Pancreatic disease early warning system based on artificial intelligence

Abstract

The invention discloses a pancreatic disease early warning system based on artificial intelligence, and relates to the technical field of artificial intelligence. According to the method, an image confidence score is generated by pre-training a basic model, an uncertain area of the model is identified, a gradient magnitude mean value and a texture entropy are combined, an image priority index is generated, a complex lesion area is accurately positioned, and clinical data of a patient, including personal information of the patient, self-symptom expression description of the patient and an examination index, are collected. According to the method and the system, a high-risk early-warning queue and a medium-risk early-warning queue are constructed, early-warning hidden danger degree classification of different patients is realized, cross-modal correlation analysis of image features and clinical data is realized, the problem of patients who are prone to missed diagnosis and have atypical image performance but abnormal clinical indexes in the prior art is solved, and the missed diagnosis rate can be effectively reduced.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to an early warning system for pancreatic diseases based on artificial intelligence. Background Art

[0002] Pancreatic diseases, especially pancreatic cancer, have hidden early symptoms, rapid disease progression, and extremely high mortality rate. Early and accurate diagnosis is crucial to improving patient prognosis.

[0003] However, the pancreatic disease early warning system in the existing technology still has the following deficiencies: Existing technologies lack cross-modal correlation analysis of imaging features, clinical symptoms, and laboratory indicators, resulting in a high risk of missed diagnosis of cases with atypical imaging but high clinical risk. In addition, during the analysis of clinical data for early warning of pancreatic diseases, it is impossible to integrate the patient's demographic characteristics, symptoms, and test index performance, making it difficult to identify early signals from subclinical states to overt lesions, resulting in delayed warning.

[0004] To this end, an artificial intelligence-based early warning system for pancreatic diseases was launched. Summary of the Invention

[0005] In view of this, the present invention provides an artificial intelligence-based early warning system for pancreatic diseases to solve the problems raised by the above-mentioned background technology.

[0006] The purpose of the present invention can be achieved through the following technical solutions: an early warning system for pancreatic diseases based on artificial intelligence, comprising: Intelligent labeling module: extracts unlabeled image data from the image library, uses pre-trained basic models, generates preliminary prediction results, and calculates voxel-level confidence scores Where p is the predicted value of the lesion probability of each image data; the regions with confidence scores below the threshold are identified and classified as uncertainty regions. The image priority index R corresponding to the uncertainty region is obtained by combining the image complexity index; the image complexity index includes the mean gradient amplitude and texture entropy; Clinical assessment module: Identify the clinical data of patients corresponding to the uncertainty area and conduct a comprehensive assessment to derive the clinical risk index U of the patients corresponding to the uncertainty area; the clinical data includes patient personal information, patient self-description of symptoms, and examination indicators; Cohort construction module: Receives the image priority index R and clinical risk index U of the uncertainty area, performs a comprehensive assessment, and divides the high-risk warning cohort and the medium-risk warning cohort based on the assessment results; Expert annotation module: receives high-risk warning queues and medium-risk warning queues, and sends them to the corresponding expert mailboxes based on the set task allocation logic.

[0007] In some embodiments, the image complexity index is combined to obtain the image priority index R corresponding to the uncertainty area, specifically: Calculate the mean three-dimensional gradient amplitude G of the voxels in the uncertainty region; that is, by the formula Calculated; where N represents the total number of voxels in the uncertainty region selected when calculating the mean gradient amplitude; I represents the image data; Represent the partial derivatives of the image data I in the x, y, and z directions respectively; The texture entropy E of the uncertainty region is calculated by the gray level co-occurrence matrix; that is, by the formula ;in Represents the probability of a pixel with gray value i and a pixel with gray value j appearing in a specific spatial relationship in the gray level co-occurrence matrix; Extract the confidence score C, gradient amplitude mean G and texture entropy E of the uncertainty area, normalize them and substitute them into the formula A weighted calculation is performed to determine the image priority index R of the uncertainty area.

[0008] In some embodiments, the identifying of the patient's personal information corresponding to the uncertainty region and performing a comprehensive evaluation may include: The patient's age group and gender are extracted from the personal information of the patient corresponding to the uncertain area, and each group of age group intervals is pre-constructed, and each group of age group intervals corresponds to an age-induced score; the extracted age group of the patient is matched with the corresponding age group interval and converted into the patient's age-induced score; A gender additional coefficient is set for each gender; the patient's age-induced score is multiplied by the corresponding gender additional coefficient to be used as the demographic characteristic score of the patient corresponding to the uncertain area, recorded as N1.

[0009] In some embodiments, the identification of the uncertainty region corresponds to a patient's self-described symptom and a comprehensive assessment, specifically: For the patient's self-symptom description corresponding to the uncertain area, text preprocessing is performed; after the preprocessing is completed, the patient's self-symptom description is input into a pre-constructed pancreatic disease symptom vocabulary for matching, the total number of matched symptom words is counted, and the total number of symptom words is multiplied by the integer 2 as the disease characteristic score of the patient corresponding to the uncertain area, recorded as N2.

[0010] In some embodiments, the identification of the patient's examination indicators corresponding to the uncertainty region and the comprehensive evaluation are specifically as follows: Extracting the patient's examination indicators from the patient's personal information corresponding to the uncertain area, extracting the normal indicator range corresponding to each examination indicator, and matching each patient's examination indicator with the corresponding normal indicator range. If a group of examination indicators is outside the normal indicator range, it is determined to be an abnormal indicator; Preset the key indicators corresponding to pancreatic diseases, match abnormal indicators with key indicators, and count the number of successful matches as the number of induced indicators; Calculate the degree of deviation of the induced index quantity from the corresponding normal index range, that is, by identifying whether the induced index is higher than the normal index range or lower than the normal index range. If the induced index is higher than the normal index range, extract the highest value of the normal index range and calculate the difference between the highest value and the induced index value. Divide the calculated difference by the highest value to obtain the deviation rate; If the induced index is lower than the normal index range, the lowest value of the normal index range is extracted and the difference between it and the induced index value is calculated. The calculated difference is divided by the lowest value to obtain the deviation rate; The deviation rate of each induced indicator is calculated by comparing it with the corresponding set deviation rate threshold, so as to determine the deviation risk value of each induced indicator corresponding to the patient; The number of induced indicators and the deviation risk value corresponding to the patient are marked as fw and ftc respectively; where c represents the number of each induced indicator, c=1,2,...,v, and v is the total number of induced indicators of the patient.

[0011] In some embodiments, the clinical risk index U of the patient corresponding to the uncertainty region is obtained as follows: After normalizing the number of induced indicators fw and the deviation hidden danger value ftc corresponding to the patient, substitute them into the formula Perform weighted calculation to obtain the indicator abnormal value fr; are the number of induced indicators fw and the weight coefficient of each induced indicator deviation from the hidden danger value ftc; the calculated indicator abnormal value fr is rounded off to an integer and used as the indicator characteristic score N3; The demographic characteristic score N1, symptom characteristic score N2, and indicator characteristic score N3 of the patients corresponding to the uncertainty area are normalized, multiplied by the corresponding set weight coefficients, and then summed to obtain the clinical hidden danger index U of the patients corresponding to the uncertainty area.

[0012] In some embodiments, the high-risk warning queue and the medium-risk warning queue are divided based on the evaluation results, specifically: The imaging priority index R and clinical hidden danger index U of each uncertainty area are normalized and multiplied by the corresponding set weight coefficients to determine the early warning review index of each patient; the threshold index range corresponding to the early warning review index is set. If the early warning review index of a patient is higher than the threshold index range, the patient is classified into the high-risk early warning queue; if the early warning review index of a patient is within the threshold index range, the patient is classified into the medium-risk early warning queue.

[0013] In some embodiments, the task allocation logic based on the settings will be sent to the corresponding expert mailbox, specifically: For high-risk warning queues, a read signal is sent to each expert. After each expert confirms the read signal, the number and type of queues to be processed in each expert's mailbox at the current time are obtained. If a high-risk warning queue does not exist in the queue type to be processed by an expert, the current high-risk warning queue is directly sent to the mailbox of the expert who does not have a high-risk warning queue, and the expert is reminded to process it as the first task. If all experts have high-risk warning queues in their mailboxes at the current time, the high-risk warning queue will be sent to the mailbox of the expert with a larger review recommendation index and will be arranged after the existing high-risk warning queue; for the medium-risk warning queue, it will be randomly sent to the mailbox of any expert and marked according to the deadline for review of the medium-risk warning queue.

[0014] In some embodiments, the specific process of obtaining the expert review recommendation index is as follows: The number of high-level warning queues for each expert is counted and recorded as the number of urgent tasks. The time required for each expert to process a single piece of data in the high-level warning queue is calculated. That is, the time taken by each expert to open and process a single patient data set M times before the current time point is extracted, and the average time taken for each group is calculated to obtain the time required for each expert to process a single patient data in the high-level warning queue. The total number of patient data in each expert's emergency task count is counted and multiplied by the corresponding processing time to obtain the estimated waiting time for each expert to process the current high-level warning queue. At the same time, the length of medical practice of each expert is also obtained. The estimated waiting time and medical practice time of each expert are normalized respectively, and then multiplied with the corresponding set weight coefficient to obtain the processing waiting weighted value and the medical practice time weighted value. The ratio is calculated with the medical practice time weighted value as the numerator and the processing waiting weighted value as the denominator to obtain the review recommendation index of each expert at the current time point.

[0015] Compared with the prior art, the present invention has the following beneficial effects: The present invention generates image confidence scores through pre-training basic models, identifies model uncertainty areas, and combines gradient amplitude mean and texture entropy to generate image priority indexes to accurately locate complex lesion areas. At the same time, it collects patients' clinical data, including personal information, self-description of symptoms, and examination indicators, and conducts a comprehensive evaluation, thereby constructing high-risk and medium-risk warning queues, classifying the degree of warning risks of different patients, and realizing cross-modal correlation analysis between image features and clinical data. This solves the problem that existing technologies easily miss diagnoses patients with atypical imaging manifestations but abnormal clinical indicators, and can effectively reduce the missed diagnosis rate. The present invention uses stratified quantification of clinical data, including demographic characteristic scores, symptom characteristic scores, and indicator characteristic scores, to reflect the patient's symptom status from multiple aspects of demographic characteristics, symptom characteristics, and indicator characteristics. This is converted into a calculable risk score through a pre-set calculation conversion logic, providing quantitative data support for early warning of pancreatic diseases. The present invention prioritizes the allocation of high-risk warning queues to experts who currently have no high-risk tasks, or dynamically allocates them according to the review recommendation index, ensuring that emergency cases are handled in a timely manner and improving the efficiency and accuracy of warning processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Further details, features and advantages of the present application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which: Figure 1 This is a principle block diagram of the present invention. DETAILED DESCRIPTION

[0017] Several embodiments of the present application will be described in more detail below with reference to the accompanying drawings so that those skilled in the art can implement the present application. The present application can be embodied in many different forms and for many different purposes and should not be limited to the embodiments described herein. These embodiments are provided to make the present application comprehensive and complete and to fully convey the scope of the present application to those skilled in the art. The embodiments do not limit the present application.

[0018] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It will be further understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the relevant art and / or the context of this specification, and will not be interpreted in an idealized or overly formal sense unless expressly defined as such herein.

[0019] Example

[0020] See also Figure 1As shown, the AI-based early warning system for pancreatic diseases includes an intelligent annotation module, a clinical evaluation module, a cohort construction module, and an expert annotation module; The intelligent labeling module is used to extract unlabeled image data in the image library, use pre-trained basic models such as 3DU-Net to generate preliminary prediction results, and calculate voxel-level confidence scores. Where p is the predicted value of the lesion probability of each image data; the regions with confidence scores below the threshold are identified and classified as uncertainty regions. The image priority index R corresponding to the uncertainty region is obtained by combining the image complexity index; the image complexity index includes the mean gradient amplitude and texture entropy; Specifically: Calculate the mean value G of the three-dimensional gradient amplitude of the voxels in the uncertainty region; reflect the intensity of the grayscale change; that is, through the formula Calculated; N represents the total number of voxels within the uncertainty region selected when calculating the mean gradient amplitude. When calculating G, it is necessary to traverse every voxel within the region. N determines the scale of the voxels involved in the calculation, thus affecting the accuracy and representativeness of the final mean. I represents the image data; it is a three-dimensional information matrix, in which each element corresponds to a voxel in the image and stores information such as the grayscale value of the voxel. In pancreatic lesion images, the distribution of I values ​​in different tissues and lesion areas has its own characteristics. Represents the partial derivatives of the image data I in the x, y, and z directions respectively; used to measure the rate of change of the image grayscale in each direction. By calculating the square root of the sum of the squares of the partial derivatives in the three directions, it can comprehensively reflect the grayscale change of the voxel in the three-dimensional space, and thus reflect the intensity of the grayscale change in the region; The texture entropy E of the uncertainty region is calculated by the gray-level co-occurrence matrix; the high entropy value corresponds to the heterogeneous lesion; that is, through the formula ;in Represents the probability of a pixel with gray value i and a pixel with gray value j in the gray level co-occurrence matrix appearing in a specific spatial relationship; a specific spatial relationship such as a set distance and direction; Gray-level co-occurrence matrix is ​​a statistical method used to describe the texture characteristics of an image. It constructs a matrix by calculating the frequency of occurrence of different gray-level pixel pairs in an image. It is the probability value obtained after the elements in the matrix are normalized, reflecting the spatial distribution relationship between different gray levels. Take the logarithm and weighted sum (the weighting coefficient is The larger the E value is, the higher the disorder of the image texture is. Extract the confidence score C, gradient amplitude mean G and texture entropy E of the uncertainty area, normalize them and substitute them into the formula Perform weighted calculation to determine the image priority index R of the uncertainty area; In addition, the image complexity index and confidence score are integrated into the priority index R, which provides data support for the subsequent formation of the annotation queue, avoiding the inefficiency of the traditional method of applying equal effort to all areas, resulting in low early warning efficiency; The clinical assessment module is used to identify the clinical data of patients corresponding to the uncertainty area and conduct a comprehensive assessment to derive the clinical risk index U of the patients corresponding to the uncertainty area. The clinical data includes patient personal information, patient self-description of symptoms, and examination indicators. Specifically: The patient's age group and gender are extracted from the personal information of the patient corresponding to the uncertain area, and each age group interval is pre-constructed, and each age group interval corresponds to an age-induced score; the age-induced score range is set to 1-10, and the older the patient, the higher the corresponding age-induced score. Pancreatic diseases are more common in the elderly population. After matching the patient's extracted age group with the corresponding age group interval, it is converted into the patient's age-induced score; A gender additional coefficient is set for each gender; the gender additional coefficient range is set between 1.092-1.138. The gender additional coefficient is set according to the male-female incidence rate of pancreatic diseases, and the gender additional coefficient corresponding to males is greater than that corresponding to females. The patient's age-induced score is multiplied by the corresponding gender additional coefficient to obtain the demographic characteristic score of the patient corresponding to the uncertain area, which is recorded as N1; It is supplemented that, under the same imaging features, the detection rate of malignant lesions in patients with high N1 scores is higher than that in patients with low scores; For the patient's self-described symptom manifestations corresponding to the uncertain area, text preprocessing was performed, including filtering irrelevant content, Chinese word segmentation, and standardization. After preprocessing, the patient's self-described symptom manifestations were input into a pre-built pancreatic disease symptom word library for matching. The total number of matched symptom words was counted, and the result of multiplying the total number of symptom words by the integer 2 was used as the disease characteristic score of the patient corresponding to the uncertain area, recorded as N2. Extracting the patient's examination indicators from the personal information of the patient corresponding to the uncertain area, including liver function indicators, blood sugar, glycosylated hemoglobin, and blood lipids; Extract the normal index range corresponding to each examination index, and match each examination index of the patient with the corresponding normal index range. If a group of examination indicators is outside the normal index range, it is determined to be an abnormal indicator; Preset key indicators corresponding to pancreatic diseases, such as serum amylase, lipase, CA19-9, blood sugar, liver function indicators, etc.; match abnormal indicators with key indicators, and count the number of successful matches as the number of induced indicators; Calculate the degree of deviation of the induced index quantity from the corresponding normal index range, that is, by identifying whether the induced index is higher than the normal index range or lower than the normal index range. If the induced index is higher than the normal index range, extract the highest value of the normal index range and calculate the difference between the highest value and the induced index value. Divide the calculated difference by the highest value to obtain the deviation rate; If the induced index is lower than the normal index range, the lowest value of the normal index range is extracted and the difference between it and the induced index value is calculated. The calculated difference is divided by the lowest value to obtain the deviation rate; Supplementary note: The above difference calculation takes the absolute value of the calculated difference by default; The deviation rate of each induced indicator is calculated by comparing it with the corresponding set deviation rate threshold, so as to determine the deviation risk value of each induced indicator corresponding to the patient; The number of induced indicators and the deviation risk value corresponding to the patient are marked as fw and ftc respectively; where c represents the number of each induced indicator, c = 1, 2, ..., v, and v is the total number of induced indicators of the patient; After normalizing the number of induced indicators fw and the deviation hidden danger value ftc corresponding to the patient, substitute them into the formula Perform weighted calculation to obtain the indicator abnormal value fr; are the number of induced indicators fw and the weight coefficient of each induced indicator deviation from the hidden danger value ftc; the calculated indicator abnormal value fr is rounded off to an integer and used as the indicator characteristic score N3; For the demographic characteristic score N1, symptom characteristic score N2, and indicator characteristic score N3 of the patients corresponding to the uncertainty area, after normalization processing, they are multiplied by the corresponding set weight coefficients respectively, and then the sum is obtained to obtain the clinical risk index U of the patients corresponding to the uncertainty area; To supplement, the above content reflects the patient's condition from multiple aspects, including demographic characteristics, symptom characteristics, and indicator characteristics. This is converted into a calculable risk score through a pre-defined calculation conversion logic, providing quantitative data support for early warning of pancreatic diseases. The cohort construction module is used to receive the image priority index R and clinical hidden danger index U of the uncertainty area, conduct a comprehensive evaluation, and divide the high-risk warning cohort and the medium-risk warning cohort based on the evaluation results; Specifically: The image priority index R and clinical hidden danger index U of each uncertainty area are normalized and multiplied by the corresponding set weight coefficient to determine the early warning review index of each patient; Set the threshold index range corresponding to the early warning review index. If a patient's early warning review index is higher than the threshold index range, the patient is classified into the high-risk early warning queue; if a patient's early warning review index is within the threshold index range, the patient is classified into the medium-risk early warning queue.

[0021] The expert annotation module is used to receive high-risk warning queues and medium-risk warning queues, and send them to the corresponding expert mailboxes based on the set task allocation logic; Specifically: For high-risk warning queues, a read signaling command is sent to each expert. After each expert confirms the read signaling command, the number and type of queues to be processed in each expert's mailbox at the current time point are obtained; If there is no high-risk warning queue in the queue type to be processed by a certain expert, the current high-risk warning queue will be directly sent to the email address of the expert who does not have a high-risk warning queue, and the expert will be reminded to process it as the first task; In addition, if the number of experts not in the high-risk warning queue is greater than one, they will be randomly selected; If all experts have high-risk warning queues in their mailboxes at the current time point, the number of high-level warning queues for each expert is counted and recorded as the number of urgent tasks; the time required for each expert to process a single piece of data in the high-level warning queue is calculated, that is, by extracting the time taken by each expert to open and complete the processing of a single patient data set M times before the current time point, and calculating the average time taken for each group, the time required for each expert to process a single patient data in the high-level warning queue is obtained; where M>10, and the specific value is set by the technical staff; The total number of patient data in each expert's emergency task count is counted and multiplied by the corresponding processing time to obtain the estimated waiting time for each expert to process the current high-level warning queue. At the same time, the length of medical practice of each expert is also obtained. The estimated waiting time and medical practice time of each expert are normalized and multiplied by the corresponding weight coefficient to obtain the weighted value of waiting time and the weighted value of medical practice time. The weighted value of medical practice time is used as the numerator and the weighted value of waiting time is used as the denominator to calculate the ratio, thereby obtaining the review recommendation index of each expert at the current time point; Send the high-level warning queue to the mailbox of the expert with the highest review recommendation index, and arrange it after the existing high-level warning queue; For the medium-risk warning queue, it will be randomly sent to any expert's mailbox and marked according to the deadline for review of the medium-risk warning queue; for example, the medium-risk warning queue can be limited to one week for review completion; The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to specific embodiments. Obviously, many modifications and variations are possible based on the contents of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. An artificial intelligence-based early warning system for pancreatic diseases, characterized by: include: Intelligent labeling module: extracts unlabeled image data from the image library, uses pre-trained basic models, generates preliminary prediction results, and calculates voxel-level confidence scores Where p is the predicted value of the lesion probability of each image data; the regions with confidence scores below the threshold are identified and classified as uncertainty regions. The image priority index R corresponding to the uncertainty region is obtained by combining the image complexity index; the image complexity index includes the mean gradient amplitude and texture entropy; Clinical assessment module: Identify the clinical data of patients corresponding to the uncertainty area and conduct a comprehensive assessment to derive the clinical risk index U of the patients corresponding to the uncertainty area; the clinical data includes patient personal information, patient self-description of symptoms, and examination indicators; Cohort construction module: Receives the image priority index R and clinical risk index U of the uncertainty area, performs a comprehensive assessment, and divides the high-risk warning cohort and the medium-risk warning cohort based on the assessment results; Expert annotation module: receives high-risk warning queues and medium-risk warning queues, and sends them to the corresponding expert mailboxes based on the set task allocation logic.

2. The artificial intelligence-based early warning system for pancreatic diseases according to claim 1, characterized in that: The image complexity index is combined to obtain the image priority index R corresponding to the uncertainty area, specifically: Calculate the mean three-dimensional gradient amplitude G of the voxels in the uncertainty region; that is, by the formula Calculated; where N represents the total number of voxels in the uncertainty region selected when calculating the mean gradient amplitude; I represents the image data; Represent the partial derivatives of the image data I in the x, y, and z directions respectively; The texture entropy E of the uncertainty region is calculated by the gray level co-occurrence matrix; that is, by the formula ;in Represents the probability of a pixel with gray value i and a pixel with gray value j appearing in a specific spatial relationship in the gray level co-occurrence matrix; Extract the confidence score C, gradient amplitude mean G and texture entropy E of the uncertainty area, normalize them and substitute them into the formula A weighted calculation is performed to determine the image priority index R of the uncertainty area.

3. The artificial intelligence-based early warning system for pancreatic diseases according to claim 2, characterized in that: The personal information of the patient corresponding to the uncertainty area is identified and a comprehensive assessment is performed, specifically: The patient's age group and gender are extracted from the personal information of the patient corresponding to the uncertain area, and each group of age group intervals is pre-constructed, and each group of age group intervals corresponds to an age-induced score; the extracted age group of the patient is matched with the corresponding age group interval and converted into the patient's age-induced score; A gender additional coefficient is set for each gender; the patient's age-induced score is multiplied by the corresponding gender additional coefficient to be used as the demographic characteristic score of the patient corresponding to the uncertain area, recorded as N1.

4. The artificial intelligence-based early warning system for pancreatic disease according to claim 3, characterized in that: The patient's self-symptom description of the patient corresponding to the identified uncertainty area is comprehensively evaluated, specifically: For the patient's self-symptom description corresponding to the uncertain area, text preprocessing is performed; after the preprocessing is completed, the patient's self-symptom description is input into a pre-constructed pancreatic disease symptom vocabulary for matching, the total number of matched symptom words is counted, and the total number of symptom words is multiplied by the integer 2 as the disease characteristic score of the patient corresponding to the uncertain area, recorded as N2.

5. The artificial intelligence-based early warning system for pancreatic diseases according to claim 4, characterized in that: The examination indicators of the patients corresponding to the identified uncertainty areas are comprehensively evaluated, specifically: Extracting the patient's examination indicators from the patient's personal information corresponding to the uncertain area, extracting the normal indicator range corresponding to each examination indicator, and matching each patient's examination indicator with the corresponding normal indicator range. If a group of examination indicators is outside the normal indicator range, it is determined to be an abnormal indicator; Preset the key indicators corresponding to pancreatic diseases, match abnormal indicators with key indicators, and count the number of successful matches as the number of induced indicators; Calculate the degree of deviation of the induced index quantity from the corresponding normal index range, that is, by identifying whether the induced index is higher than the normal index range or lower than the normal index range. If the induced index is higher than the normal index range, extract the highest value of the normal index range and calculate the difference between the highest value and the induced index value. Divide the calculated difference by the highest value to obtain the deviation rate; If the induced index is lower than the normal index range, the lowest value of the normal index range is extracted and the difference between it and the induced index value is calculated. The calculated difference is divided by the lowest value to obtain the deviation rate; The deviation rate of each induced indicator is calculated by comparing it with the corresponding set deviation rate threshold, so as to determine the deviation risk value of each induced indicator corresponding to the patient; The number of induced indicators and the deviation risk value corresponding to the patient are marked as fw and ftc respectively; where c represents the number of each induced indicator, c=1,2,...,v, and v is the total number of induced indicators of the patient.

6. The artificial intelligence-based early warning system for pancreatic diseases according to claim 5, characterized in that: The clinical risk index U of the patient corresponding to the uncertainty area is specifically: After normalizing the number of induced indicators fw and the deviation hidden danger value ftc corresponding to the patient, substitute them into the formula Perform weighted calculation to obtain the indicator abnormal value fr; are the number of induced indicators fw and the weight coefficient of each induced indicator deviation from the hidden danger value ftc; the calculated indicator abnormal value fr is rounded off to an integer and used as the indicator characteristic score N3; The demographic characteristic score N1, symptom characteristic score N2, and indicator characteristic score N3 of the patients corresponding to the uncertainty area are normalized, multiplied by the corresponding set weight coefficients, and then summed to obtain the clinical hidden danger index U of the patients corresponding to the uncertainty area.

7. The artificial intelligence-based early warning system for pancreatic diseases according to claim 6, characterized in that: The high-risk warning cohort and the medium-risk warning cohort are divided based on the evaluation results, specifically: The imaging priority index R and clinical hidden danger index U of each uncertainty area are normalized and multiplied by the corresponding set weight coefficients to determine the early warning review index of each patient; the threshold index range corresponding to the early warning review index is set. If the early warning review index of a patient is higher than the threshold index range, the patient is classified into the high-risk early warning queue; if the early warning review index of a patient is within the threshold index range, the patient is classified into the medium-risk early warning queue.

8. The artificial intelligence-based early warning system for pancreatic diseases according to claim 7, characterized in that: The task allocation logic based on the settings will be sent to the corresponding expert mailbox, specifically: For high-risk warning queues, a read signaling command is sent to each expert. After each expert confirms the read signaling command, the number and type of queues to be processed in each expert's mailbox at the current time point are obtained; If there is no high-risk warning queue in the queue type to be processed by a certain expert, the current high-risk warning queue will be directly sent to the email address of the expert who does not have a high-risk warning queue, and the expert will be reminded to process it as the first task; If all experts have high-risk warning queues in their mailboxes at the current time, the high-risk warning queue will be sent to the mailbox of the expert with the higher review recommendation index, and will be placed after the one with the existing high-risk warning queue; For the medium-risk warning queue, it will be randomly sent to the mailbox of any expert and marked according to the deadline for review of the medium-risk warning queue.

9. The artificial intelligence-based early warning system for pancreatic diseases according to claim 8, characterized in that: The specific process of obtaining the expert review and recommendation index is as follows: The number of high-level warning queues for each expert is counted and recorded as the number of urgent tasks. The time required for each expert to process a single piece of data in the high-level warning queue is calculated. That is, the time taken by each expert to open and process a single patient data set M times before the current time point is extracted, and the average time taken for each group is calculated to obtain the time required for each expert to process a single patient data in the high-level warning queue. The total number of patient data in each expert's emergency task count is counted and multiplied by the corresponding processing time to obtain the estimated waiting time for each expert to process the current high-level warning queue. At the same time, the length of medical practice of each expert is also obtained. The estimated waiting time and medical practice time of each expert are normalized respectively, and then multiplied with the corresponding set weight coefficient to obtain the processing waiting weighted value and the medical practice time weighted value. The ratio is calculated with the medical practice time weighted value as the numerator and the processing waiting weighted value as the denominator to obtain the review recommendation index of each expert at the current time point.

Images