma and the stoma bag within a first preset number of dates. The fifth sample hidden state information can be processed by the fifth fully connected layer and then processed by the activation layer to obtain the first sample stoma disease probability value. Similarly, the second sample stoma disease probability value can be obtained. Further, the probability labeling of the stoma disease of the first sample patient and the probability labeling of the stoma disease of the second sample patient can be obtained, and the cross entropy loss function of the first sample stoma disease probability value and the probability labeling of the stoma disease of the first sample patient and the cross entropy loss function of the second sample stoma disease probability value and the probability labeling of the stoma disease of the second sample patient can be obtained respectively, and then the weighted summation is performed to obtain the stoma health state loss function.

[0131] According to one embodiment of the present invention, in step S110, based on the first sample hidden state information, the second sample hidden state information, the third sample hidden state information, the fourth sample hidden state information, the first stoma sample feature information, the first ostomy bag sample feature information, the second stoma sample feature information and the second ostomy bag sample feature information, and the third recurrent neural network model, a stoma care status loss function is obtained, including: based on the first sample hidden state information, the second sample hidden state information, the first stoma sample feature information, the first ostomy bag sample feature information and the third recurrent neural network model, obtaining the first sample stoma disease trend global state information; based on the third sample hidden state information, the fourth sample hidden state information, the second stoma sample feature information, the second ostomy bag sample feature information and the third recurrent neural network model type, obtain the global state information of the second sample stoma disease trend; input the first sample stoma disease trend global state information into the first fully connected layer, obtain the first sample stoma nursing treatment status probability value, and input the second sample stoma disease trend global state information into the first fully connected layer, obtain the second sample stoma nursing treatment status probability value; input the first sample stoma disease trend global state information into the sixth fully connected layer, obtain the first probability value that the first sample patient is a patient with a mushroom tube implanted, and input the second sample stoma disease trend global state information into the sixth fully connected layer, obtain the second probability value that the second sample patient is a patient with a mushroom tube implanted; according to the first sample stoma nursing treatment status probability value, the second sample stoma nursing treatment status probability value, the first probability value, and the second probability value, obtain the stoma nursing status loss function.

[0132] According to one embodiment of the present invention, the first sample stoma disease trend global status information and the second sample stoma disease trend global status information are obtained in a similar manner to the above-mentioned stoma care status characteristic information, which will not be repeated here. Moreover, similar to the above-mentioned stoma care treatment status information, the first sample stoma care treatment status probability value and the second sample stoma care treatment status probability value can be obtained through the first fully connected layer and the activation function. In addition, the first sample stoma disease trend global status information and the second sample stoma disease trend global status information can also be input into the sixth fully connected layer. After being processed by the activation function, a first probability value of the first sample patient being a patient with a fungus tube implanted and a second probability value of the second sample patient being a patient with a fungus tube implanted can be obtained. That is, the model determines the possibility of stoma disease and the frequency of ostomy bag replacement based on the various information previously obtained, and then determines whether the patient has a fungus tube implanted. Therefore, the judgment result can be used to assist in the training of each model, thereby improving the accuracy of the information obtained by each model.

[0133] According to one embodiment of the present invention, a stoma care treatment condition loss function is obtained based on the first sample stoma care treatment condition probability value, the second sample stoma care treatment condition probability value, the first probability value, and the second probability value, including: obtaining the stoma care condition loss function according to formula (3) , (3), in, is the first probability value, is the second probability value, The probability of stoma disease in the first sample patient is marked, The probability of stoma disease in the second sample patients is marked. is the probability value of the stoma care treatment status of the first sample, is the probability value of stoma care treatment status of the second sample, and is the preset weight.

[0134] According to one embodiment of the present invention, in formula (3), is a cross-entropy loss function based on the first probability value and the second probability value, wherein the first sample patient is a patient with a mushroom canal implant, whose label is 1, and the second sample patient is a patient without a mushroom canal implant, whose label is 0. Based on the respective labels and the first probability value and the second probability value, the above cross-entropy loss function can be obtained, which can represent the loss function for predicting whether a mushroom canal is implanted. is the cross entropy loss function of the probability value of the first sample stoma care treatment condition and the probability value of the first stoma diseased sample, The cross entropy loss function of the probability value of the second sample ostomy care treatment status and the probability value of the second ostomy disease sample is weighted and summed to obtain the loss function for predicting the disease probability. Since the training of the first fully connected layer is necessary, that is, the first fully connected layer is a necessary layer for obtaining ostomy care treatment status information, therefore, when constructing the ostomy care status loss function, the maximum function cannot be used for selection as in formula (1). Otherwise, it may be impossible to select the branch where the first fully connected layer is located for training, resulting in the inability to train the first fully connected layer and thus inaccurate ostomy care treatment status information. Therefore, the above loss function for predicting whether the mushroom tube is implanted can be multiplied with the loss function for predicting the disease probability, so that the branch where the first fully connected layer is located can be trained, and the loss function for predicting whether the mushroom tube is implanted can be used to enhance the training strength of the model before the first fully connected layer, further improving the accuracy of the model in determining the possibility of ostomy disease and the frequency of changing ostomy bags.

[0135] In this way, when constructing the loss function for stoma care status, the loss function for predicting whether a mushroom tube is implanted can be used for auxiliary training, thereby improving the training intensity of the model before the first fully connected layer. By multiplying, the situation where the first fully connected layer cannot be trained can be avoided, which can improve the training intensity and efficiency and comprehensively speed up the model accuracy.

[0136] According to one embodiment of the present invention, in step S111, a comprehensive loss function can be obtained based on the multiple loss functions determined above, for example, a weighted sum of the multiple loss functions is performed to obtain a sum loss function, and in step S112, the comprehensive loss function is back-propagated to train a convolutional neural network model, a second convolutional neural network model, a first recurrent neural network model, a second recurrent neural network model and a third recurrent neural network model, and the first fully connected layer can be trained. After multiple trainings, verification can be performed in a validation set consisting of multiple positive samples and negative samples. After the accuracy of the verification model meets the requirements (for example, after the accuracy of determining whether the ostomy care treatment status information in the validation set is abnormal meets the predetermined requirements), the training can be completed, and the trained first convolutional neural network model, the second convolutional neural network model, the first recurrent neural network model, the second recurrent neural network model and the third recurrent neural network model and the first fully connected layer can be used in the above process of determining the ostomy care treatment status information.

[0137] According to an embodiment of the present invention, the wound negative pressure nursing treatment status monitoring method utilizes a mushroom tube for negative pressure therapy, which not only reduces excreta flowing into the ostomy bag, thus reducing the frequency of bag changes, but also reduces contamination of the skin around the stoma by excreta, thereby lowering the probability of peristomal skin infection. Furthermore, by photographing the stoma and ostomy bag, the stoma's health status can be determined by comparing multiple dates of stoma and ostomy bag images to determine changes in the stoma's status. This eliminates the need to extract exudate from the peristomal skin, thus reducing patient pain. Based on the stoma and ostomy bag images, the stoma bag change frequency and stoma status can be comprehensively determined, thereby improving the accuracy of stoma nursing treatment status information. The fungiform tube can also be used to divert intestinal fluid and excreta flowing into the ostomy bag, thereby reducing the possibility of local skin irritation caused by excreta leakage. The frequency of ostomy bag replacement can also be reduced. The condition of the stoma and ostomy bag can be monitored by taking images of the stoma and ostomy bag daily and using multiple models. By utilizing the characteristics of the recurrent neural network model, the potential for health conditions in the stoma can be expressed using latent state information, improving monitoring accuracy and providing accurate alarm information to doctors and caregivers. When determining the stoma state loss function, a maximum function can be selected, thereby directly reducing the error between the first sample latent state information and the second sample latent state information, and indirectly strengthening model training by reducing the error of information such as the first sample state similarity, thereby reducing the error between the first sample latent state information and the second sample latent state information. This can increase training intensity and improve training efficiency and model performance. When determining the pouch status loss function, the loss function value can be selected using a conditional function based on the theoretically faster pouch usage rate of the second sample patient. This increases training intensity and efficiency when pouch usage rate predictions are incorrect, reduces the error between the first and second usage rates, and improves model accuracy, thereby more accurately representing the impact of the fungus tube on pouch replacement frequency. Furthermore, when constructing the stoma care status loss function, a loss function for predicting whether a fungus tube has been implanted can be used for auxiliary training, thereby increasing training intensity for the model before the first fully connected layer. By multiplying the loss function, this avoids situations where the first fully connected layer cannot be trained, thereby increasing training intensity and efficiency and overall improving model accuracy.

[0138] Figure 3 A block diagram of a wound negative pressure care treatment status monitoring system according to an embodiment of the present invention is exemplarily shown, wherein the system includes:

[0139] an implantation module, configured to capture an image of the stoma and an image of the stoma bag at the first preset moment on each day after a fungus-shaped tube is implanted at the stoma of an ileostomy patient for negative pressure therapy;

[0140] A first feature information module, configured to input the stoma image into a first convolutional neural network model to obtain first feature information of the stoma image;

[0141] A second feature information module is used to input the ostomy bag image into the second convolutional neural network model to obtain second feature information of the ostomy bag image;

[0142] Hidden state information module, used to store the current date and the previous date The first feature information of each date is input into the first recurrent neural network model in date order to obtain the first hidden state information corresponding to multiple dates, and the current date and the dates before it are combined into a single feature information. The first feature information of the dates is input into the second recurrent neural network model in reverse order of the dates to obtain the second hidden state information corresponding to the multiple dates, wherein, is the first preset quantity;

[0143] A status characteristic information module, configured to concatenate the first characteristic information and the second characteristic information to obtain status characteristic information for each date;

[0144] The third hidden state information module is used to store the current date and the date before it. The state feature information is input into the third recurrent neural network model to obtain the third latent state information corresponding to the current date;

[0145] The stoma nursing treatment status information module is used to obtain the stoma nursing treatment status information corresponding to the current date based on the first latent state information, the second latent state information, and the third latent state information;

[0146] The alarm information module is used to continuously obtain daily stoma care and treatment status information and generate alarm information when the stoma care and treatment status information is abnormal.

[0147] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0148] Those skilled in the art will appreciate that the embodiments of the present invention described above and shown in the accompanying drawings are intended to be illustrative only and are not intended to limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and illustrated in the embodiments. Any variations or modifications may be made to the embodiments of the present invention without departing from the principles described.

[0149] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for monitoring the status of wound negative pressure nursing treatment, characterized in that: include: After a fungus-shaped tube was implanted at the stoma for negative pressure therapy in patients with ileostomy, images of the stoma and stoma bag were taken at the first preset moment on each day; Inputting the stoma image into a first convolutional neural network model to obtain first feature information of the stoma image; Inputting the ostomy bag image into a second convolutional neural network model to obtain second feature information of the ostomy bag image; The current date and the date before it The first feature information of each date is input into the first recurrent neural network model in date order to obtain the first hidden state information corresponding to multiple dates, and the current date and the dates before it are combined into a single feature information. The first feature information of the dates is input into the second recurrent neural network model in reverse order of the dates to obtain the second hidden state information corresponding to the multiple dates, wherein, is the first preset quantity; The first characteristic information and the second characteristic information are combined to obtain the status characteristic information of each date; The current date and the date before it The state feature information is input into the third recurrent neural network model to obtain the third latent state information corresponding to the current date; Obtaining stoma care treatment status information corresponding to the current date according to the first latent state information, the second latent state information, and the third latent state information; Continuously obtain daily stoma care treatment status information, and generate alarm information when the stoma care treatment status information is abnormal.

2. The method for monitoring the wound negative pressure nursing treatment status according to claim 1, characterized in that: According to the first latent state information, the second latent state information, and the third latent state information, stoma care treatment status information corresponding to the current date is obtained, including: Obtain the state similarity between the first latent state information and the second latent state information of the same date; The state similarities corresponding to multiple dates are combined into a state similarity vector; The first hidden state information corresponding to the current date and the The second latent state information corresponding to the dates is spliced ​​to obtain the stoma disease trend state information, where the current date is the jth date; The state similarity vector and the stoma disease trend state information are spliced ​​to obtain the stoma disease trend global state information; The global state information of stoma disease trend and the third latent state information corresponding to the current date are combined to obtain characteristic information of stoma care status; The stoma care status feature information is input into the first fully connected layer to obtain the stoma care treatment status information.

3. The method for monitoring the wound negative pressure nursing treatment status according to claim 1, characterized in that: The method further comprises: acquiring first sample stoma images and first sample ostomy bag images on multiple dates for multiple first sample patients implanted with a fungiform tube for negative pressure therapy; acquiring second sample stoma images and second sample ostomy bag images on multiple dates for multiple second sample patients who do not have a fungiform tube implanted for negative pressure therapy; Obtaining, by a first convolutional neural network model, first stoma sample feature information of the first sample stoma image and second stoma sample feature information of the second sample stoma image; Obtaining, by a second convolutional neural network model, first ostomy bag sample feature information of the first sample ostomy bag image and second ostomy bag sample feature information of the second sample ostomy bag image; Processing the first stoma sample feature information through a first recurrent neural network model to obtain first sample latent state information, and processing the first stoma sample feature information through a second recurrent neural network model to obtain second sample latent state information; Processing the second stoma sample feature information through the first recurrent neural network model to obtain third sample hidden state information, and processing the second stoma sample feature information through the second recurrent neural network model to obtain fourth sample hidden state information; Obtaining a stoma state loss function according to the first sample hidden state information, the second sample hidden state information, the third sample hidden state information, and the fourth sample hidden state information; Obtaining an ostomy bag state loss function according to the first ostomy bag sample feature information and the second ostomy sample feature information; Obtaining a stoma health state loss function according to the first stoma sample feature information, the first stoma bag sample feature information, the second stoma sample feature information, and the second stoma bag sample feature information, and a third recurrent neural network model; Obtaining a stoma care status loss function according to the first sample latent state information, the second sample latent state information, the third sample latent state information, the fourth sample latent state information, the first stoma sample feature information, the first stoma bag sample feature information, the second stoma sample feature information, and the second stoma bag sample feature information, and a third recurrent neural network model; Obtain a comprehensive loss function based on the stoma state loss function, the stoma bag state loss function, the stoma health state loss function, and the stoma care state loss function; Through the comprehensive loss function, the first convolutional neural network model, the second convolutional neural network model, the first recurrent neural network model, the second recurrent neural network model and the third recurrent neural network model are trained.

4. The method for monitoring the wound negative pressure nursing treatment status according to claim 3, characterized in that: According to the first sample hidden state information, the second sample hidden state information, the third sample hidden state information and the fourth sample hidden state information, a stoma state loss function is obtained, including: Input the first sample hidden state information corresponding to the last date into the second fully connected layer to obtain the first stoma disease sample probability value of the first sample patient, and input the third sample hidden state information corresponding to the last date into the second fully connected layer to obtain the second stoma disease sample probability value of the second sample patient; Obtaining the first sample state similarity between the first sample hidden state information and the second sample hidden state information corresponding to the same date, and obtaining the second sample state similarity between the third sample hidden state information and the fourth sample hidden state information corresponding to the same date; Obtaining global state information of the stoma disease trend of the first sample according to the first sample state similarity, the first sample hidden state information, and the second sample hidden state information; Obtaining global state information of the stoma disease trend of the second sample according to the second sample state similarity, the third sample hidden state information, and the fourth sample hidden state information; Inputting the global state information of the stoma disease trend of the first sample into the third fully connected layer to obtain a third stoma disease sample probability value, and inputting the global state information of the stoma disease trend of the second sample into the third fully connected layer to obtain a fourth stoma disease sample probability value; A stoma state loss function is obtained according to the first stoma diseased sample probability value, the second stoma diseased sample probability value, the third stoma diseased sample probability value, and the fourth stoma diseased sample probability value.

5. The method for monitoring the wound negative pressure nursing treatment status according to claim 4, characterized in that: According to the first stoma disease sample probability value, the second stoma disease sample probability value, the third stoma disease sample probability value, and the fourth stoma disease sample probability value, a stoma state loss function is obtained, including: According to the formula ， Obtain stoma state loss function ,in, The probability of stoma disease in the first sample patient is marked, The probability of stoma disease in the second sample patients is marked. is the probability value of the first stoma disease sample, is the probability value of the third stoma disease sample, is the probability value of the second stoma disease sample, is the probability value of the fourth stoma disease sample, and is the preset weight, and max is the maximum value function.

6. The method for monitoring the wound negative pressure nursing treatment status according to claim 3, characterized in that: According to the first stoma bag sample feature information and the second stoma sample feature information, an stoma bag state loss function is obtained, including: Input the first ostomy bag sample feature information of each date into the fourth fully connected layer to obtain the first usage ratio of ostomy bags for the first sample patients on each date, and input the second ostomy bag sample feature information of each date into the fourth fully connected layer to obtain the second usage ratio of ostomy bags for the second sample patients on each date; According to the formula ， Obtaining the pouch state loss function ,in, is the first usage ratio of ostomy bag for the first sample patient on the i-th day, The first usage ratio of ostomy bag for the first sample patients on date 1, is the first usage ratio of ostomy bag for the first sample patient on the i-1th date, The labeling information for the proportion of stoma bags used by the first sample patients on the first day, The labeling information of the usage ratio of ostomy bags for the first sample patient on the i-th day, is the second usage ratio of ostomy bag for the second sample patients on the i-th day, is the second usage ratio of ostomy bags for the second sample patients on the i-1th date, The labeling information for the usage ratio of ostomy bags for the second sample patients on the i-th date, The second usage ratio of ostomy bag for the second sample patients on date 1, The labeling information for the proportion of stoma bags used by the second sample patients on the first day, is the first preset number, and is the preset weight, is the preset parameter, and if is the conditional function.

7. The method for monitoring the wound negative pressure nursing treatment status according to claim 3, characterized in that: According to the first stoma sample feature information, the first stoma bag sample feature information, the second stoma sample feature information, the second stoma bag sample feature information, and the third recurrent neural network model, a stoma health state loss function is obtained, including: The first stoma sample characteristic information and the first stoma bag sample characteristic information of the same date are spliced ​​together to obtain the first state characteristic information of each date, and the second stoma sample characteristic information and the second stoma bag sample characteristic information of the same date are spliced ​​together to obtain the second state characteristic information of each date; The first state feature information of each date is processed by the third recurrent neural network model to obtain the fifth sample hidden state information of the last date, and the second state feature information of each date is processed to obtain the sixth sample hidden state information of the last date; Inputting the fifth sample hidden state information into the fifth fully connected layer to obtain the first sample stoma disease probability value, and inputting the sixth sample hidden state information into the fifth fully connected layer to obtain the second sample stoma disease probability value; A stoma health state loss function is obtained according to the stoma disease probability value of the first sample, the stoma disease probability value of the second sample, the probability labeling of the stoma disease of the first sample patient, and the probability labeling of the stoma disease of the second sample patient.

8. The method for monitoring wound negative pressure nursing treatment status according to claim 3, characterized in that: According to the first sample hidden state information, the second sample hidden state information, the third sample hidden state information, the fourth sample hidden state information, the first stoma sample feature information, the first ostomy bag sample feature information, the second stoma sample feature information and the second ostomy bag sample feature information, and the third recurrent neural network model, a stoma care status loss function is obtained, including: Obtaining global state information of the stoma disease trend of the first sample according to the first sample hidden state information, the second sample hidden state information, the first stoma sample feature information, the first stoma bag sample feature information, and the third recurrent neural network model; Obtaining global state information of the stoma disease trend of the second sample according to the third sample hidden state information, the fourth sample hidden state information, the second stoma sample feature information, the second stoma bag sample feature information, and the third recurrent neural network model; Inputting the first sample stoma disease trend global state information into the first fully connected layer to obtain the first sample stoma nursing treatment status probability value, and inputting the second sample stoma disease trend global state information into the first fully connected layer to obtain the second sample stoma nursing treatment status probability value; Inputting the first sample stoma disease trend global state information into the sixth fully connected layer to obtain a first probability value that the first sample patient is a patient with a mushroom tube implanted, and inputting the second sample stoma disease trend global state information into the sixth fully connected layer to obtain a second probability value that the second sample patient is a patient with a mushroom tube implanted; A stoma care condition loss function is obtained according to the first sample stoma care treatment condition probability value, the second sample stoma care treatment condition probability value, the first probability value, and the second probability value.

9. The method for monitoring the wound negative pressure nursing treatment status according to claim 8, characterized in that: Obtaining a stoma care status loss function according to the first sample stoma care treatment status probability value, the second sample stoma care treatment status probability value, the first probability value, and the second probability value includes: According to the formula ， Obtain stoma care status loss function ,in, is the first probability value, is the second probability value, The probability of stoma disease in the first sample patient is marked, The probability of stoma disease in the second sample patients is marked. is the probability value of the stoma care treatment status of the first sample, is the probability value of stoma care treatment status of the second sample, and is the preset weight.

10. A wound negative pressure nursing treatment status monitoring system, characterized in that: include: an implantation module, configured to capture an image of the stoma and an image of the stoma bag at the first preset moment on each day after a fungus-shaped tube is implanted at the stoma of an ileostomy patient for negative pressure therapy; A first feature information module, configured to input the stoma image into a first convolutional neural network model to obtain first feature information of the stoma image; A second feature information module is used to input the ostomy bag image into the second convolutional neural network model to obtain second feature information of the ostomy bag image; Hidden state information module, used to store the current date and the previous date The first feature information of each date is input into the first recurrent neural network model in date order to obtain the first hidden state information corresponding to multiple dates, and the current date and the dates before it are combined into a single feature information. The first feature information of the dates is input into the second recurrent neural network model in reverse order of the dates to obtain the second hidden state information corresponding to the multiple dates, wherein, is the first preset quantity; A status characteristic information module, configured to concatenate the first characteristic information and the second characteristic information to obtain status characteristic information for each date; The third hidden state information module is used to store the current date and the date before it. The state feature information is input into the third recurrent neural network model to obtain the third latent state information corresponding to the current date; The stoma nursing treatment status information module is used to obtain the stoma nursing treatment status information corresponding to the current date based on the first latent state information, the second latent state information, and the third latent state information; The alarm information module is used to continuously obtain daily stoma care and treatment status information and generate alarm information when the stoma care and treatment status information is abnormal.

Images