Image processing method and device for judging hemolysis of suspended red blood cells

Abstract

The application discloses an image processing method and device for judging hemolysis of suspended red blood cells, and the method comprises the following steps: acquiring a blood image set; the blood image set comprises a plurality of blood images and corresponding label information; acquiring a blood parameter information set; the blood parameter information set comprises a plurality of blood parameters and corresponding label information; the blood parameters comprise hemoglobin concentration, reticulocyte concentration, indirect bilirubin concentration, lactate dehydrogenase concentration and globin concentration; a blood hemolysis composite detection model is constructed by using the blood image set and the blood parameter information set; the collected blood image and blood parameter are processed based on the blood hemolysis composite detection model, and hemolysis detection result information is obtained; the hemolysis detection result information is used for representing the red blood cell hemolysis information of the corresponding blood of the collected blood image and blood parameter.

Description

Technical Field

[0001] This invention relates to the fields of blood quality testing and image processing, specifically to an image processing method and apparatus for determining hemolysis of suspended red blood cells. Background Technology

[0002] During blood use, such as when a hospital's transfusion medicine department sends stored blood to clinical settings, the blood quality must be checked before use. A crucial indicator of blood quality is the presence of hemolysis of suspended red blood cells. Normal suspended red blood cells in a static state, compared to those in a state of hemolysis, will exhibit completely different color characteristics. Currently, pre-use blood quality checks often rely on visually observing the color of the plasma layer in static suspended red blood cells to determine hemolysis. However, this method is often influenced by the operator's subjectivity, resulting in inaccurate and subjective assessments of hemolysis. Summary of the Invention

[0003] This invention addresses the problem that current detection methods for hemolysis of suspended red blood cells are often affected by the subjective factors of the operator, resulting in unobjective and inaccurate test results. This invention discloses an image processing method and apparatus for determining hemolysis of suspended red blood cells.

[0004] In a first aspect, this application discloses an image processing method for determining hemolysis of suspended red blood cells, comprising:

[0005] S1, acquire a blood image set; the blood image set includes several blood images and corresponding label information;

[0006] S2, Obtain a set of blood parameter information; the set of blood parameter information includes several blood parameters and corresponding label information; the blood parameters include hemoglobin concentration, reticulocyte concentration, indirect bilirubin concentration, lactate dehydrogenase concentration, and globin concentration;

[0007] S3, using the blood image set and blood parameter information set, a blood hemolysis complex detection model is constructed;

[0008] S4. Based on the blood hemolysis composite detection model, the acquired blood images and blood parameters are processed to obtain hemolysis detection result information; the hemolysis detection result information is used to characterize the degree of erythrocyte hemolysis in the blood corresponding to the acquired blood images and blood parameters.

[0009] The blood hemolysis composite detection model is used to process the acquired blood images and blood parameters to obtain hemolysis detection results, including:

[0010] The acquired blood images are preprocessed to obtain preprocessed images;

[0011] Using the blood hemolysis composite detection model, the preprocessed image and blood parameters are processed to obtain hemolysis detection results.

[0012] The blood hemolysis composite detection model includes: a hemolysis image detection sub-model, a parameter detection sub-model, and a fusion sub-model;

[0013] The input terminal of the hemolysis image detection sub-model is used to receive the obtained blood image;

[0014] The input terminal of the parameter detection sub-model is used to receive blood parameters.

[0015] The fusion sub-model is connected to the output terminals of the hemolysis image detection sub-model and the parameter detection sub-model, respectively, and is used to fuse the output information of the two sub-models to obtain hemolysis detection result information.

[0016] The method of constructing a blood hemolysis complex detection model using the blood image set and blood parameter information set includes:

[0017] Initialize the blood hemolysis complex detection model;

[0018] The blood image set is used to train the hemolysis image detection sub-model to obtain the trained hemolysis image detection sub-model.

[0019] Using the blood parameter information set, the parameter detection sub-model is trained to obtain the trained parameter detection sub-model;

[0020] Using the trained hemolysis image detection sub-model, the trained parameter detection sub-model, and the fusion sub-model, a blood hemolysis composite detection model is constructed.

[0021] The hemolysis image detection sub-model includes:

[0022] Color features are calculated on the input blood image to obtain a set of color feature moments; the set of color feature moments includes the first-order color moments, second-order color moments, and third-order color moments of the three color channels: R channel, G channel, and B channel;

[0023] Texture features are calculated on the input blood image to obtain contrast and entropy values;

[0024] The shape and structure features of the input blood image are calculated to obtain the pixel values ​​of corner points, texture feature points, and SIFT feature points.

[0025] Using a preset color moment fusion model, the color feature moment set is calculated to obtain color feature values; the expression of the preset color moment fusion model is:

[0026] ,

[0027] ,

[0028] in, This is the color composite value for one color channel. Let i be the i-th order color moment of a color channel. For a color channel, the preset i-th order standard color moment. and These are all model parameters to be solved during the training process. For color feature values, This represents the color composite value of the j-th color channel. When j=1, 2, and 3, it represents the R channel, G channel, and B channel, respectively.

[0029] A feature vector is constructed using the color feature value, contrast value, entropy value, corner pixel value, texture feature point pixel value, and SIFT feature point pixel value. ;

[0030] The first hemolysis detection calculation is performed on the feature vector to obtain the first hemolysis detection result;

[0031] The expression for calculating the first hemolysis detection is:

[0032]

[0033] in, This is the first result of the hemolysis test. and All are model vectors. and These are all model parameters to be solved during the training process.

[0034] The parameter detection sub-model includes:

[0035] Blood parameters collected at several time points are merged to obtain a set of blood parameter sequences; the set of blood parameter sequences includes hemoglobin concentration sequence, reticulocyte concentration sequence, indirect bilirubin concentration sequence, lactate dehydrogenase concentration sequence, and globin concentration sequence; the elements in the sequence are blood parameters collected at each time point;

[0036] Obtain the standard value for each blood parameter;

[0037] By subtracting the standard value of the corresponding blood parameter from each sequence in the blood parameter sequence set, the corresponding difference sequence is obtained;

[0038] By using all the difference sequences as row vectors, a difference matrix is ​​constructed.

[0039] Extract the maximum and minimum values ​​of all row vectors in the difference matrix;

[0040] The difference matrix is ​​used to perform a second hemolysis detection calculation to obtain the second hemolysis detection result;

[0041] The expression for calculating the second hemolysis detection is:

[0042] ,

[0043] In the formula, For the second hemolysis detection result, N and M are the row and column dimensions of the difference matrix, respectively. and Let represent the maximum and minimum values ​​of the i-th row vector of the difference matrix, respectively. The element in the i-th row and j-th column of the difference matrix is... and , j=1,2,…,M, are all model parameters to be solved during the training process.

[0044] The hemolysis image detection sub-model includes: a feature detection network and a fusion detection network;

[0045] The feature detection network is used to process the blood image to obtain image feature information; the fusion detection network is used to perform hemolysis prediction processing on the image feature information to obtain a second hemolysis detection result.

[0046] The feature detection network includes a first input module, a first convolution module, a depthwise separable convolution module, a first up-dimensional convolution module, a second up-dimensional convolution module, a third up-dimensional convolution module, a fourth up-dimensional convolution module, a second convolution module, a first pooling module, a third convolution module, and a first fully connected module.

[0047] The input terminal of the first input module of the feature detection network is used to receive the blood image; the output terminal of the first input module of the feature detection network is connected to the input terminal of the first convolution module of the feature detection network; the output terminal of the first convolution module of the feature detection network is connected to the input terminal of the depth-separable convolution module of the feature detection network; the output terminal of the depth-separable convolution module of the feature detection network is connected to the input terminal of the first up-dimensional convolution module of the feature detection network; the output terminal of the first up-dimensional convolution module of the feature detection network is connected to the input terminal of the second up-dimensional convolution module of the feature detection network; the output terminal of the second up-dimensional convolution module of the feature detection network is connected to the input terminal of the feature detection image. The input of the third up-dimensional convolutional module of the feature detection network is connected to the input of the fourth up-dimensional convolutional module of the feature detection network; the output of the fourth up-dimensional convolutional module of the feature detection network is connected to the input of the second convolutional module of the feature detection network; the output of the second convolutional module of the feature detection network is connected to the input of the first pooling module of the feature detection network; the output of the first pooling module of the feature detection network is connected to the input of the third convolutional module of the feature detection network; and the output of the third convolutional module of the feature detection network is connected to the input of the first fully connected module of the feature detection network.

[0048] The fusion detection network includes a second input module, a fourth convolution module, a fifth convolution module, a sixth convolution module, a second pooling module, a seventh convolution module, and a second fully connected module.

[0049] The input terminal of the second input module of the fusion detection network is connected to the output terminal of the first fully connected module of the feature detection network; the output terminal of the second input module of the fusion detection network is connected to the input terminal of the fourth convolution module of the fusion detection network; the output terminal of the fourth convolution module of the fusion detection network is connected to the input terminal of the fifth convolution module of the fusion detection network; the output terminal of the fifth convolution module of the fusion detection network is connected to the input terminal of the sixth convolution module of the fusion detection network; the output terminal of the sixth convolution module of the fusion detection network is connected to the input terminal of the second pooling module of the fusion detection network; the output terminal of the second pooling module of the fusion detection network is connected to the input terminal of the seventh convolution module of the fusion detection network; the output terminal of the seventh convolution module of the fusion detection network is connected to the input terminal of the second fully connected module of the fusion detection network; the output terminal of the second fully connected module of the fusion detection network is used to output the second hemolysis detection result of the predicted blood image.

[0050] A second aspect of the present invention discloses an image processing device for determining hemolysis of suspended red blood cells, the device comprising:

[0051] Memory containing executable program code;

[0052] A processor coupled to the memory;

[0053] The processor calls the executable program code stored in the memory to execute the image processing method for determining hemolysis of suspended red blood cells.

[0054] In a third aspect of this invention, a computer-storable medium is disclosed, the computer-storable medium storing computer instructions, which, when invoked by a computer, are used to execute the image processing method for determining hemolysis of suspended red blood cells.

[0055] In a fourth aspect of this invention, an information data processing terminal is disclosed, which is used to implement the image processing method for determining hemolysis of suspended red blood cells.

[0056] The beneficial effects of this invention are as follows:

[0057] This invention constructs a blood hemolysis composite detection model by combining blood images and multiple key blood parameters, breaking the limitations of relying solely on image features or blood parameters for hemolysis judgment. It achieves the synergistic fusion of multi-dimensional information, effectively improving the comprehensiveness and accuracy of hemolysis detection results. It can more accurately characterize the degree of erythrocyte hemolysis in the blood, providing a reliable basis for clinical hemolysis detection.

[0058] In the hemolysis image detection sub-model, this invention comprehensively extracts the color features, texture features, and shape and structure features of blood images, and optimizes and fuses the multi-channel color feature moments through a preset color moment fusion model. At the same time, combined with the targeted calculation of feature vectors, it fully explores the subtle feature information related to hemolysis in blood images, greatly improves the ability of image features to represent the hemolysis state, and makes image-based hemolysis detection more sensitive and reliable.

[0059] The parameter detection sub-model of this invention serializes multiple blood parameters collected at multiple time points, constructs a difference matrix, and extracts the maximum and minimum values ​​of the row vectors. Combined with a specially designed second hemolysis detection calculation method, it considers both the deviation of blood parameters from standard values ​​and the dynamic characteristics of parameter changes over time. It can accurately capture the changing patterns of blood parameters during hemolysis, providing solid parameter support for judging the degree of hemolysis.

[0060] The detection method of this invention has a clear process and is easy to operate. The blood images and blood parameters to be collected are all routine clinical test indicators. There is no need to add complicated test procedures or special equipment. It is easy to promote and apply in clinical practice and can be widely used in hemolysis detection scenarios in various medical institutions. It has high clinical practical value and promotion prospects. Attached Figure Description

[0061] Figure 1 This is a flowchart illustrating the implementation of the method of the present invention. Detailed Implementation

[0062] To better understand the content of this invention, an embodiment is provided here.

[0063] Figure 1 This is a flowchart illustrating the implementation of the method of the present invention.

[0064] To address the problem that current detection methods for hemolysis of suspended red blood cells are often affected by the subjective factors of the operator, resulting in unobjective and inaccurate test results, this invention discloses an image processing method and apparatus for determining hemolysis of suspended red blood cells.

[0065] In a first aspect, this application discloses an image processing method for determining hemolysis of suspended red blood cells, comprising:

[0066] S1, acquire a blood image set; the blood image set includes several blood images and corresponding label information;

[0067] S2, Obtain a set of blood parameter information; the set of blood parameter information includes several blood parameters and corresponding label information; the blood parameters include hemoglobin concentration, reticulocyte concentration, indirect bilirubin concentration, lactate dehydrogenase concentration, and globin concentration;

[0068] S3, using the blood image set and blood parameter information set, a blood hemolysis complex detection model is constructed;

[0069] S4. Based on the blood hemolysis composite detection model, the acquired blood images and blood parameters are processed to obtain hemolysis detection result information; the hemolysis detection result information is used to characterize the degree of erythrocyte hemolysis in the blood corresponding to the acquired blood images and blood parameters.

[0070] The label information is used to characterize the blood parameters corresponding to the occurrence of erythrocyte hemolysis. The larger the value, the more severe the erythrocyte hemolysis.

[0071] The hemolysis detection results are used to characterize the degree of hemolysis of red blood cells; the higher the value, the more obvious the hemolysis.

[0072] The blood hemolysis composite detection model is used to process the acquired blood images and blood parameters to obtain hemolysis detection results, including:

[0073] The acquired blood images are preprocessed to obtain preprocessed images;

[0074] Using the blood hemolysis composite detection model, the preprocessed image and blood parameters are processed to obtain hemolysis detection results.

[0075] The preprocessing of the acquired blood images to obtain preprocessed images includes:

[0076] The acquired blood images are normalized to obtain normalized images;

[0077] The normalized image is subjected to boundary limiting processing to obtain a preprocessed image.

[0078] The blood hemolysis composite detection model includes: a hemolysis image detection sub-model, a parameter detection sub-model, and a fusion sub-model;

[0079] The input terminal of the hemolysis image detection sub-model is used to receive the obtained blood image;

[0080] The input terminal of the parameter detection sub-model is used to receive blood parameters.

[0081] The fusion sub-model is connected to the output terminals of the hemolysis image detection sub-model and the parameter detection sub-model, respectively, and is used to fuse the output information of the two sub-models to obtain hemolysis detection result information.

[0082] The fusion sub-model performs fusion calculations on the output information of the two sub-models. This can be achieved by using preset weight values ​​to perform a weighted sum of the output information of the two sub-models. Alternatively, the weight values ​​can be the reciprocal of the mean square error between the model output and the label information during the training process of the two sub-models.

[0083] The method of constructing a blood hemolysis complex detection model using the blood image set and blood parameter information set includes:

[0084] Initialize the blood hemolysis complex detection model;

[0085] The blood image set is used to train the hemolysis image detection sub-model to obtain the trained hemolysis image detection sub-model.

[0086] Using the blood parameter information set, the parameter detection sub-model is trained to obtain the trained parameter detection sub-model;

[0087] Using the trained hemolysis image detection sub-model, the trained parameter detection sub-model, and the fusion sub-model, a blood hemolysis composite detection model is constructed.

[0088] The hemolysis image detection sub-model includes:

[0089] Color features are calculated on the input blood image to obtain a set of color feature moments; the set of color feature moments includes the first-order color moments, second-order color moments, and third-order color moments of the three color channels: R channel, G channel, and B channel;

[0090] Texture features are calculated on the input blood image to obtain contrast and entropy values;

[0091] The shape and structure features of the input blood image are calculated to obtain the pixel values ​​of corner points, texture feature points, and SIFT feature points.

[0092] Using a preset color moment fusion model, the color feature moment set is calculated to obtain color feature values; the expression of the preset color moment fusion model is:

[0093] ,

[0094] ,

[0095] in, This is the color composite value for one color channel. Let i be the i-th order color moment of a color channel. For a color channel, the preset i-th order standard color moment. and These are all model parameters to be solved during the training process. For color feature values, This represents the color composite value of the j-th color channel. When j=1, 2, and 3, it represents the R channel, G channel, and B channel, respectively.

[0096] A feature vector is constructed using the color feature value, contrast value, entropy value, corner pixel value, texture feature point pixel value, and SIFT feature point pixel value. ;

[0097] The first hemolysis detection calculation is performed on the feature vector to obtain the first hemolysis detection result;

[0098] The expression for calculating the first hemolysis detection is:

[0099]

[0100] in, This is the first result of the hemolysis test. and All are model vectors. and These are all model parameters to be solved during the training process.

[0101] The expression used for color feature fusion in this invention addresses the differences between the characteristic moments of different color channels and the standard characteristic moments by specifically processing them. At the same time, it combines logarithmic operations to achieve comprehensive quantification of multi-channel color information. This effectively integrates color features of different dimensions, highlights the differences in color changes related to hemolysis in blood images, avoids the limitations of single color features or channel information, and allows the fused color feature values ​​to more accurately reflect the color characteristics of blood images under hemolysis conditions, providing more recognizable image feature support for subsequent hemolysis detection.

[0102] The expression used in this invention for calculating the first hemolysis detection result can transform a high-dimensional feature vector into an intuitive quantitative result of hemolysis detection by performing targeted operations on a vector that integrates multiple features such as color, texture, and shape structure. This fully utilizes the correlation between various features and the hemolysis state, achieves efficient mapping of feature information to detection results, reduces interference from irrelevant features, improves the accuracy of hemolysis detection based on image features, and makes the detection result more directly characterize the degree of hemolysis.

[0103] The parameter detection sub-model includes:

[0104] Blood parameters collected at several time points are merged to obtain a set of blood parameter sequences; the set of blood parameter sequences includes hemoglobin concentration sequence, reticulocyte concentration sequence, indirect bilirubin concentration sequence, lactate dehydrogenase concentration sequence, and globin concentration sequence; the elements in the sequence are blood parameters collected at each time point;

[0105] Obtain the standard value for each blood parameter;

[0106] By subtracting the standard value of the corresponding blood parameter from each sequence in the blood parameter sequence set, the corresponding difference sequence is obtained;

[0107] By using all the difference sequences as row vectors, a difference matrix is ​​constructed.

[0108] Extract the maximum and minimum values ​​of all row vectors in the difference matrix;

[0109] The difference matrix is ​​used to perform a second hemolysis detection calculation to obtain the second hemolysis detection result;

[0110] The expression for calculating the second hemolysis detection is:

[0111] ,

[0112] In the formula, For the second hemolysis detection result, N and M are the row and column dimensions of the difference matrix, respectively. and Let represent the maximum and minimum values ​​of the i-th row vector of the difference matrix, respectively. The element in the i-th row and j-th column of the difference matrix is... and , j=1,2,…,M, are all model parameters to be solved during the training process.

[0113] The expression used in this invention for calculating the second hemolysis detection result, by performing hierarchical operations on the elements in the difference matrix, fully considers the deviation of each blood parameter from the standard value, and also takes into account the parameter fluctuation range reflected by the maximum and minimum values ​​in each parameter sequence. It can comprehensively capture the overall deviation and dynamic change characteristics of blood parameters during hemolysis, avoid the one-sidedness brought by a single parameter or static value, and make the hemolysis detection results based on blood parameters more scientific and reliable, accurately reflecting the change law of hemolysis-related parameters.

[0114] The hemolysis image detection sub-model includes: a feature detection network and a fusion detection network;

[0115] The feature detection network is used to process the blood image to obtain image feature information; the fusion detection network is used to perform hemolysis prediction processing on the image feature information to obtain a second hemolysis detection result.

[0116] The feature detection network includes a first input module, a first convolution module, a depthwise separable convolution module, a first up-dimensional convolution module, a second up-dimensional convolution module, a third up-dimensional convolution module, a fourth up-dimensional convolution module, a second convolution module, a first pooling module, a third convolution module, and a first fully connected module.

[0117] The input terminal of the first input module of the feature detection network is used to receive the blood image; the output terminal of the first input module of the feature detection network is connected to the input terminal of the first convolution module of the feature detection network; the output terminal of the first convolution module of the feature detection network is connected to the input terminal of the depth-separable convolution module of the feature detection network; the output terminal of the depth-separable convolution module of the feature detection network is connected to the input terminal of the first up-dimensional convolution module of the feature detection network; the output terminal of the first up-dimensional convolution module of the feature detection network is connected to the input terminal of the second up-dimensional convolution module of the feature detection network; the output terminal of the second up-dimensional convolution module of the feature detection network is connected to the input terminal of the feature detection image. The input of the third up-dimensional convolutional module of the feature detection network is connected to the input of the fourth up-dimensional convolutional module of the feature detection network; the output of the fourth up-dimensional convolutional module of the feature detection network is connected to the input of the second convolutional module of the feature detection network; the output of the second convolutional module of the feature detection network is connected to the input of the first pooling module of the feature detection network; the output of the first pooling module of the feature detection network is connected to the input of the third convolutional module of the feature detection network; and the output of the third convolutional module of the feature detection network is connected to the input of the first fully connected module of the feature detection network.

[0118] The fusion detection network includes a second input module, a fourth convolution module, a fifth convolution module, a sixth convolution module, a second pooling module, a seventh convolution module, and a second fully connected module;

[0119] The input terminal of the second input module of the fusion detection network is connected to the output terminal of the first fully connected module of the feature detection network; the output terminal of the second input module of the fusion detection network is connected to the input terminal of the fourth convolution module of the fusion detection network; the output terminal of the fourth convolution module of the fusion detection network is connected to the input terminal of the fifth convolution module of the fusion detection network; the output terminal of the fifth convolution module of the fusion detection network is connected to the input terminal of the sixth convolution module of the fusion detection network; the output terminal of the sixth convolution module of the fusion detection network is connected to the input terminal of the second pooling module of the fusion detection network; the output terminal of the second pooling module of the fusion detection network is connected to the input terminal of the seventh convolution module of the fusion detection network; the output terminal of the seventh convolution module of the fusion detection network is connected to the input terminal of the second fully connected module of the fusion detection network; the output terminal of the second fully connected module of the fusion detection network is used to output the second hemolysis detection result of the predicted blood image.

[0120] The blood hemolysis composite detection model of the present invention achieves deep fusion of image feature information and blood parameter information by organically combining the hemolysis image detection sub-model, parameter detection sub-model and fusion sub-model. It leverages the complementary advantages of the two detection methods, effectively avoids the errors that may exist in a single detection method, significantly reduces the probability of false positives and false negatives of hemolysis, and further improves the robustness and practicality of the detection model.

[0121] This invention employs various convolutional neural network structures, such as depthwise separable convolution, multi-stage upscaling convolution, and pooling, in the feature detection network. This enables efficient extraction of deep features from blood images. Simultaneously, through targeted processing of the fusion detection network, it achieves accurate mapping from image features to hemolysis detection results. While ensuring detection accuracy, it improves the computational efficiency of the model, enabling hemolysis detection to be completed quickly and meeting the timeliness requirements of clinical scenarios.

[0122] The convolution module can be implemented using 3D multichannel convolution;

[0123] The depthwise separable convolution module can be implemented by connecting channel splitting submodules and single-channel convolution submodules. Specifically, it can be implemented using the depthwise separable convolution module in the MobileNet network.

[0124] The upscaling convolution module can be implemented using the upscaling convolution module in the ResNet network;

[0125] The pooling module can be implemented using a maximized pooling operation.

[0126] Specifically, when the convolution module, depthwise separable convolution module, dimension-scaling convolution module, and pooling module are implemented using multiple three-dimensional filters connected in parallel, the data characteristics of the corresponding three-dimensional filter matrices are different. The data characteristics of the matrices include the mean, variance, and eigenvalues ​​of the matrices.

[0127] The output of the first fully connected module of the feature detection network is used to output the image feature information.

[0128] The output of the second fully connected module of the fusion detection network is used to output the hemolysis detection result information of the blood image.

[0129] The first convolutional module, depthwise separable convolutional module, first upscaling convolutional module, second upscaling convolutional module, third upscaling convolutional module, fourth upscaling convolutional module, second convolutional module, first pooling module, and third convolutional module of the feature detection network are all implemented using N parallel three-dimensional filters; the three-dimensional filters in adjacent modules of the feature detection network are connected according to their sequence numbers;

[0130] The fourth, fifth, sixth, second, and seventh convolutional modules of the fusion detection network are all implemented using N parallel three-dimensional filters; the three-dimensional filters in adjacent modules of the fusion detection network are connected according to their sequence numbers.

[0131] The 3D filters in adjacent modules of the feature detection network or fusion detection network are connected according to their sequence numbers. Specifically, the first 3D filter of the preceding convolutional module is connected to the first 3D filter of the following convolutional module, the second 3D filter of the preceding convolutional module is connected to the second 3D filter of the following convolutional module, the third 3D filter of the preceding convolutional module is connected to the third 3D filter of the following convolutional module, and so on. The last 3D filter of the preceding convolutional module is connected to the last 3D filter of the following convolutional module. Taking the connection relationship between the first and second up-dimensional convolutional modules in the feature detection network as an example, it is assumed that both the first and second up-dimensional convolutional modules include three 3D filters.

[0132] The specific structure of the hemolysis image detection sub-model of this invention revolves around the core objective of accurately extracting features related to erythrocyte hemolysis from blood images and efficiently predicting the degree of hemolysis. Through the hierarchical design of the feature detection network and the fusion detection network, the orderly connection of each functional module, and the targeted structural selection, it forms multi-dimensional technical advantages, as detailed below:

[0133] The feature detection network constructs an efficient feature extraction channel from basic information input of blood images to deep feature output. The initial convolutional module quickly filters redundant noise in the image, preserving basic visual features related to hemolysis, laying the foundation for subsequent deep feature extraction. The introduction of depthwise separable convolution significantly reduces the computational load and parameter size of the model while ensuring feature extraction effectiveness, avoiding the inefficiency caused by high image feature dimensionality. It also improves the model's ability to capture subtle features in blood images, which is particularly important for subtle feature changes in early-stage hemolysis.

[0134] The progressive design of the multi-stage dimensionality-increasing convolution module can gradually increase the dimensionality of the feature map, enabling layer-by-layer feature mining of blood images from shallow to deep layers. During hemolysis, the color, texture, shape, and structure of blood images undergo complex and subtle changes. Single-dimensional feature extraction is insufficient to fully cover these changes, while multi-stage dimensionality-increasing convolution can capture feature information in different dimensions, preserving both local details and integrating global features. This makes the extracted features more representative and fully reflects the differences in image features corresponding to the degree of hemolysis, providing rich and effective feature support for subsequent hemolysis prediction.

[0135] In the feature detection network, the pooling module, in conjunction with subsequent convolutional and fully connected modules, enables the selection and refinement of features. The pooling module can reduce the size of the feature map while retaining key features, further improving computational efficiency and avoiding overfitting. The subsequent convolutional module further refines and extracts the pooled features, while the fully connected module maps high-dimensional features into more compact feature vectors, completing the transformation from image features to feature information. This ensures that the output feature information is both concise and rich in hemolysis-related key information, providing high-quality input for the accurate prediction of the fusion detection network.

[0136] The fusion detection network adopts an input-stage convolution-pooling-convolution-fully connected structure, performing specialized hemolysis prediction processing on the image feature information output by the feature detection network. The continuous computation of the multi-stage convolution module enables deep processing of the input feature information, further strengthening hemolysis-related features and suppressing irrelevant interference information. The pooling module further filters core features, improving feature robustness. Finally, the fully connected module is specifically responsible for mapping the processed feature information to hemolysis detection results, achieving a precise conversion from feature information to hemolysis degree prediction, ensuring that the output detection results can directly and accurately represent the hemolysis state corresponding to the blood image.

[0137] The structural design of the entire hemolysis image detection sub-model closely aligns with the application's goal of "accurately detecting the degree of hemolysis in red blood cells." Through the collaborative division of labor between the feature detection network and the fusion detection network, it achieves efficient extraction of multi-dimensional and deep-level features from blood images. Furthermore, targeted feature processing and mapping ensure the accuracy and reliability of hemolysis prediction. The orderly connection and structural selection of each module balance computational efficiency and model performance while fully leveraging key information related to hemolysis in blood images. This effectively improves the sensitivity and accuracy of image-based hemolysis detection, providing high-quality image-side detection support for the entire blood hemolysis composite detection model.

[0138] The blood parameters can be collected using a fully automated coagulation analyzer.

[0139] The parameter detection sub-model can be trained using a set of blood parameter information as the training dataset. The objective function is to minimize the error between the output predicted value and the label information. The parameters of the parameter detection sub-model are then solved to obtain the model parameters. The steepest descent method can be used as the specific algorithm for solving the model parameters.

[0140] The step of training a pre-defined hemolysis image detection sub-model using the blood image set to obtain a trained hemolysis image detection sub-model includes:

[0141] Initialize the number of training iterations;

[0142] The blood images in the blood image set are used as input data and input into the hemolysis image detection sub-model;

[0143] The input data is processed using the hemolysis image detection sub-model to obtain a second hemolysis detection result;

[0144] The difference calculation process is performed on the obtained second hemolysis detection result and the label information corresponding to the input data to obtain the difference value;

[0145] Determine whether the difference value satisfies the convergence condition to obtain the first determination result;

[0146] When the first judgment result is negative, it is determined whether the training iteration count value is equal to the training count threshold to obtain the second judgment result;

[0147] When the second judgment result is negative, the model training state is determined to be that the termination training condition is not met.

[0148] When the second judgment result is yes, the model training state is determined to meet the termination training condition;

[0149] When the first judgment result is yes, it is determined that the model training state meets the termination training condition;

[0150] When the training state of the model does not meet the termination training condition, the parameters of the hemolysis image detection sub-model are updated using the parameter update model, the training iteration number is increased by 1, and the blood images in the blood image set are used as input data to the hemolysis image detection sub-model.

[0151] When the model training state meets the termination training condition, the training process of the hemolysis image detection sub-model is completed, and the trained hemolysis image detection sub-model is obtained.

[0152] The parameter update model is as follows:

[0153] ,

[0154] ;

[0155] In the formula, For the blood image set, the first A blood image, For the blood image set, the first Label information for each blood image, For loss function, Update the parameter value. For the parameters of the module to be updated, The initial parameter is the learning rate. For momentum angle parameters, , Indicates the variable Find the partial derivative. This indicates that the hemolysis image detection sub-model is effective for the first hemolysis image in the training dataset. The second hemolysis detection result is calculated from a blood image. This is the computation function corresponding to the hemolysis image detection sub-model; Represents a constant Exponentiation; and This is a preset constant value; and These are the preset weighting coefficients.

[0156] The difference value satisfies the convergence condition when it is less than a preset convergence threshold; the difference value does not satisfy the convergence condition when it is not less than a preset convergence threshold.

[0157] The difference calculation process can be implemented using a loss function.

[0158] The loss function can be the cross-entropy loss function.

[0159] The boundary limitation process involves determining whether the value of each data point in the normalized image is within a preset value range, and deleting data points outside the preset value range from the normalized image to obtain a preprocessed image.

[0160] A second aspect of the present invention discloses an image processing device for determining hemolysis of suspended red blood cells, the device comprising:

[0161] Memory containing executable program code;

[0162] A processor coupled to the memory;

[0163] The processor calls the executable program code stored in the memory to execute the image processing method for determining hemolysis of suspended red blood cells.

[0164] In a third aspect of this invention, a computer-storable medium is disclosed, the computer-storable medium storing computer instructions, which, when invoked by a computer, are used to execute the image processing method for determining hemolysis of suspended red blood cells.

[0165] In a fourth aspect of this invention, an information data processing terminal is disclosed, which is used to implement the image processing method for determining hemolysis of suspended red blood cells.

[0166] The above description is merely an embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of this application should be included within the scope of the claims of this application.

Claims

1. An image processing method for determining hemolysis of suspended red blood cells, characterized in that, include: S1, acquire the blood image set; The blood image set includes several blood images and corresponding label information; S2, obtain the set of blood parameter information; The blood parameter information set includes several blood parameters and corresponding label information; the blood parameters include hemoglobin concentration, reticulocyte concentration, indirect bilirubin concentration, lactate dehydrogenase concentration, and globin concentration. S3, using the blood image set and blood parameter information set, a blood hemolysis complex detection model is constructed; S4, Based on the blood hemolysis composite detection model, the acquired blood images and blood parameters are processed to obtain hemolysis detection result information; the hemolysis detection result information is used to characterize the degree of erythrocyte hemolysis in the blood corresponding to the acquired blood images and blood parameters; The blood hemolysis composite detection model includes: a hemolysis image detection sub-model, a parameter detection sub-model, and a fusion sub-model; The input terminal of the hemolysis image detection sub-model is used to receive the obtained blood image; The input terminal of the parameter detection sub-model is used to receive blood parameters. The fusion sub-model is connected to the output of the hemolysis image detection sub-model and the output of the parameter detection sub-model, respectively, and is used to perform fusion calculation on the output information of the two sub-models to obtain hemolysis detection result information. The hemolysis image detection sub-model includes: Color features are calculated on the input blood image to obtain a set of color feature moments; the set of color feature moments includes the first-order color moments, second-order color moments, and third-order color moments of the three color channels: R channel, G channel, and B channel; Texture features are calculated on the input blood image to obtain contrast and entropy values; The shape and structure features of the input blood image are calculated to obtain the pixel values ​​of corner points, texture feature points, and SIFT feature points. Using a preset color moment fusion model, the color feature moment set is calculated to obtain color feature values; the expression of the preset color moment fusion model is: ， ， in, This is the color composite value for one color channel. Let i be the i-th order color moment of a color channel. For a color channel, the preset i-th order standard color moment is... and These are all model parameters to be solved during the training process. For color feature values, This represents the color composite value of the j-th color channel. When j=1, 2, and 3, it represents the R channel, G channel, and B channel, respectively. A feature vector is constructed using the color feature value, contrast value, entropy value, corner pixel value, texture feature point pixel value, and SIFT feature point pixel value. ; The first hemolysis detection calculation is performed on the feature vector to obtain the first hemolysis detection result; The expression for calculating the first hemolysis detection is: in, This is the first result of the hemolysis test. and All are model vectors. and These are all model parameters to be solved during the training process.

2. The image processing method for determining hemolysis of suspended red blood cells as described in claim 1, characterized in that, The blood hemolysis composite detection model is used to process the acquired blood images and blood parameters to obtain hemolysis detection results, including: The acquired blood images are preprocessed to obtain preprocessed images; Using the blood hemolysis composite detection model, the preprocessed image and blood parameters are processed to obtain hemolysis detection results.

3. The image processing method for determining hemolysis of suspended red blood cells as described in claim 1, characterized in that, The method of constructing a blood hemolysis complex detection model using the blood image set and blood parameter information set includes: Initialize the blood hemolysis complex detection model; The blood image set is used to train the hemolysis image detection sub-model to obtain the trained hemolysis image detection sub-model. Using the blood parameter information set, the parameter detection sub-model is trained to obtain the trained parameter detection sub-model; Using the trained hemolysis image detection sub-model, the trained parameter detection sub-model, and the fusion sub-model, a blood hemolysis composite detection model is constructed.

4. The image processing method for determining hemolysis of suspended red blood cells as described in claim 1, characterized in that, The parameter detection sub-model includes: Blood parameters collected at several time points are merged to obtain a set of blood parameter sequences; the set of blood parameter sequences includes hemoglobin concentration sequence, reticulocyte concentration sequence, indirect bilirubin concentration sequence, lactate dehydrogenase concentration sequence, and globin concentration sequence; the elements in the sequence are blood parameters collected at each time point; Obtain the standard value for each blood parameter; By subtracting the standard value of the corresponding blood parameter from each sequence in the blood parameter sequence set, the corresponding difference sequence is obtained; By using all the difference sequences as row vectors, a difference matrix is ​​constructed. Extract the maximum and minimum values ​​of all row vectors in the difference matrix; The difference matrix is ​​used to perform a second hemolysis detection calculation to obtain the second hemolysis detection result; The expression for calculating the second hemolysis detection is: ， In the formula, For the second hemolysis detection result, N and M are the row and column dimensions of the difference matrix, respectively. and Let represent the maximum and minimum values ​​of the i-th row vector of the difference matrix, respectively. The element in the i-th row and j-th column of the difference matrix is... and , j=1,2,…,M, are all model parameters to be solved during the training process.

5. The image processing method for determining hemolysis of suspended red blood cells as described in claim 1, characterized in that, The hemolysis image detection sub-model includes: a feature detection network and a fusion detection network; The feature detection network is used to process the blood image to obtain image feature information; the fusion detection network is used to perform hemolysis prediction processing on the image feature information to obtain a second hemolysis detection result. The feature detection network includes a first input module, a first convolution module, a depthwise separable convolution module, a first up-dimensional convolution module, a second up-dimensional convolution module, a third up-dimensional convolution module, a fourth up-dimensional convolution module, a second convolution module, a first pooling module, a third convolution module, and a first fully connected module. The input terminal of the first input module of the feature detection network is used to receive the blood image; the output terminal of the first input module of the feature detection network is connected to the input terminal of the first convolution module of the feature detection network; the output terminal of the first convolution module of the feature detection network is connected to the input terminal of the depth-separable convolution module of the feature detection network; the output terminal of the depth-separable convolution module of the feature detection network is connected to the input terminal of the first up-dimensional convolution module of the feature detection network; the output terminal of the first up-dimensional convolution module of the feature detection network is connected to the input terminal of the second up-dimensional convolution module of the feature detection network; the output terminal of the second up-dimensional convolution module of the feature detection network is connected to the input terminal of the feature detection image. The input of the third up-dimensional convolutional module of the feature detection network is connected to the input of the fourth up-dimensional convolutional module of the feature detection network; the output of the fourth up-dimensional convolutional module of the feature detection network is connected to the input of the second convolutional module of the feature detection network; the output of the second convolutional module of the feature detection network is connected to the input of the first pooling module of the feature detection network; the output of the first pooling module of the feature detection network is connected to the input of the third convolutional module of the feature detection network; and the output of the third convolutional module of the feature detection network is connected to the input of the first fully connected module of the feature detection network. The fusion detection network includes a second input module, a fourth convolution module, a fifth convolution module, a sixth convolution module, a second pooling module, a seventh convolution module, and a second fully connected module; The input terminal of the second input module of the fusion detection network is connected to the output terminal of the first fully connected module of the feature detection network; the output terminal of the second input module of the fusion detection network is connected to the input terminal of the fourth convolution module of the fusion detection network; the output terminal of the fourth convolution module of the fusion detection network is connected to the input terminal of the fifth convolution module of the fusion detection network; the output terminal of the fifth convolution module of the fusion detection network is connected to the input terminal of the sixth convolution module of the fusion detection network; the output terminal of the sixth convolution module of the fusion detection network is connected to the input terminal of the second pooling module of the fusion detection network; the output terminal of the second pooling module of the fusion detection network is connected to the input terminal of the seventh convolution module of the fusion detection network; the output terminal of the seventh convolution module of the fusion detection network is connected to the input terminal of the second fully connected module of the fusion detection network; the output terminal of the second fully connected module of the fusion detection network is used to output the second hemolysis detection result of the predicted blood image.

6. An image processing device for determining hemolysis of suspended red blood cells, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the image processing method for determining hemolysis of suspended red blood cells as described in any one of claims 1 to 5.

7. A computer-storable medium, characterized in that, The computer storage medium stores computer instructions, which, when invoked by the computer, are used to execute the image processing method for determining hemolysis of suspended red blood cells as described in any one of claims 1 to 5.

8. An information data processing terminal, characterized in that, The information data processing terminal is used to implement the image processing method for determining hemolysis of suspended red blood cells as described in any one of claims 1 to 5.

Images