Artificial Intelligence-Based Method and System for Identifying Postoperative Stoma Complications

Abstract

The present invention provides a method and system for identifying postoperative stoma complications based on artificial intelligence, relating to the technical field of image processing. The method includes: obtaining a stoma image through an image acquisition device; determining whether a stoma patient has postoperative stoma complications; if the stoma patient has postoperative stoma complications, obtaining nutritional index data of the stoma patient by extracting the blood of the stoma patient; inputting the nutritional index data and the stoma image into a trained probability prediction model for peristomal skin complications to obtain predicted peristomal skin complication probability data; obtaining the stoma location and stoma type; inputting the stoma location, the stoma type and the stoma image into a trained probability prediction model for stoma complications to obtain predicted stoma complication probability data; and determining the complication type. According to the present invention, the type of postoperative stoma complications can be identified, and the accuracy and comprehensiveness of the identification of the type of postoperative stoma complications can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to a method and system for identifying postoperative stoma complications based on artificial intelligence. Background Art

[0002] In the related art, CN117982115A discloses an intelligent memory function stoma belt and its usage method, including: obtaining the physiological data of stoma patients to judge whether there is a risk of complications for stoma patients; further analyzing the physiological characterization values; analyzing the number of times of complications when different physiological characterization values occur to judge whether there is a risk of complications for stoma patients; analyzing the historical data of the target physiological data before each occurrence of complications to evaluate the correlation between the target physiological data and the occurrence of complications; when obtaining the target physiological data related to the occurrence of complications, further analyzing the target physiological data to judge whether there is a risk of complications for stoma patients; this solution can establish personalized management for each patient, make the early warning more accurate and reliable, reduce the risks brought by complications, and thus improve the safety and rehabilitation effect of patients.

[0003] CN118557275A discloses a stoma system. The stoma system includes a stoma device. The stoma device can cut tissues to form a stoma. The stoma device is also used for electrically connecting to an external energy source to ablate the tissues around the stoma during and / or after the stoma formation process. By strengthening the tissues around the stoma after the stoma formation, it is possible to prevent stoma retraction or closure, thereby ensuring better postoperative effects.

[0004] Therefore, in the related art, although it is possible to judge whether there is a risk of complications for stoma patients, the related art does not consider the influence of the predicted probability data of skin complications around the stoma and the predicted probability data of stoma complications on the identification of postoperative stoma complication types, that is, it is impossible to identify the types of postoperative stoma complications based on the predicted probability data of skin complications around the stoma and the predicted probability data of stoma complications.

[0005] The information disclosed in the background art part of the present application is only intended to deepen the understanding of the general background art of the present application, and should not be regarded as an admission or any form of suggestion that this information constitutes the prior art already known to those skilled in the art. Summary of the Invention

[0006] The present invention provides a method and system for identifying postoperative stoma complications based on artificial intelligence, which can solve the technical problem that the related art cannot identify the types of postoperative stoma complications based on the predicted probability data of skin complications around the stoma and the predicted probability data of stoma complications.

[0007] According to a first aspect of the present invention, there is provided an artificial intelligence-based method for identifying postoperative stoma complications, including: acquiring a stoma image through an image acquisition device; determining whether a stoma patient has postoperative stoma complications based on the stoma image; if the stoma patient has postoperative stoma complications, acquiring nutritional index data of the stoma patient by extracting the stoma patient's blood; inputting the nutritional index data and the stoma image into a trained probability prediction model for peristomal skin complications to obtain predicted probability data for peristomal skin complications; acquiring the stoma location and stoma type; inputting the stoma location, the stoma type, and the stoma image into a trained probability prediction model for stoma complications to obtain predicted probability data for stoma complications; and determining the complication type based on the predicted probability data for peristomal skin complications and the predicted probability data for stoma complications.

[0008] Further, in the stoma image, identifying whether the stoma height is within a preset range through an image detection model to determine the stoma height recognition result; in the stoma image, identifying whether the stoma color is beef red or pink through an image detection model to determine the stoma color recognition result; in the stoma image, identifying whether the stoma shape is circular or elliptical through an image detection model to determine the stoma shape recognition result; in the stoma image, identifying whether the skin around the stoma is intact or flat through an image detection model to determine the peristomal skin recognition result; determining a first stoma judgment parameter based on the stoma height recognition result, the stoma color recognition result, the stoma shape recognition result, and the peristomal skin recognition result; if the first stoma judgment parameter is less than a set threshold, determining that the postoperative stoma patient has postoperative stoma complications; and if the first stoma judgment parameter is greater than or equal to the set threshold, determining that the postoperative stoma patient has no postoperative stoma complications.

[0009] Further, determining a first stoma judgment parameter based on the stoma height recognition result, the stoma color recognition result, the stoma shape recognition result, and the peristomal skin recognition result includes: determining the first stoma judgment parameter A1 according to the formula A1 = α1H o +α2C o +α3F o +α4D o wherein H o is the stoma height recognition result, C o is the stoma color recognition result, F o is the stoma shape recognition result, D o is the peristomal skin recognition result, and α1, α2, α3, and α4 are preset weights.

[0010] Further, the training steps of the peristomal skin complication probability prediction model include: extracting multiple sample stoma patients to obtain sample nutritional index data, sample stoma images, and sample peristomal skin complication probability data; determining, among multiple pixel points of the sample stoma images, multiple abnormal color pixel points corresponding to chromaticity higher than a first preset chromaticity threshold; performing connected component analysis on the multiple abnormal color pixel points to obtain a first sample abnormal color region and determining the area of the first sample abnormal color region; performing feature extraction processing on the peristomal skin region through the skin feature encoding layer of a trained image recognition neural network model to obtain a sample feature vector of the peristomal skin in the sample stoma image; processing the sample nutritional index data and the sample stoma image through the peristomal skin complication probability prediction model to obtain sample predicted peristomal skin complication probability data; determining a loss function of the peristomal skin complication probability prediction model according to the sample nutritional index data, the area of the first sample abnormal color region, the sample feature vector of the peristomal skin, the sample peristomal skin complication probability data, and the sample predicted peristomal skin complication probability data; and training the peristomal skin complication probability prediction model according to the loss function of the peristomal skin complication probability prediction model to obtain the trained peristomal skin complication probability prediction model.

[0011] Further, determining the loss function of the peristomal skin complication probability prediction model according to the sample nutritional index data, the area of the first sample abnormal color region, the sample feature vector of the peristomal skin, the sample peristomal skin complication probability data, and the sample predicted peristomal skin complication probability data includes: according to the formula determining the loss function Loss1 of the peristomal skin complication probability prediction model, where R j is the sample feature vector of the peristomal skin in the sample stoma image of the j-th sample stoma patient, R a is the feature vector of the peristomal skin in the stoma image of the stoma patient, R j T is the transposed vector of R j , S 1,j is the area of the first sample abnormal color region of the j-th sample stoma patient, S 1,a is the area of the first abnormal color region of the stoma patient, e j,i is the i-th sample nutritional index data of the j-th sample stoma patient, e T is the standard nutritional index data, G j is the sample peristomal skin complication probability data of the j-th sample stoma patient, G j,pThe probability data of the sample predicted peristomal skin complications for the j-th sample stoma patient, N is the number of types of nutritional indicators, m x is the number of sample stoma patients in the x-th batch, U is the number of sample batches, i ≤ N, j ≤ m x , x ≤ U, and i, j, x, N, m x and U are all positive integers.

[0012] Furthermore, the training steps of the stoma complication probability prediction model include: by extracting multiple sample stoma patients, obtaining the sample stoma location, sample stoma type, sample stoma image, and sample stoma complication probability data; through the stoma feature encoding layer of the trained image recognition neural network model, performing feature extraction processing on the stoma to obtain the sample feature vector of the stoma in the sample stoma image; through the stoma complication probability prediction model, processing the sample stoma location, the sample stoma type, and the sample stoma image to obtain the sample predicted stoma complication probability data; according to the sample stoma location, the sample stoma type, the sample feature vector of the stoma, the sample stoma complication probability data, and the sample predicted stoma complication probability data, determining the loss function of the stoma complication probability prediction model; according to the loss function of the stoma complication probability prediction model, training the stoma complication probability prediction model to obtain the trained stoma complication probability prediction model.

[0013] Furthermore, determining the loss function of the stoma complication probability prediction model according to the sample stoma location, the sample stoma type, the sample feature vector of the stoma, the sample stoma complication probability data, and the sample predicted stoma complication probability data includes: according to the formula determining the loss function Loss2 of the stoma complication probability prediction model, where W j is the sample stoma location of the j-th sample stoma patient, W a is the stoma location of the stoma patient, W j T is the transposed vector of W j K j is the sample stoma type of the j-th sample stoma patient, K a is the stoma type of the stoma patient, K j T is the transposed vector of K j O j is the sample feature vector of the stoma in the sample stoma image of the j-th sample stoma patient, O a is the feature vector of the stoma in the stoma image of the stoma patient, O j T is O jThe transposed vector of, B j is the sample stoma complication probability data of the j-th sample stoma patient, B j,p is the sample predicted stoma complication probability data of the j-th sample stoma patient, m x is the number of sample stoma patients in the x-th batch, U is the number of sample batches, j ≤ m x , x ≤ U, and j, x, m x and U are all positive integers.

[0014] According to the second aspect of the present invention, there is provided an artificial intelligence-based postoperative stoma complication recognition system, including: a stoma image module for acquiring a stoma image through an image acquisition device; a judgment module for determining whether a stoma patient has a postoperative stoma complication according to the stoma image; a nutritional index data module for, if a stoma patient has a postoperative stoma complication, acquiring the nutritional index data of the stoma patient by extracting the blood of the stoma patient; a predicted peristomal skin complication probability data module for inputting the nutritional index data and the stoma image into a trained peristomal skin complication probability prediction model to obtain predicted peristomal skin complication probability data; a stoma position and stoma type module for acquiring the stoma position and stoma type; a predicted stoma complication probability data module for inputting the stoma position, the stoma type and the stoma image into a trained stoma complication probability prediction model to obtain predicted stoma complication probability data; a complication type module for determining the complication type according to the predicted peristomal skin complication probability data and the predicted stoma complication probability data.

[0015] Technical effects: According to the present invention, by using the stoma image, it can be determined whether a stoma patient has postoperative stoma complications, which can improve the inspection efficiency and at the same time reduce the probability of misidentification. Through the information in multiple dimensions of the stoma image, nutritional index data, stoma location, and stoma type, the condition of the stoma patient can be more comprehensively understood, so as to predict the probabilities of peristomal skin complications and stoma complications. According to the predicted peristomal skin complication probability data and the predicted stoma complication probability data, the type of postoperative stoma complications is identified, improving the accuracy and comprehensiveness of the identification of the type of postoperative stoma complications. When determining the first stoma judgment parameter, the first stoma judgment parameter can be determined through the stoma height recognition result, stoma color recognition result, stoma shape recognition result, and peristomal skin recognition result. An artificial intelligence model is used to detect the height, color, shape, and the condition of the surrounding skin of the stoma, and based on the detection results, it is determined whether the stoma patient has postoperative stoma complications, which helps to more accurately reflect the overall health condition of the stoma, thereby improving the comprehensiveness and accuracy of the first judgment parameter. When determining the loss function of the peristomal skin complication probability prediction model, the influence of the sample nutritional index data on the probability of peristomal skin complications can be used to determine the influence of the above data on the error of the sample predicted peristomal skin complication probability data. Then, based on this influence and the relative error between the sample peristomal skin complication probability data and the sample predicted peristomal skin complication probability data, and based on the characteristic that the more similar the peristomal skin condition of the stoma patient is to that of the sample stoma patient, the greater the reference value, and the characteristic that the shorter the time interval from the first batch, the lower the accuracy, weights are set. Thus, the errors output by the peristomal skin complication probability prediction model of multiple sample nutritional index data of each sample stoma patient in the x-th batch are weighted and summed to obtain the loss function, so as to improve the design accuracy and objectivity of the loss function, thereby improving the training efficiency during the training process and the accuracy of the peristomal skin complication probability prediction model. When determining the loss function of the stoma complication probability prediction model, based on the relative error between the sample stoma complication probability data and the sample predicted stoma complication probability data, and based on the characteristic that the more similar the stoma condition of the stoma patient is to that of the sample stoma patient, the greater the reference value, and the characteristic that the shorter the time interval from the first batch, the lower the accuracy, weights are set. Thus, the errors output by the stoma complication probability prediction model of each sample stoma patient in the x-th batch are weighted and summed to obtain the loss function, so as to improve the design accuracy and objectivity of the loss function, thereby improving the training efficiency during the training process and the accuracy of the stoma complication probability prediction model.

[0016] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and do not limit the present invention. According to the following detailed description of the exemplary embodiments with reference to the accompanying drawings, other features and aspects of the present invention will be clearer. Brief Description of the Drawings

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other embodiments can also be obtained based on these drawings;

[0018] Figure 1 Exemplarily shows a schematic flowchart of an artificial intelligence-based method for identifying postoperative stoma complications according to an embodiment of the present invention;

[0019] Figure 2 Exemplarily shows a block diagram of an artificial intelligence-based system for identifying postoperative stoma complications according to an embodiment of the present invention. Detailed Embodiments

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, rather than all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0021] The following will detail the technical solutions of the present invention with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0022] Figure 1 Exemplarily shows a schematic flowchart of an artificial intelligence-based method for identifying postoperative stoma complications according to an embodiment of the present invention, and the method includes:

[0023] Step S101, obtaining a stoma image through an image acquisition device;

[0024] Step S102, determining whether the stoma patient has postoperative stoma complications according to the stoma image;

[0025] Step S103, if the stoma patient has postoperative stoma complications, obtaining the nutritional index data of the stoma patient by extracting the blood of the stoma patient;

[0026] Step S104, inputting the nutritional index data and the stoma image into a trained probability prediction model for peristomal skin complications to obtain probability data for predicting peristomal skin complications;

[0027] Step S105, obtain the stoma location and stoma type;

[0028] Step S106, input the stoma location, the stoma type, and the stoma image into a trained stoma complication probability prediction model to obtain predicted stoma complication probability data;

[0029] Step S107, determine the complication type according to the predicted peristomal skin complication probability data and the predicted stoma complication probability data.

[0030] According to the artificial intelligence-based postoperative stoma complication identification method of the embodiments of the present invention, by using a stoma image to determine whether a stoma patient has postoperative stoma complications, the inspection efficiency can be improved, and at the same time, the probability of misidentification can be reduced. Through information from multiple dimensions such as the stoma image, nutritional index data, stoma location, and stoma type, the condition of the stoma patient can be more comprehensively understood, thereby predicting the probabilities of peristomal skin complications and stoma complications. According to the predicted peristomal skin complication probability data and the predicted stoma complication probability data, the identification of postoperative stoma complication types is performed to improve the accuracy and comprehensiveness of the identification of postoperative stoma complication types.

[0031] According to an embodiment of the present invention, in step S101, a high-definition camera can be used to shoot the stoma location of the stoma patient to collect the stoma image of the stoma patient, and the height, shape, color of the stoma, and the condition of the surrounding skin can be clearly captured.

[0032] According to an embodiment of the present invention, in step S102, according to the stoma image, it is determined whether the stoma patient has postoperative stoma complications.

[0033] According to an embodiment of the present invention, step S102 includes: in the stoma image, using an image detection model to identify whether the stoma height is within a preset range to determine the stoma height recognition result; in the stoma image, using an image detection model to identify whether the stoma color is beef red or pink to determine the stoma color recognition result; in the stoma image, using an image detection model to identify whether the stoma shape is circular or elliptical to determine the stoma shape recognition result; in the stoma image, using an image detection model to identify whether the skin around the stoma is intact or flat to determine the peristomal skin recognition result; according to the stoma height recognition result, the stoma color recognition result, the stoma shape recognition result, and the peristomal skin recognition result, determine the first stoma judgment parameter; if the first stoma judgment parameter is less than a set threshold, it is determined that the postoperative stoma patient has postoperative stoma complications; if the first stoma judgment parameter is greater than or equal to the set threshold, it is determined that the postoperative stoma patient has no postoperative stoma complications.

[0034] According to an embodiment of the present invention, the image detection model belongs to a type of deep learning model. The image detection model is trained with historical data so that it can identify whether the stoma in the stoma image is a normal stoma. An abnormal stoma will cause skin complications and stoma complications around the stoma. The image detection model is used to identify whether the stoma height is within a preset range. If the stoma height is within the preset range, the stoma height recognition result is 1. If the stoma height is not within the preset range, the stoma height recognition result is 0. The image detection model is used to identify whether the stoma color is beef red or pink. If the stoma color is beef red or pink, the stoma color recognition result is 1. If the stoma color is not beef red or pink, the stoma color recognition result is 0. The image detection model is used to identify whether the stoma shape is circular or elliptical. If the stoma shape is circular or elliptical, the stoma shape recognition result is 1. If the stoma shape is not circular or elliptical, the stoma shape recognition result is 0. The image detection model is used to identify whether the skin around the stoma is intact or flat. If the skin around the stoma is intact or flat, the stoma surrounding skin recognition result is 1. If the skin around the stoma is not intact or not flat, the stoma surrounding skin recognition result is 0. According to the stoma height recognition result, the stoma color recognition result, the stoma shape recognition result, and the stoma surrounding skin recognition result, a first stoma judgment parameter is determined. The first stoma judgment parameter is compared with a set threshold. If the first stoma judgment parameter is less than the set threshold, it is determined that the postoperative stoma patient has postoperative stoma complications. If the first stoma judgment parameter is greater than or equal to the set threshold, it is determined that the postoperative stoma patient has no postoperative stoma complications.

[0035] According to an embodiment of the present invention, determining the first stoma judgment parameter according to the stoma height recognition result, the stoma color recognition result, the stoma shape recognition result, and the stoma surrounding skin recognition result includes: determining the first stoma judgment parameter A1 according to formula (1).

[0036] A1 = α1H o + α2C o + α3F o + α4D o (1).

[0037] Wherein, H o is the stoma height recognition result, C o is the stoma color recognition result, F o is the stoma shape recognition result, D o is the stoma surrounding skin recognition result. α1, α2, α3, and α4 are preset weights.

[0038] According to an embodiment of the present invention, in formula (1), H o is the recognition result of the stoma height. When the stoma height is within the preset range, it indicates that the stoma height is similar to that of a normal stoma, and the value of H o is 1. On the contrary, it indicates that the stoma height is abnormal and there may be postoperative stoma complications, and the value of H o is 0. C o is the recognition result of the stoma color. When the stoma color is beef red or pink, it indicates that the stoma color is similar to that of a normal stoma, and the value of C o is 1. On the contrary, it indicates that the stoma color is abnormal and there may be postoperative stoma complications, and the value of C o is 0. F o is the recognition result of the stoma shape. When the stoma shape is circular or oval, it indicates that the stoma shape is similar to that of a normal stoma, and the value of F o is 1. On the contrary, it indicates that the stoma shape is abnormal and there may be postoperative stoma complications, and the value of F o is 0. D o is the recognition result of the skin around the stoma. When the skin around the stoma is intact or flat, it indicates that the skin around the stoma is similar to that of a normal stoma, and the value of D o is 1. On the contrary, it indicates that the skin around the stoma is abnormal and there may be postoperative stoma complications, and the value of D o is 0. By performing weighted summation on H o , C o , F o and D o , the first stoma judgment parameter can be obtained. The larger the first stoma judgment parameter, the lower the possibility that the postoperative stoma patient has postoperative stoma complications. The smaller the first stoma judgment parameter, the higher the possibility that the postoperative stoma patient has postoperative stoma complications.

[0039] In this way, the first stoma judgment parameter can be determined through the recognition results of the stoma height, stoma color, stoma shape, and the skin around the stoma. The artificial intelligence model is used to detect the height, color, shape, and the condition of the skin around the stoma, and based on the detection results, it is determined whether the stoma patient has postoperative stoma complications, which helps to more accurately reflect the overall health status of the stoma, thereby improving the comprehensiveness and accuracy of the first judgment parameter.

[0040] According to an embodiment of the present invention, in step S103, when it is confirmed that the stoma patient has postoperative stoma complications, preparations are made for blood extraction. The various nutritional components (such as hemoglobin, albumin, electrolytes, etc.) in the blood of the stoma patient are accurately measured to obtain the nutritional index data of the stoma patient.

[0041] According to an embodiment of the present invention, in step S104, the nutritional index data and the stoma image are input into a trained probability prediction model for peristomal skin complications. The probability prediction model for peristomal skin complications can be a neural network model, which is trained based on a large amount of sample data through machine learning or deep learning techniques to obtain the probability data for predicting peristomal skin complications.

[0042] According to an embodiment of the present invention, the training steps of the probability prediction model for peristomal skin complications include: by extracting multiple sample stoma patients, obtaining sample nutritional index data, sample stoma images, and sample peristomal skin complication probability data; determining, among multiple pixel points of the sample stoma image, multiple abnormal color pixel points whose corresponding chromaticity is higher than a first preset chromaticity threshold; performing connected component analysis on the multiple abnormal color pixel points to obtain a first sample abnormal color region and determining the area of the first sample abnormal color region; through the skin feature encoding layer of a trained image recognition neural network model, performing feature extraction processing on the peristomal skin region to obtain a sample feature vector of the peristomal skin in the sample stoma image; through the probability prediction model for peristomal skin complications, processing the sample nutritional index data and the sample stoma image to obtain sample predicted peristomal skin complication probability data; according to the sample nutritional index data, the area of the first sample abnormal color region, the sample feature vector of the peristomal skin, the sample peristomal skin complication probability data, and the sample predicted peristomal skin complication probability data, determining the loss function of the probability prediction model for peristomal skin complications; according to the loss function of the probability prediction model for peristomal skin complications, training the probability prediction model for peristomal skin complications to obtain the trained probability prediction model for peristomal skin complications.

[0043] According to an embodiment of the present invention, the multiple extracted sample stoma patients are divided into different batches of sample stoma patients, and the number of sample stoma patients in each batch is the same, so as to perform training of the probability prediction model for peristomal skin complications in different batches. For example, m1 + m2 + … + m U = M, where M is the number of sample stoma patients, and m1, m2, …, m UThe number of sample stoma patients in the 1st, 2nd, …, Uth batches respectively. For each sample stoma patient, sample stoma peristomal skin complication probability data can be obtained from the recorded occurrence probability of peristomal skin complications. For example, if a sample stoma patient has peristomal skin complications, the sample stoma peristomal skin complication probability data of this sample stoma patient is 1; if a sample stoma patient has no peristomal skin complications, the sample stoma peristomal skin complication probability data of this sample stoma patient is 0. The first preset chromaticity threshold can be set to be 20% or 30% higher than the chromaticity of normal skin color, etc., and the present invention does not limit this. The connectivity relationship between multiple abnormal color pixel points can be determined to form a connected domain. Herein, the connected domain refers to a set of abnormal color pixel points with an interconnected or adjacent relationship, and the boundary and scope of the connected domain can be used as the boundary and scope of the first sample abnormal color area. Based on this, the area of the first sample abnormal color area is determined. The image recognition neural network model can be a convolutional neural network model. Using the skin feature encoding layer of the image recognition neural network model, feature extraction processing is performed on the peristomal skin area to obtain the sample feature vector of the peristomal skin in the sample stoma image. The sample feature vector of the peristomal skin usually includes various features of the peristomal skin of the stoma patient, such as skin color, skin flatness, etc., and can be used to analyze and identify the peristomal skin condition. The sample nutritional index data affects the probability of peristomal skin complications. For example, when the sample nutritional index data does not meet the standard, it may lead to poor healing of the peristomal skin and increase the occurrence probability of complications such as mold infection and dermatitis. The peristomal skin complication probability prediction model can predict the sample predicted peristomal skin complication probability data based on the relationship between the above sample nutritional index data and the probability of peristomal skin complications, based on the area of the first sample abnormal color area and the sample feature vector of the peristomal skin. Determine the loss function according to the relative difference between the sample predicted peristomal skin complication probability data and the sample recorded sample stoma peristomal skin complication probability data. By performing feedback adjustment on the loss function, the trained peristomal skin complication probability prediction model is obtained.

[0044] According to an embodiment of the present invention, determining the loss function of the peristomal skin complication probability prediction model according to the sample nutritional index data, the area of the first sample abnormal color area, the sample feature vector of the peristomal skin, the sample stoma peristomal skin complication probability data, and the sample predicted peristomal skin complication probability data includes: determining the loss function Loss1 of the peristomal skin complication probability prediction model according to formula (2),

[0045]

[0046] wherein, R jFor the j-th sample stoma patient, the sample feature vector of the skin around the stoma in the sample stoma image, R a For the stoma patient, the feature vector of the skin around the stoma in the stoma image, R j T Is R j The transposed vector of, S 1,j For the first sample abnormal color area area of the j-th sample stoma patient, S 1,a For the first abnormal color area area of the stoma patient, e j,i For the i-th sample nutrition index data of the j-th sample stoma patient, e T For the standard nutrition index data, G j For the sample stoma surrounding skin complication probability data of the j-th sample stoma patient, G j,p For the sample predicted stoma surrounding skin complication probability data of the j-th sample stoma patient, N is the number of types of nutrition indexes, m x For the number of sample stoma patients in the x-th batch, U is the number of sample batches, i ≤ N, j ≤ m x , x ≤ U, and i, j, x, N, m x And U are all positive integers.

[0047] According to an embodiment of the present invention, in formula (2), Is the relative error between the sample stoma surrounding skin complication probability data of the j-th sample stoma patient and the sample predicted stoma surrounding skin complication probability data of the j-th sample stoma patient. Is the relative difference between the i-th sample nutrition index data of the j-th sample stoma patient and the standard nutrition index data. The larger the relative difference, the greater the probability of stoma surrounding skin complications. That is, the sample nutrition index data is positively correlated with the sample stoma surrounding skin complication probability data of the sample stoma patient. When the sample nutrition index data does not meet the standard, it may lead to poor healing of the skin around the stoma and increase the occurrence probability of complications such as mold infection and dermatitis. Therefore, the larger the sample nutrition index data relative to the standard nutrition index data, that is, The larger the value of, the greater the influence on the error of the sample predicted stoma surrounding skin complication probability data. Is the similarity between the sample feature vector of the skin around the stoma in the sample stoma image of the j-th sample stoma patient and the feature vector of the skin around the stoma in the stoma image of the stoma patient. Is the relative difference between the first abnormal color area area of the stoma patient and the first sample abnormal color area area of the j-th sample stoma patient. is the similarity between the area of the first abnormal color region of the stoma patient and the area of the first sample abnormal color region of the j-th sample stoma patient. To achieve a similar detection effect of the probability of peristomal skin complications, if the feature vector and the area of the first abnormal color region of the peristomal skin in the stoma image of the stoma patient are closer to the sample feature vector and the area of the first abnormal color region of the peristomal skin in the sample stoma image of the sample stoma patient, that is, and the larger the value, the more similar the peristomal skin condition in the stoma image of the stoma patient is to the peristomal skin condition in the sample stoma image of the sample stoma patient, and the greater its reference value. Therefore, its weight is higher. represents the sum of the relative errors of the probability of peristomal skin complications of multiple sample nutrient index data of each sample stoma patient in the x-th batch multiplied by the corresponding weights. is the weight of the x-th batch, which is used to reasonably weight the relative errors of different training batches in the loss function. For the sample stoma patients in the (x + 1)-th batch, the accuracy of the predicted probability data of peristomal skin complications in the (x + 1)-th batch output by the peristomal skin complication probability prediction model is usually higher than that of the predicted probability data of peristomal skin complications in the x-th batch. That is, the shorter the time interval between a certain batch and the first batch, the less accurate its prediction result. To improve the training efficiency, its weight is set higher. On the contrary, the more accurate the prediction result, the lower its weight. Thus, a higher weight can be assigned to the item with lower accuracy, thereby enhancing the training intensity and training efficiency. Multiply the above and two items, which can represent the loss function of the peristomal skin complication probability prediction model.

[0048] According to an embodiment of the present invention, during the process of training the peristomal skin complication probability prediction model, by performing backpropagation on the loss function, some parameters inside the model are adjusted to reduce the value of the loss function of the peristomal skin complication probability prediction model, thereby improving the accuracy of the peristomal skin complication probability prediction model and obtaining the trained peristomal skin complication probability prediction model.

[0049] In this way, the influence of the sample nutritional index data on the probability of peristomal skin complications can be used to determine the influence of the above data on the error of the sample prediction data of the probability of peristomal skin complications. Then, based on this influence and the relative error between the sample peristomal skin complication probability data and the sample prediction peristomal skin complication probability data, and based on the characteristics that the more similar the peristomal skin condition of the stoma patient is to that of the sample stoma patient, the greater the reference value, and the shorter the time interval from the first batch, the lower the accuracy, weights are set. Thus, the errors output by the peristomal skin complication probability prediction model for multiple sample nutritional index data of each sample stoma patient in the x-th batch are weighted and summed to obtain a loss function, so as to improve the design accuracy and objectivity of the loss function, thereby improving the training efficiency during the training process and enhancing the accuracy of the peristomal skin complication probability prediction model.

[0050] According to an embodiment of the present invention, in step S105, the stoma location can be located in multiple parts of the body, for example, the abdomen, perineum, etc. The stoma type can be divided into multiple types according to its different functions and morphologies, for example, ileostomy, colostomy, urostomy, etc.

[0051] According to an embodiment of the present invention, in step S106, the stoma location, stoma type, and stoma image are input into the trained stoma complication probability prediction model. This stoma complication probability prediction model can be a neural network model, which is trained based on a large amount of sample data through machine learning or deep learning techniques to obtain the predicted stoma complication probability data.

[0052] According to an embodiment of the present invention, the training steps of the stoma complication probability prediction model include: extracting multiple sample stoma patients to obtain sample stoma locations, sample stoma types, sample stoma images, and sample stoma complication probability data; performing feature extraction processing on the stoma through the stoma feature encoding layer of the trained image recognition neural network model to obtain the sample feature vector of the stoma in the sample stoma image; processing the sample stoma location, the sample stoma type, and the sample stoma image through the stoma complication probability prediction model to obtain sample predicted stoma complication probability data; determining the loss function of the stoma complication probability prediction model according to the sample stoma location, the sample stoma type, the sample feature vector of the stoma, the sample stoma complication probability data, and the sample predicted stoma complication probability data; training the stoma complication probability prediction model according to the loss function of the stoma complication probability prediction model to obtain the trained stoma complication probability prediction model.

[0053] According to an embodiment of the present invention, a plurality of extracted sample stoma patients are divided into different batches of sample stoma patients, and the number of sample stoma patients in each batch is the same, so as to train the stoma complication probability prediction model for different batches. For example, m1 + m2 + … + m U = M, where M is the number of sample stoma patients, and m1, m2, …, m U are the numbers of sample stoma patients in the 1st, 2nd, …, Uth batches respectively. For each sample stoma patient, the sample stoma complication probability data can be obtained from the recorded occurrence probability of stoma complications. For example, if the sample stoma patient has a stoma complication, the sample stoma complication probability data of this sample stoma patient is 1; if the sample stoma patient has no stoma complication, the sample stoma complication probability data of this sample stoma patient is 0. The image recognition neural network model can be a convolutional neural network model. Using the stoma feature encoding layer of the image recognition neural network model, feature extraction processing is performed on the stoma to obtain the sample feature vector of the stoma in the sample stoma image. The sample feature vector of the stoma usually includes stoma color, stoma shape, secretion properties, etc., and can be used to analyze and identify the health status of the stoma. The stoma complication probability prediction model can predict the sample predicted stoma complication probability data based on the sample stoma location, sample stoma type, and sample feature vector of the stoma. The loss function is determined according to the relative difference between the sample predicted stoma complication probability data and the sample recorded sample stoma complication probability data. By performing feedback adjustment on the loss function, the trained stoma complication probability prediction model is obtained.

[0054] According to an embodiment of the present invention, determining the loss function of the stoma complication probability prediction model according to the sample stoma location, the sample stoma type, the sample feature vector of the stoma, the sample stoma complication probability data, and the sample predicted stoma complication probability data includes: determining the loss function Loss2 of the stoma complication probability prediction model according to formula (3),

[0055]

[0056] where W j is the sample stoma location of the jth sample stoma patient, W a is the stoma location of the stoma patient, W j T is the transposed vector of W j K j is the sample stoma type of the jth sample stoma patient, K a is the stoma type of the stoma patient, K j T is the transposed vector of K j O jis the sample feature vector of the stoma in the sample stoma image of the j-th sample stoma patient, O a is the feature vector of the stoma in the stoma image of the stoma patient, O j T is O j 's transposed vector, B j is the sample stoma complication probability data of the j-th sample stoma patient, B j,p is the sample predicted stoma complication probability data of the j-th sample stoma patient, m x is the number of sample stoma patients in the x-th batch, U is the number of sample batches, j ≤ m x , x ≤ U, and j, x, m x and U are all positive integers.

[0057] According to an embodiment of the present invention, in formula (3), is the relative error between the sample stoma complication probability data of the j-th sample stoma patient and the sample predicted stoma complication probability data of the j-th sample stoma patient. is the similarity between the sample stoma position of the j-th sample stoma patient and the stoma position of the stoma patient. is the similarity between the sample stoma type of the j-th sample stoma patient and the stoma type of the stoma patient. is the similarity between the sample feature vector of the stoma in the sample stoma image of the j-th sample stoma patient and the feature vector of the stoma in the stoma image of the stoma patient. To achieve a similar stoma complication probability detection effect, if the stoma position, stoma type, and feature vector of the stoma of the stoma patient are closer to the sample stoma position, sample stoma type, and sample feature vector of the stoma of the sample stoma patient, that is, and the greater the value, the more similar the stoma condition of the stoma of the stoma patient is to the stoma condition of the sample stoma patient, and the greater its reference value, so its weight is higher. is the weight of the x-th batch, used to reasonably weight the relative errors of different training batches in the loss function. For the sample stoma patients in the x + 1-th batch, the accuracy of the sample predicted peristomal skin complication probability data of the x + 1-th batch output by the peristomal skin complication probability prediction model is usually higher than that of the sample predicted peristomal skin complication probability data of the x-th batch, that is, the shorter the time interval between a certain batch and the 1st batch, the less accurate its prediction result. To improve the training efficiency, its weight is set higher, otherwise, the prediction result is more accurate, and its weight is also lower, so that higher weights can be assigned to the terms with lower accuracy, thereby enhancing the training intensity and training efficiency.

[0058] According to an embodiment of the present invention, using and The relative errors of the stoma complication probabilities of each sample stoma patient in the x-th batch are weighted and summed to obtain a training loss function. During the process of training the stoma complication probability prediction model, by backpropagating the loss function, some parameters inside the model are adjusted to reduce the value of the loss function of the stoma complication probability prediction model, thereby improving the accuracy of the stoma complication probability prediction model and obtaining the trained stoma complication probability prediction model.

[0059] In this way, based on the relative error between the sample stoma complication probability data and the sample predicted stoma complication probability data, weights are set based on the characteristic that the more similar the stoma condition of the stoma patient is to that of the sample stoma patient, the greater the reference value, and weights are set based on the characteristic that the lower the accuracy is with the shorter time interval from the first batch. Thus, the errors output by the stoma complication probability prediction model for each sample stoma patient in the x-th batch are weighted and summed to obtain a loss function, so as to improve the design accuracy and objectivity of the loss function, thereby improving the training efficiency during the training process and enhancing the accuracy of the stoma complication probability prediction model.

[0060] According to an embodiment of the present invention, in step S107, when the predicted peristomal skin complication probability data is greater than the preset peristomal skin complication probability data and the predicted stoma complication probability data is less than or equal to the preset stoma complication probability data, the complication type is peristomal skin complication. When the predicted peristomal skin complication probability data is less than or equal to the preset peristomal skin complication probability data and the predicted stoma complication probability data is greater than the preset stoma complication probability data, the complication type is stoma complication. When the predicted peristomal skin complication probability data is greater than the preset peristomal skin complication probability data and the predicted stoma complication probability data is greater than the preset stoma complication probability data, the complication type is peristomal skin complication and stoma complication.

[0061] An artificial intelligence-based method for identifying postoperative stoma complications according to an embodiment of the present invention can determine whether a stoma patient has postoperative stoma complications through stoma images, improve the inspection efficiency, and reduce the probability of misidentification at the same time. Through information in multiple dimensions such as stoma images, nutritional index data, stoma location, and stoma type, the condition of the stoma patient can be more comprehensively understood, so as to predict the probabilities of peristomal skin complications and stoma complications. According to the predicted peristomal skin complication probability data and the predicted stoma complication probability data, the type of postoperative stoma complications is identified, improving the accuracy and comprehensiveness of the identification of the type of postoperative stoma complications. When determining the first stoma judgment parameter, the first stoma judgment parameter can be determined through the stoma height recognition result, stoma color recognition result, stoma shape recognition result, and peristomal skin recognition result. An artificial intelligence model is used to detect the height, color, shape, and the condition of the surrounding skin of the stoma, and whether the stoma patient has postoperative stoma complications is determined according to the detection results, which helps to more accurately reflect the overall health condition of the stoma, thereby improving the comprehensiveness and accuracy of the first judgment parameter. When determining the loss function of the peristomal skin complication probability prediction model, the influence of the sample nutritional index data on the probability of peristomal skin complications can be used to determine the influence of the above data on the error of the sample predicted peristomal skin complication probability data. Based on this influence and the relative error between the sample peristomal skin complication probability data and the sample predicted peristomal skin complication probability data, and based on the characteristic that the more similar the peristomal skin condition of the stoma patient is to that of the sample stoma patient, the greater the reference value, weights are set, and based on the characteristic that the shorter the time interval from the first batch, the lower the accuracy, weights are set. Thus, the errors output by the peristomal skin complication probability prediction model of multiple sample nutritional index data of each sample stoma patient in the xth batch are weighted and summed to obtain the loss function, so as to improve the design accuracy and objectivity of the loss function, thereby improving the training efficiency during the training process and the accuracy of the peristomal skin complication probability prediction model. When determining the loss function of the stoma complication probability prediction model, based on the relative error between the sample stoma complication probability data and the sample predicted stoma complication probability data, and based on the characteristic that the more similar the stoma condition of the stoma patient is to that of the sample stoma patient, the greater the reference value, weights are set, and based on the characteristic that the shorter the time interval from the first batch, the lower the accuracy, weights are set. Thus, the errors output by the stoma complication probability prediction model of each sample stoma patient in the xth batch are weighted and summed to obtain the loss function, so as to improve the design accuracy and objectivity of the loss function, thereby improving the training efficiency during the training process and the accuracy of the stoma complication probability prediction model.

[0062] Figure 2 Exemplarily shown is a block diagram of an artificial intelligence-based postoperative stoma complication identification system according to an embodiment of the present invention. The system includes:

[0063] A stoma image module for obtaining a stoma image through an image acquisition device;

[0064] A judgment module for determining whether a stoma patient has postoperative stoma complications according to the stoma image;

[0065] A nutritional index data module for obtaining the nutritional index data of a stoma patient by extracting the blood of the stoma patient if the stoma patient has postoperative stoma complications;

[0066] A module for predicting the probability data of peristomal skin complications for inputting the nutritional index data and the stoma image into a trained peristomal skin complication probability prediction model to obtain the probability data of peristomal skin complications;

[0067] A stoma position and stoma type module for obtaining the stoma position and stoma type;

[0068] A module for predicting the probability data of stoma complications for inputting the stoma position, the stoma type and the stoma image into a trained stoma complication probability prediction model to obtain the probability data of stoma complications;

[0069] A complication type module for determining the complication type according to the probability data of peristomal skin complications and the probability data of stoma complications;

[0070] 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 having thereon computer-readable program instructions for performing various aspects of the present invention.

[0071] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the drawings are only examples and do not limit the present invention. The object of the present invention has been fully and effectively achieved. The functions and structural principles of the present invention have been shown and described in the embodiments, and the embodiments of the present invention may have any deformation or modification without departing from the principle.

Claims

1. A method for identifying postoperative stoma complications based on artificial intelligence, characterized in that: include: Obtain stoma images through image acquisition equipment; determining whether the stoma patient suffers from postoperative stoma complications according to the stoma image; If the stoma patient suffers from postoperative stoma complications, the nutritional index data of the stoma patient will be obtained by drawing blood from the stoma patient; Inputting the nutrition index data and the stoma image into a trained peristomal skin complication probability prediction model to obtain predicted peristomal skin complication probability data; Obtain stoma location and stoma type; Inputting the stoma position, the stoma type and the stoma image into a trained stoma complication probability prediction model to obtain predicted stoma complication probability data; According to the predicted peristomal skin complication probability data and the predicted stoma complication probability data, the type of complication is determined, specifically including: when the predicted peristomal skin complication probability data is greater than the preset peristomal skin complication probability data, and the predicted stoma complication probability data is less than or equal to the preset stoma complication probability data, the complication type is peristomal skin complication; when the predicted peristomal skin complication probability data is less than or equal to the preset peristomal skin complication probability data, and the predicted stoma complication probability data is greater than the preset stoma complication probability data, the complication type is stoma complication; when the predicted peristomal skin complication probability data is greater than the preset peristomal skin complication probability data, and the predicted stoma complication probability data is greater than the preset stoma complication probability data, the complication types are peristomal skin complication and stoma complication.

2. The method for identifying postoperative stoma complications based on artificial intelligence according to claim 1, characterized in that: According to the stoma image, determining whether the postoperative stoma patient suffers from postoperative stoma complications, including: In the stoma image, identifying whether the stoma height is within a preset range through an image detection model, and determining a stoma height recognition result; In the stoma image, identifying whether the stoma color is beef red or pink by an image detection model, and determining a stoma color recognition result; In the stoma image, identifying whether the stoma shape is circular or elliptical through an image detection model, and determining a stoma shape recognition result; In the stoma image, identifying whether the peristomal skin is intact or flat by using an image detection model, and determining a peristomal skin recognition result; Determining a first stoma assessment parameter according to the stoma height recognition result, the stoma color recognition result, the stoma shape recognition result, and the stoma peristomal skin recognition result; If the stoma first evaluation parameter is less than a set threshold, it is determined that the postoperative stoma patient suffers from postoperative stoma complications; If the first stoma assessment parameter is greater than or equal to a set threshold, it is determined that the postoperative stoma patient has no postoperative stoma complications.

3. The method for identifying postoperative stoma complications based on artificial intelligence according to claim 2, characterized in that: Determining a first stoma assessment parameter according to the stoma height recognition result, the stoma color recognition result, the stoma shape recognition result, and the stoma peristomal skin recognition result includes: According to the formula A1=α1H o +α2C o +α3F o +α4D o Determine the first stoma judgment parameter A1, where H o is the stoma height recognition result, C o is the stoma color recognition result, F o is the stoma shape recognition result, D o This is the peristomal skin identification result. α1, α2, α3 and α4 are preset weights.

4. The method for identifying postoperative stoma complications based on artificial intelligence according to claim 1, characterized in that: The training steps of the peristomal skin complication probability prediction model include: By sampling multiple sample stoma patients, sample nutritional index data, sample stoma images and sample peristomal skin complication probability data are obtained; Determine, among the plurality of pixels of the sample stoma image, a plurality of abnormal color pixels whose corresponding chromaticity is higher than a first preset chromaticity threshold; Performing a connected domain analysis on the multiple abnormal color pixels to obtain a first sample abnormal color region and determine the area of ​​the first sample abnormal color region; Through the skin feature encoding layer of the trained image recognition neural network model, feature extraction processing is performed on the peristomal skin area to obtain a sample feature vector of the peristomal skin in the sample stoma image; The sample nutrition index data and the sample stoma image are processed by a peristomal skin complication probability prediction model to obtain sample predicted peristomal skin complication probability data; Determine the loss function of the peristomal skin complication probability prediction model according to the sample nutrition index data, the area of ​​the first sample abnormal color region, the sample feature vector of the peristomal skin, the sample peristomal skin complication probability data and the sample predicted peristomal skin complication probability data; According to the loss function of the peristomal skin complications probability prediction model, the peristomal skin complications probability prediction model is trained to obtain the trained peristomal skin complications probability prediction model.

5. The artificial intelligence-based postoperative stoma complication identification method according to claim 4, characterized in that: Determining a loss function of the peristomal skin complication probability prediction model according to the sample nutritional index data, the area of ​​the first sample abnormal color region, the sample feature vector of the peristomal skin, the sample peristomal skin complication probability data, and the sample predicted peristomal skin complication probability data includes: According to the formula Determine the loss function Loss1 of the peristomal skin complication probability prediction model, where R j is the sample feature vector of the skin around the stoma in the sample stoma image of the jth sample stoma patient, R a is the feature vector of the skin around the stoma in the stoma image of the stoma patient, R j T For R j The transposed vector, S 1,j is the area of ​​the abnormal color region of the first sample of the jth sample stoma patient, S 1,a is the area of ​​the first abnormal color region of the stoma patient, e j,i is the nutritional index data of the i-th sample of the j-th sample stoma patient, e T is the standard nutritional index data, G j is the probability data of peristomal skin complications of the jth sample stoma patient, G j,p Predict the probability data of peristomal skin complications for the jth sample of stoma patients, N is the number of nutritional indicator types, m is x is the number of stoma patients in the xth batch of samples, U is the number of sample batches, i≤N, j≤m x , x≤U, and i, j, x, N, m x and U are both positive integers.

6. The method for identifying postoperative stoma complications based on artificial intelligence according to claim 1, characterized in that: The training steps of the stoma complication probability prediction model include: By sampling a plurality of sample stoma patients, sample stoma locations, sample stoma types, sample stoma images and sample stoma complication probability data are obtained; Through the stoma feature encoding layer of the trained image recognition neural network model, feature extraction processing is performed on the stoma to obtain a sample feature vector of the stoma in the sample stoma image; The sample stoma position, the sample stoma type and the sample stoma image are processed by a stoma complication probability prediction model to obtain sample predicted stoma complication probability data; Determining a loss function of the stoma complication probability prediction model according to the sample stoma location, the sample stoma type, the sample feature vector of the stoma, the sample stoma complication probability data, and the sample predicted stoma complication probability data; The stoma complication probability prediction model is trained according to the loss function of the stoma complication probability prediction model to obtain the trained stoma complication probability prediction model.

7. The method for identifying postoperative stoma complications based on artificial intelligence according to claim 6, characterized in that: Determining a loss function of the stoma complication probability prediction model according to the sample stoma position, the sample stoma type, the sample feature vector of the stoma, the sample stoma complication probability data, and the sample predicted stoma complication probability data includes: According to the formula Determine the loss function Loss2 of the stoma complication probability prediction model, where W j is the sample stoma position of the jth sample stoma patient, W a is the stoma position of stoma patients, W j T W j The transposed vector, K j is the sample stoma type of the jth sample stoma patient, K a is the stoma type of the stoma patient, K j T K j The transposed vector of j is the sample feature vector of the stoma in the sample stoma image of the jth sample stoma patient, O a is the feature vector of the stoma in the stoma image of the stoma patient, O j T O j The transposed vector, B j is the sample stoma complication probability data of the jth sample stoma patient, B j,p Predict the probability of stoma complications for the jth sample of stoma patients, m x is the number of stoma patients in the xth batch of samples, U is the number of sample batches, j≤m x , x≤U, and j, x, m x and U are both positive integers.

8. An artificial intelligence-based postoperative stoma complication identification system for executing the artificial intelligence-based postoperative stoma complication identification method according to any one of claims 1 to 7, characterized in that: include: A stoma image module, used for acquiring a stoma image through an image acquisition device; A judgment module, used for determining whether the stoma patient suffers from postoperative stoma complications according to the stoma image; A nutrition index data module is used to obtain nutrition index data of a stoma patient by drawing blood from the stoma patient if the stoma patient suffers from postoperative stoma complications; A module for predicting the probability data of peristomal skin complications, used for inputting the nutrition index data and the stoma image into a trained peristomal skin complication probability prediction model to obtain predicted peristomal skin complication probability data; A stoma location and stoma type module, used to obtain the stoma location and stoma type; A stoma complication probability prediction data module is used to input the stoma position, the stoma type and the stoma image into a trained stoma complication probability prediction model to obtain stoma complication probability prediction data; The complication type module is used to determine the complication type based on the predicted peristomal skin complication probability data and the predicted stoma complication probability data.

Images