A diagnostic device for infectious hydronephrosis based on CT images

Abstract

The present invention discloses a diagnostic device for infectious hydronephrosis based on CT images, comprising: a segmentation model training module, which is used to train a segmentation model based on a segmentation image set and a report text description set of a segmented area of ​​a kidney-related area; a hydronephrosis diagnostic model training module, which is used to train an infectious hydronephrosis diagnostic model based on the segmentation image set, the report text description set and the patient's clinical information description set; and a hydronephrosis diagnostic module, which is used to input the CT image of the patient to be diagnosed into the segmentation model and the infectious hydronephrosis diagnostic model to obtain a hydronephrosis diagnosis result. The present invention combines the CT image set and the description information of the area to be segmented to perform image segmentation, and can improve the segmentation model to distinguish between the real lesion area and the illusion caused by the imaging technology or conditions through the description information in the report text, thereby reducing the false positives and false negatives in the segmentation process, and is conducive to improving the accuracy of the segmentation of the kidney organ area and the hydronephrosis area in the kidney.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging technology, and in particular to a CT image-based infectious hydronephrosis diagnosis device. Background Art

[0002] Hydronephrosis is a common disease in urology. It is mostly caused by urinary tract obstruction due to urinary stones, congenital malformations and ureteral stenosis, which makes urine flow out poorly and accumulate in the renal pelvis. Hydronephrosis is prone to infection. When hydronephrosis occurs, the pressure in the renal pelvis increases, the renal tissue is easily damaged, and bacterial colonization, invasion and proliferation often occur. Bacteria or endotoxins in infectious hydronephrosis can flow back into the blood, causing systemic infection.

[0003] Urinary tract obstruction caused by urinary stones is often complicated by hydronephrosis. When patients with stones and infected hydronephrosis undergo lithotripsy, fluid infusion may lead to high pressure in the renal pelvis, inducing reflux of infected hydronephrosis into the renal tissue, leading to systemic inflammatory response syndrome and even urinary sepsis, which endangers the patient's life. According to studies, the proportion of patients with fever after percutaneous nephrolithotomy is about 26%-37%, and the incidence of systemic inflammatory response syndrome is about 9.8%-34.7%; the incidence of infectious complications and systemic inflammatory response syndrome after ureteroscopy lithotripsy is 8.1-8.7%. Among stone patients, 3%-5.8% of systemic inflammatory response syndrome can develop into life-threatening septic shock, urinary sepsis is associated with death in 43% of stone patients, and the mortality rate of patients with septic shock after lithotripsy is as high as 50%-66%. Therefore, it is particularly important to evaluate the infection of hydronephrosis before lithotripsy.

[0004] Existing diagnostic devices for infectious hydronephrosis usually use bladder urine routine examination and bladder urine bacterial culture to determine whether the hydronephrosis is infected. However, the test samples of these two methods are bladder urine specimens. In fact, it is the mixed urine from urine from both kidneys that enters the bladder to diagnose whether the hydronephrosis is infected. It is difficult to evaluate whether unilateral hydronephrosis is complicated by infection, resulting in low accuracy in the diagnosis of infectious hydronephrosis. Summary of the invention

[0005] The present invention provides an infectious hydronephrosis diagnostic device based on CT images to solve the technical problem that the existing infectious hydronephrosis diagnostic device diagnoses whether the hydronephrosis is infected by using the mixed urine after the urine from both kidneys enters the bladder, and it is difficult to evaluate whether the unilateral hydronephrosis is complicated with infection, resulting in low diagnostic accuracy of infectious hydronephrosis.

[0006] The present invention provides a CT image-based infectious hydronephrosis diagnostic device, comprising:

[0007] An image set segmentation module is used to segment the segmented areas of the CT images of the hydronephrosis patient to obtain a segmented image set; the segmented areas include the kidney organ area and the hydronephrosis infection area in the kidney;

[0008] A report text description set extraction module is used to obtain an inspection report matching the CT image, extract the report text of the segmented area from the inspection report, and obtain a report text description set of the segmented area;

[0009] A segmentation model training module, used for training a segmentation model based on the segmented image set and the report text description set;

[0010] A clinical information description set acquisition module is used to acquire basic clinical data and hydronephrosis infection examination results of the patient with hydronephrosis to obtain a clinical information description set of the patient;

[0011] A hydronephrosis diagnostic model training module is used to train an infectious hydronephrosis diagnostic model based on the segmented image set, the report text description set and the patient clinical information description set;

[0012] The hydronephrosis diagnosis module is used to input the CT image of the patient to be diagnosed into the segmentation model to obtain a segmented image set to be diagnosed, and input the segmented image set to be diagnosed into the infectious hydronephrosis diagnosis model to obtain a diagnostic result of infectious hydronephrosis of the patient to be diagnosed.

[0013] Furthermore, the image set segmentation module is also used for:

[0014] Acquire a CT image of a patient with hydronephrosis, acquire a plurality of key frame images of the segmented area based on the CT image, and form a CT image key frame set with the plurality of key frame images;

[0015] The segmented regions in the CT image key frame set are labeled and segmented to obtain a segmented image set.

[0016] Furthermore, the marking and segmentation processing of the segmented areas in the CT image key frame set is also used for:

[0017] Performing windowing processing on the CT image key frame set, and standardizing the windowed CT image key frame set using a z-score method to obtain a standard CT image key frame set;

[0018] The segmented areas in the standard CT image key frame set are labeled and segmented.

[0019] Furthermore, the segmentation model training module is also used for:

[0020] Using an image encoder to encode the segmented image set to obtain an encoded image set;

[0021] Using a text encoder to encode the report text description set to obtain an encoded text set;

[0022] The encoded image set and the encoded text set are input into the pre-trained large model for model training. During the model training process, the decoder of the SAM model is used to output the target segmentation mask prediction result, and the IOU loss function is used as the training loss function.

[0023] Furthermore, the clinical information description set acquisition module is also used to:

[0024] The basic clinical data and hydronephrosis infection examination results are organized into tabular data, and the patient ID is associated with the CT examination image number and the CT examination report information to establish a clinical information description set; the basic clinical data include age, gender, BMI, body temperature and whether there is low back pain; the hydronephrosis infection examination results include urine leukocyte examination results, blood leukocyte examination results and hydronephrosis bacterial culture results.

[0025] Furthermore, the hydronephrosis diagnostic model training module is also used for:

[0026] After performing data enhancement operation on the segmented image set, the segmented image set is input into a pre-trained ResNet34 model for embedding representation to obtain segmented image features; the input channel of the ResNet34 model is 1, and an attention mechanism is added after the spatial attention module and the channel attention module of the ResNet34 model, and the segmented image features include kidney organ image features and kidney hydronephrosis area image features;

[0027] Inputting the report text description set into the Chinese pre-trained Bert model to obtain text description features;

[0028] Encoding the patient clinical information description set to obtain a patient clinical information description encoding feature;

[0029] The coding features of the patient's clinical information description, the segmented image features and the text description are serially concatenated to obtain the overall features of the patient;

[0030] The overall characteristics of the patient are input into the classifier network for training to obtain a diagnosis prediction result; wherein, during the training process, cross entropy is used as a training patrol function of the classifier network, and the training ends when the loss converges.

[0031] Furthermore, the patient clinical information description set includes component variable features and continuous variable features, and the encoding process of the patient clinical information description set to obtain the patient clinical information description encoding features includes:

[0032] The categorical variable feature information is converted into digital representation according to the preset label code; the categorical variable features include the patient's gender, whether there is low back pain, and the bacterial culture result of hydronephrosis;

[0033] The digital features of the continuous variable features are extracted, and the digital representation and the digital features are combined to obtain the patient's clinical information description coding features, wherein the continuous variable features include the patient's age, BMI, body temperature, urine leukocyte test results, and blood leukocyte test results.

[0034] The embodiment of the present invention trains a segmentation model based on a segmented image set and a report text description set, and performs image segmentation in combination with a CT image set and description information of the area to be segmented. The segmentation model can be improved through the description information in the report text to distinguish between real lesion areas and illusions caused by imaging technology or conditions, thereby reducing false positives and false negatives in the segmentation process, which is beneficial to improving the accuracy of segmentation of kidney organ areas and hydronephrosis areas in the kidney.

[0035] Furthermore, the embodiment of the present invention trains an infectious hydronephrosis diagnostic model based on the segmented image set, the report text description set and the patient clinical information description set. The combination of image data and multimodal data enables multimodal learning of the diagnostic model, and comprehensively considers the images of the report text description set and the patient clinical information description set for the diagnosis of hydronephrosis, thereby further improving the accuracy of the diagnosis of infectious hydronephrosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 Schematic diagram of the structure of a CT imaging-based infectious hydronephrosis diagnostic device provided by an embodiment of the present invention;

[0037] Figure 2 is a schematic diagram of hydronephrosis region segmentation provided by an embodiment of the present invention;

[0038] Figure 3 It is a schematic diagram of the architecture of the segmentation model provided by an embodiment of the present invention.

[0039] Figure 4 It is a schematic diagram of the architecture of the diagnosis model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0041] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the feature. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0042] In the description of this application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0043] See also Figure 1 The present invention provides a diagnostic device for infectious hydronephrosis based on CT images, comprising:

[0044] The image set segmentation module 10 is used to segment the segmented areas of the CT images of the hydronephrosis patient to obtain a segmented image set; the segmented areas include the kidney organ area and the hydronephrosis infection area in the kidney;

[0045] A report text description set extraction module 20 is used to obtain an inspection report matching the CT image, extract the report text of the segmented area from the inspection report, and obtain a report text description set of the segmented area;

[0046] In an embodiment of the present invention, the text examination report matching the CT image can be queried based on the patient's CT image examination number list, and then the radiologist can extract the description of the kidney organ and the hydronephrosis area in the examination report, and refine and restate it in combination with the specific imaging and professional background knowledge to form a coherent, accurate, independent and detailed text description of the imaging report, and then establish a text description set of the imaging report corresponding to the kidney organ and the hydronephrosis area in the kidney in the CT image.

[0047] The segmentation model training module 30 is used to train a segmentation model based on the segmentation image set and the report text description set;

[0048] A clinical information description set acquisition module 40 acquires basic clinical data of the hydronephrosis patient and the hydronephrosis infection examination results to obtain the patient's clinical information description set;

[0049] The hydronephrosis diagnosis model training module 50 is used to train an infectious hydronephrosis diagnosis model based on the segmented image set, the report text description set and the patient clinical information description set;

[0050] The hydronephrosis diagnosis module 60 is used to input the CT image of the patient to be diagnosed into the segmentation model to obtain the segmented image set to be diagnosed, and input the segmented image set to be diagnosed into the infectious hydronephrosis diagnosis model to obtain the infectious hydronephrosis diagnosis result of the patient to be diagnosed.

[0051] The embodiment of the present invention trains a segmentation model based on a segmented image set and a report text description set, and performs image segmentation in combination with a CT image set and description information of the area to be segmented. The segmentation model can be improved through the description information in the report text to distinguish between real lesion areas and illusions caused by imaging technology or conditions, thereby reducing false positives and false negatives in the segmentation process, which is beneficial to improving the accuracy of segmentation of kidney organ areas and hydronephrosis areas in the kidney.

[0052] Furthermore, the embodiment of the present invention trains an infectious hydronephrosis diagnostic model based on the segmented image set, the report text description set and the patient clinical information description set. The combination of image data and multimodal data enables multimodal learning of the diagnostic model, and comprehensively considers the images of the report text description set and the patient clinical information description set for the diagnosis of hydronephrosis, thereby further improving the accuracy of the diagnosis of infectious hydronephrosis.

[0053] In one embodiment, the image set segmentation module 10 is further used for:

[0054] Acquire a CT image of a patient with hydronephrosis, acquire a plurality of key frame images of a segmented region based on the CT image, and form a CT image key frame set with the plurality of key frame images;

[0055] In a specific embodiment of the present invention, 351 is collected for CT plain scan image information of patients, of which 234 patients are negative for bacterial infection and 117 patients are positive for bacterial infection. The CT images of the corresponding patients with hydronephrosis are queried according to the list of CT image examination numbers of the included patients, and the key frame images representative of the hydronephrosis area in the kidney are selected to obtain the CT image key frame set associated with the patient set. In a specific implementation, the key frame images can be selected by an radiologist.

[0056] The segmented regions in the CT image key frame set are labeled and segmented to obtain a segmented image set.

[0057] In the embodiment of the present invention, an image segmentation and annotation tool may be used to perform fine boundary annotation on the segmented area of ​​the key frame of the CT image to obtain an annotated key frame image.

[0058] Then, the labeled key frame images are segmented. According to the labeling results, the kidney organs and hydronephrosis areas are segmented and extracted to obtain segmented image sets. The segmented image sets obtained by segmentation can provide a visual display effect, thereby effectively improving the recognition effect of the morphology, size, position and other characteristics of the kidney organs and hydronephrosis, thereby effectively improving the accuracy of the diagnosis of infectious hydronephrosis.

[0059] In the embodiment of the present invention, the mask information of the kidney organ and the mask information of the hydronephrosis region in the kidney can be obtained through segmentation, and a segmented image set of the kidney organ and the hydronephrosis region in the kidney is established based on these two types of information.

[0060] In one embodiment, the segmented regions in the CT image key frame set are labeled and segmented, and are also used to:

[0061] Performing windowing processing on the CT image key frame set, and using a z-score method to standardize the windowed CT image key frame set to obtain a standard CT image key frame set;

[0062] The segmented areas in the standard CT image key frame set are labeled and segmented.

[0063] In an embodiment of the present invention, the windowing process of the CT plain scan image set can be realized by adjusting the window width and window level. For example, the windowing process can set the window width to 250 and the window level to 80. The contrast between the hydronephrosis area and the surrounding normal tissue can be increased by the windowing process, so that it is easier to observe and segment. Moreover, the gray value of the hydronephrosis area can be reasonably distributed within the display range by the windowing process, which is conducive to observing and identifying the morphology and position of the hydronephrosis. At the same time, the z-score method is used to standardize the windowed CT image, so that the gray value of the image is converted into a standard normal distribution, so that the influence of different scanning devices, parameters and other factors on the image gray can be effectively eliminated, and the gray value range of different images can be unified, which is convenient for quantitative analysis and comparison, and the format conversion can be unified to obtain a standard CT image key frame set in PNG format.

[0064] See also Figure 2 , is a schematic diagram of hydronephrosis region segmentation provided by an embodiment of the present invention. In the embodiment of the present invention, by marking and segmenting the segmented regions in the standard CT image key frame set, the segmentation accuracy of the kidney organ and the hydronephrosis region in the kidney can be effectively improved.

[0065] In one embodiment, the segmentation model training module 30 is further used to:

[0066] The segmented image set is encoded by using an image encoder to obtain an encoded image set; the encoded image is an image embedding representation corresponding to the segmented image.

[0067] In an embodiment of the present invention, the image encoder may be an OpenAI CLIP ViT model, and the image size may be uniformly processed to 384*384.

[0068] A text encoder is used to encode the report text description set to obtain an encoded text set; the encoded text set is the embedding representation corresponding to the report text.

[0069] In an embodiment of the present invention, the text encoder may be a Chinese pre-trained Bert model, and the maximum length of the text does not exceed 512.

[0070] See also Figure 3 , is a schematic diagram of the architecture of a segmentation model provided in an embodiment of the present invention.

[0071] The encoded image set and the encoded text set are input into the pre-trained large model for model training. During the model training process, the decoder of the SAM model is used to output the target segmentation mask prediction result. The IOU loss function is used as the training loss function to measure the overlap accuracy of the predicted output kidney organ mask and the kidney hydronephrosis area mask with the manually annotated data.

[0072] In an embodiment of the present invention, the above two embedded representations, namely the encoded image set and the encoded text set, are input into the pre-trained large model LLaVA for training, and when convergence occurs, a segmentation model that can be used to segment the kidney organ and the water accumulation area in the kidney is obtained.

[0073] In one embodiment, the clinical information description set acquisition module 40 is further used to:

[0074] Basic clinical data and hydronephrosis infection examination results were organized into tabular data, and the patient ID was associated with the CT examination image number and CT examination report information to establish a clinical information description set; basic clinical data included age, gender, BMI, body temperature, and whether there was low back pain; hydronephrosis infection examination results included urine leukocyte examination results, blood leukocyte examination results, and hydronephrosis bacterial culture results.

[0075] In an embodiment of the present invention, the outpatient and hospital admission information of the patient at the time of the CT examination can be retrieved in the information system according to the patient ID information, so as to obtain the basic clinical data required for the subsequent establishment of the diagnostic model. The basic clinical information included in the clinical information description set specifically includes: gender, age, BMI, body temperature, and whether there is low back pain; the urine, blood, and hydronephrosis examination results of the patient at the time of the CT examination can be retrieved to obtain the examination results required for the subsequent establishment of the diagnostic model. The clinical examination result indicators included in the clinical information description set specifically include the urine leukocyte examination results, the blood leukocyte examination results, and the hydronephrosis bacterial culture results.

[0076] The bacterial culture results of hydronephrosis are divided into positive bacterial infection and negative bacterial infection (which can be used as the gold standard for whether the patient's hydronephrosis is complicated by infection). The above clinical data and examination results are organized into a table, and the patient ID is associated with the CT examination image number and CT examination report information to establish a patient clinical information description set, among which gender, age, BMI, body temperature, whether there is low back pain, urine white blood cell examination results and blood white blood cell examination results can be used as modeling features.

[0077] In one embodiment, the hydronephrosis diagnosis model training module 50 is further used to:

[0078] After data enhancement, the segmented image set is input into the pre-trained ResNet34 model for embedding representation to obtain the segmented image features. The input channel of the ResNet34 model is 1, and an attention mechanism is added after the spatial attention module and channel attention module of the ResNet34 model. The segmented image features include the image features of the kidney organ and the image features of the hydronephrosis area in the kidney.

[0079] In the embodiment of the present invention, the data enhancement operation may include rotation, translation, scaling, cropping, elastic deformation, etc.

[0080] See also Figure 4 , which is a schematic diagram of the architecture of a diagnostic model provided in an embodiment of the present invention. In an embodiment of the present invention, by modifying the input channel of the ResNet34 model to 1, it is possible to process grayscale images through a single channel, thereby improving the segmentation effect of the segmented image set. And the spatial attention module and channel attention module of the ResNet34 model in the embodiment of the present invention are followed by an attention mechanism, which can enable the model to focus on the most important areas in the image for the segmentation task and ignore irrelevant background or other structures, thereby improving the expressiveness of features, and further effectively improving segmentation accuracy and diagnostic accuracy.

[0081] Input the report text description set into the Chinese pre-trained Bert model to obtain the text description features;

[0082] Encoding the patient's clinical information description set to obtain the patient's clinical information description coding features;

[0083] The coding features, segmented image features and text descriptions of the patient's clinical information are concatenated and spliced ​​to obtain the patient's overall characteristics;

[0084] The overall characteristics of the patient are input into the classifier network for training to obtain the diagnosis prediction results; during the training process, cross entropy is used as the training patrol function of the classifier network, and the training ends when the loss converges.

[0085] In an embodiment of the present invention, the coding features of the patient's clinical information description, the segmented image features and the text description are serially spliced ​​to obtain the overall characteristics of the patient. By combining a variety of feature information, the model can obtain comprehensive patient information, and different feature information can complement each other. At the end of training, a hydronephrosis diagnosis model that integrates multimodal information can be obtained, thereby effectively improving the accuracy of diagnosis of infectious hydronephrosis.

[0086] The diagnostic model trained by the embodiment of the present invention can capture subtle feature differences, avoid the impact of human errors and subjective errors on the diagnostic results, and thus further improve the accuracy of the diagnosis of hepatoelastic hydronephrosis.

[0087] In one embodiment, the patient clinical information description set includes component variable features and continuous variable features. The patient clinical information description set is coded to obtain the patient clinical information description coding features, including:

[0088] The categorical variable feature information is converted into digital representation according to the preset label coding; the categorical variable features include the patient's gender, whether there is low back pain, and the bacterial culture results of hydronephrosis;

[0089] The digital features of continuous variable features were extracted, and the digital representation and digital features were combined to obtain the patient's clinical information description coding features. The continuous variable features included patient age, BMI, body temperature, urine leukocyte test results, and blood leukocyte test results.

[0090] In one embodiment, after obtaining the infectious hydronephrosis diagnostic model, the method further comprises:

[0091] The verification patient data is input into the fusion hydronephrosis diagnosis model 60, and the corresponding verification diagnosis index is calculated based on the output result. In the embodiment of the present invention, the pre-processed data set of all patients can be randomly divided into a training set and a verification set according to a ratio of 2:1, and the training set has 234 people (including 113 patients with negative bacterial infection and 121 patients with negative bacterial infection, with a total of 14960 frames of kidney CT plain scan images containing hydronephrosis), and the verification set has 117 people (including 61 patients with negative bacterial infection and 56 patients with negative bacterial infection, with a total of 7480 frames of kidney CT plain scan images containing hydronephrosis).

[0092] Both the test diagnostic indicators and the validation diagnostic indicators include: sensitivity, specificity, accuracy, positive predictive value, negative predictive value and AUC value of the ROC curve. The above test diagnostic indicators obtained from the training set are 0.867, 0.891, 0.838, 0.882, 0.888, and 0.886, respectively; the above validation diagnostic indicators obtained from the validation set are 0.882, 0.894, 0.846, 0.885, 0.873, and 0.872, respectively.

[0093] The infectious hydronephrosis diagnostic device based on CT images provided by the embodiment of the present invention can achieve an accuracy of 86.4%, a sensitivity of 85.9%, and a specificity of 88.1% for diagnosing infectious hydronephrosis.

[0094] The implementation of the embodiments of the present invention has the following beneficial effects:

[0095] The embodiment of the present invention trains a segmentation model based on a segmented image set and a report text description set, and performs image segmentation in combination with a CT image set and description information of the area to be segmented. The segmentation model can be improved through the description information in the report text to distinguish between real lesion areas and illusions caused by imaging technology or conditions, thereby reducing false positives and false negatives in the segmentation process, which is beneficial to improving the accuracy of segmentation of kidney organ areas and hydronephrosis areas in the kidney.

[0096] Furthermore, the embodiment of the present invention trains an infectious hydronephrosis diagnostic model based on the segmented image set, the report text description set and the patient clinical information description set. The combination of image data and multimodal data enables multimodal learning of the diagnostic model, and comprehensively considers the images of the report text description set and the patient clinical information description set for the diagnosis of hydronephrosis, thereby further improving the accuracy of the diagnosis of infectious hydronephrosis.

[0097] The above specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A diagnostic device for infectious hydronephrosis based on CT images, characterized in that: include: An image set segmentation module is used to segment the segmented areas of the CT images of the hydronephrosis patient to obtain a segmented image set; the segmented areas include the kidney organ area and the hydronephrosis infection area in the kidney; A report text description set extraction module is used to obtain an inspection report matching the CT image, extract the report text of the segmented area from the inspection report, and obtain a report text description set of the segmented area; The segmentation model training module is used to train a segmentation model based on the segmented image set and the report text description set; it is also used to: use an image encoder to encode the segmented image set to obtain an encoded image set; use a text encoder to encode the report text description set to obtain an encoded text set; input the encoded image set and the encoded text set into a pre-trained large model for model training, and use the decoder of the SAM model to output the target segmentation mask prediction result during the model training process, and use the IOU loss function as the training loss function; A clinical information description set acquisition module is used to acquire the basic clinical data and hydronephrosis infection examination results of the hydronephrosis patient to obtain the patient's clinical information description set; it is also used to: organize the basic clinical data and hydronephrosis infection examination results into table data, associate the patient ID with the CT examination image number and the CT examination report information, and establish a clinical information description set; the basic clinical data includes age, gender, BMI, body temperature and whether there is low back pain; the hydronephrosis infection examination results include urine leukocyte examination results, blood leukocyte examination results and hydronephrosis bacterial culture results; The hydronephrosis diagnostic model training module is used to train an infectious hydronephrosis diagnostic model based on the segmented image set, the report text description set and the patient clinical information description set; it is also used to: after performing data enhancement operation on the segmented image set, input it into the pre-trained ResNet34 model for embedding representation to obtain segmented image features; the input channel of the ResNet34 model is 1, and the spatial attention module and channel attention module of the ResNet34 model are followed by an attention mechanism, and the segmented image features include kidney organ image features and kidney hydronephrosis area image features; the report text description set is input into the Chinese pre-trained Bert model to obtain text description features; the patient clinical information description set is encoded to obtain patient clinical information description encoding features; the patient clinical information description encoding features, the segmented image features and the text description are encoded The patient's overall characteristics are obtained by performing serial concatenation processing; the patient's overall characteristics are input into the classifier network for training to obtain a diagnosis prediction result; wherein, during the training process, cross entropy is used as the training patrol function of the classifier network, and the training ends when the loss converges; the patient's clinical information description set includes categorical variable characteristics and continuous variable characteristics, and the patient's clinical information description set is encoded to obtain the patient's clinical information description encoding characteristics, including: converting the categorical variable characteristic information into a digital representation according to the label preset encoding; the categorical variable characteristics include the patient's gender, whether there is low back pain, and the bacterial culture results of hydronephrosis; extracting the digital characteristics of the continuous variable characteristics, combining the digital representation with the digital characteristics to obtain the patient's clinical information description encoding characteristics, and the continuous variable characteristics include the patient's age, BMI, body temperature, urine leukocyte test results, and blood leukocyte test results; The hydronephrosis diagnosis module is used to input the CT image of the patient to be diagnosed into the segmentation model to obtain a segmented image set to be diagnosed, and input the segmented image set to be diagnosed into the infectious hydronephrosis diagnosis model to obtain a diagnostic result of infectious hydronephrosis of the patient to be diagnosed.

2. The infectious hydronephrosis diagnostic device based on CT images according to claim 1, characterized in that: The image set segmentation module is also used for: Acquire a CT image of a patient with hydronephrosis, acquire a plurality of key frame images of the segmented area based on the CT image, and form a CT image key frame set with the plurality of key frame images; The segmented regions in the CT image key frame set are labeled and segmented to obtain a segmented image set.

3. The infectious hydronephrosis diagnostic device based on CT images according to claim 2, characterized in that: The marking and segmentation processing of the segmented areas in the CT image key frame set is also used for: Performing windowing processing on the CT image key frame set, and standardizing the windowed CT image key frame set using a z-score method to obtain a standard CT image key frame set; The segmented areas in the standard CT image key frame set are labeled and segmented.

Images