Methods for predicting diabetes after acute pancreatitis based on a multi-modal lipid accumulation score

Abstract

The application discloses a method for predicting diabetes after acute pancreatitis based on a multi-modal lipid accumulation score, comprising obtaining baseline data of a patient and determining the data type, if abdominal plain CT image data is included, entering path A, if only conventional anthropometry and blood biochemical data are included, entering path B. Path A extracts visceral adipose tissue area (VAT), liver attenuation value, triglyceride (TG) and low-density lipoprotein (LDL); path B calculates fatty liver index (FLI) and lipid accumulation product (LAP) according to body mass index (BMI), waist circumference (WC), glutamyl transpeptidase (GGT), TG and total cholesterol (TC). After standardizing each characteristic variable, the imaging lipid accumulation score (LAS) or the estimated lipid accumulation score (e-LAS) is calculated according to the path calling principal component analysis (PCA) weighting formula, and then combined with the clinical basic risk factor to input the regression prediction model, and the PPDM-A occurrence probability and risk stratification report are output.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, specifically a method for predicting post-acute pancreatitis diabetes based on a multimodal lipid accumulation score. Background Technology

[0002] Among newly diagnosed diabetes in adults, pancreatic diabetes mellitus (DEP), which is secondary to pancreatic exocrine dysfunction, is the second most common type after type 2 diabetes. Among these, post-acute pancreatitis diabetes mellitus (PPDM-A) is the most prevalent subtype, accounting for about one-quarter of patients with acute pancreatitis (AP).

[0003] Traditional understanding of the pathogenesis of PPDM-A mainly focuses on inflammatory impact and its secondary damage. However, this view cannot explain the high incidence of PPDM-A in mild AP patients without pancreatic necrosis, which is as high as 14%. This indicates that the severity of AP is not entirely related to the risk of developing PPDM-A.

[0004] Recent clinical studies suggest that lipid metabolism abnormalities play a central role in this process. Overweight and obese patients have a significantly increased risk of developing PPDM-A, and patients with PPDM-A have a significantly increased incidence of visceral fat (VAT) and non-alcoholic fatty liver disease (NAFLD), while there is no significant difference in subcutaneous fat.

[0005] Systemic lipid metabolism disorders characterized by visceral fat expansion and ectopic fat deposition are key drivers of PPDM-A; however, there is currently a lack of tools in clinical practice to quantitatively assess complex lipid metabolism characteristics.

[0006] Furthermore, different levels of medical institutions face different technical conditions. In high-level medical settings, high-precision abdominal CT images can be used to construct accurate imaging lipid accumulation scores (LAS), while in primary care settings or long-term follow-up settings, only routine anthropometric measurements and blood biochemical indicators can be obtained to derive estimated lipid accumulation scores (e-LAS). Currently, there is a lack of a comprehensive solution that can simultaneously support both clinical scenarios for accurate prediction of PPDM-A. Summary of the Invention

[0007] The purpose of this invention is to provide a graded, dual-track lipid accumulation scoring framework for the early identification of high-risk patients with PPDM-A.

[0008] The specific technical solution of the present invention is as follows:

[0009] A method for predicting post-acute pancreatitis diabetes based on a multimodal lipid accumulation score includes the following steps:

[0010] S1: Data Acquisition and Judgment. Acquire baseline data of patients with acute pancreatitis and determine the current data type. If abdominal plain CT image data is included, proceed to Path A, which is for precise image and biochemical feature extraction. If only routine anthropometric data and blood biochemical data are included, proceed to Path B, which is for routine estimation of surrogate indicators extraction. Fasting blood sample indicators must be collected in both paths.

[0011] S2: Dual-path feature extraction, extracting corresponding multimodal feature data based on the determined path;

[0012] S3: Data standardization, which standardizes the feature variables extracted in S2 for each dimension and converts them into Z-scores;

[0013] S4: Calculate the comprehensive lipid accumulation score. The system calls the preset weighted formula derived from large-sample principal component analysis to calculate the corresponding lipid accumulation score.

[0014] S5: Joint assessment and risk output. The lipid accumulation score obtained in S4 is used as the core input feature. Combined with the patient's clinical baseline risk factors, it is input into a pre-trained regression prediction model to calculate the final probability of the patient developing diabetes after acute pancreatitis and generate a corresponding risk warning stratification report.

[0015] Preferably, in path A, S2 specifically includes: automatically segmenting abdominal CT images using a pre-trained deep learning AI model, extracting the visceral adipose tissue area from the second to the third lumbar vertebrae, performing three-dimensional liver segmentation to calculate the average liver attenuation value, and simultaneously obtaining the values ​​of triglycerides and low-density lipoprotein in the blood.

[0016] Preferably, in path B, S2 specifically includes: obtaining the patient's body mass index, waist circumference, gamma-glutamyl transferase, triglycerides, and total cholesterol, and using a predetermined clinical formula to calculate the fatty liver index and lipid accumulation product based on body mass index, waist circumference, gamma-glutamyl transferase, and triglycerides, respectively.

[0017] Preferably, S4 specifically includes: calculating the imaging lipid accumulation score according to the following formula:

[0018] The imaging lipid accumulation score = 0.541 × Z-score of visceral adipose tissue area − 0.630 × Z-score of liver attenuation value + 0.452 × Z-score of triglycerides + 0.324 × Z-score of low-density lipoprotein.

[0019] Preferably, S4 specifically includes: calculating the estimated lipid accumulation score according to the following formula:

[0020] Estimated lipid accumulation score = 0.203 × Z-score of total cholesterol + 0.686 × Z-score of fatty liver index + 0.699 × Z-score of lipid accumulation product.

[0021] Preferably, the pre-trained deep learning AI model is based on the nnU-Net framework; the three-dimensional liver segmentation includes calculating the average liver attenuation value for the liver region.

[0022] Preferably, the clinical baseline risk factors mentioned in S5 include a history of recurrent pancreatitis and baseline glycated hemoglobin levels.

[0023] The beneficial effects of this invention are as follows:

[0024] First, this invention breaks the limitations of a single data source and provides a solution that includes "Precise Image Standard Version (LAS)" and "Simplified Estimation Universal Version (e-LAS)". Whether it is a top-tier hospital with advanced radiology equipment or a community clinic conducting non-invasive screening, it can match the best prediction path based on existing data to achieve full-scenario coverage.

[0025] Secondly, this invention extracts core principal components through principal component analysis dimensionality reduction technology, eliminating the multicollinearity problem between various lipids and physical indicators. In particular, in the e-LAS model, the algorithm automatically assigns extremely high weights to the fatty liver index (FLI) and lipid accumulation product (LAP) while weakening the weight of total cholesterol (TC), successfully overcoming the prediction paradox interference of low total cholesterol accompanied by high morbidity caused by statins, and ensuring the prediction robustness in real-world medication environments.

[0026] Meanwhile, this invention successfully transforms the invisible state of metabolic disorder into an objective, one-dimensional risk numerical score (LAS or e-LAS), which greatly improves the clinical utility of early identification of high-risk patients for PPDM-A.

[0027] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0028] Figure 1 This is a flowchart of the steps of the present invention; Detailed Implementation

[0029] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] like Figure 1 As shown, this embodiment provides a method for predicting post-acute pancreatitis diabetes based on a multimodal lipid accumulation score, mainly including the following steps:

[0031] S1: Data Acquisition and Judgment

[0032] First, the baseline data of patients with acute pancreatitis is acquired. The baseline data includes abdominal plain CT image data, routine anthropometric data, and blood biochemical data of fasting blood samples. The system automatically determines the currently available data modality based on the acquired data types and selects the corresponding prediction path accordingly.

[0033] If the baseline data includes abdominal plain CT images, then path A is selected, which is the precise image and biochemical feature extraction path; if the baseline data only includes routine anthropometric data (BMI, waist circumference) and blood biochemistry data, then path B is selected, which is the routine estimation surrogate index extraction path.

[0034] Both pathways require collecting fasting blood samples from patients to obtain relevant blood biochemical indicators. The above decision-making logic can be adapted to the actual technical conditions of different levels of medical institutions, namely high-level medical scenarios and basic scenarios such as primary care or long-term follow-up. Specifically: in high-level medical scenarios, patients often have high-precision abdominal CT images, so pathway A can be used; in basic scenarios such as primary care or long-term follow-up, only routine anthropometric measurements and blood biochemical indicators can be obtained, so pathway B should be used.

[0035] S2: Dual-path feature extraction

[0036] Based on the path determined by S1, perform the following feature extraction operations respectively.

[0037] In path A, a pre-trained deep learning AI model is used to automatically segment abdominal CT images, extracting the visceral adipose tissue area (VAT, cm²) from the second lumbar vertebra (L2) to the third lumbar vertebra (L3); and performing three-dimensional liver segmentation to calculate the average liver attenuation (HU). Simultaneously, measured values ​​of triglycerides (TG) and low-density lipoprotein (LDL) are obtained from the patient's fasting blood biochemical test results. In this embodiment, the pre-trained deep learning AI model is based on the nnU-Net framework.

[0038] In pathway B, routine anthropometric data of the patient are obtained, including body mass index (BMI) and waist circumference (WC); simultaneously, gamma-glutamyl transferase (GGT), triglycerides (TG), and total cholesterol (TC) are obtained from the patient's fasting blood sample. Subsequently, the fatty liver index (FLI) and lipid accumulation product (LAP) are calculated using established clinical formulas based on BMI, WC, GGT, and TG, respectively.

[0039] S3: Data Standardization

[0040] The feature variables extracted from S2 are standardized and converted into Z-scores.

[0041] In path A, the four feature variables VAT, Liver Attenuation, TG and LDL are standardized by Z-score to obtain their respective Z scores, namely Z(VAT), Z(Liver Attenuation), Z(TG) and Z(LDL).

[0042] In path B, the three indicators TC, FLI and LAP are standardized by Z-score to obtain their respective Z scores, namely Z(TC), Z(FLI) and Z(LAP).

[0043] S4: Comprehensive Lipid Accumulation Score Calculation

[0044] Based on the path determined by S1, the system calls a preset weighted formula derived from large-sample principal component analysis (PCA) to calculate the comprehensive lipid accumulation score.

[0045] In path A, substitute the Z(VAT), Z(Liver Attenuation), Z(TG), and Z(LDL) obtained from S3 into the following formula to calculate the imaging lipid accumulation score (LAS):

[0046] .

[0047] In path B, substitute the Z(TC), Z(FLI), and Z(LAP) obtained from S3 into the following formula to calculate the estimated lipid accumulation score (e-LAS):

[0048] .

[0049] In the weighted formula for path B, fat indicators (FLI, LAP) are given a high weight, while the weight of TC is weakened.

[0050] Based on the above weighting, when patients take statins and their TC levels decrease, since TC has a relatively small weight in the e-LAS, while FLI and LAP, as the main contributing factors, can still stably reflect the patient's lipid accumulation status, thus ensuring that the predictive efficacy of the e-LAS score for PPDMA-A risk is not affected by statins, and guaranteeing the predictive robustness in real-world medication settings.

[0051] S5: Joint Assessment and Risk Output

[0052] The LAS or e-LAS score calculated by S4 is used as the core input feature. Combined with the patient's basic clinical risk factors (such as history of recurrent pancreatitis and baseline glycated hemoglobin (HbA1c) value), they are input into a pre-trained regression prediction model (nested logistic regression model) to calculate the final probability of the patient developing post-pancreatitis diabetes mellitus (PPDM-A). After calculating the probability of the patient developing PPDM-A through the regression prediction model, a corresponding risk warning stratification report is generated according to the preset risk stratification threshold. The risk warning stratification report includes three levels: low risk, medium risk, and high risk.

[0053] The embodiments described above are merely illustrative of implementation methods of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A method for predicting post-acute pancreatitis diabetes based on a multimodal lipid accumulation score, characterized in that, Includes the following steps: S1: Data Acquisition and Judgment. Acquire baseline data of patients with acute pancreatitis and determine the current data type. If abdominal plain CT image data is included, proceed to path A, which is for precise image and biochemical feature extraction. If only routine anthropometric data and blood biochemistry data are included, proceed to Path B, which is for extracting routine estimated alternative indicators; both paths require the collection of fasting blood sample indicators. S2: Dual-path feature extraction, extracting corresponding multimodal feature data based on the determined path; S3: Data standardization, which standardizes the feature variables extracted in S2 for each dimension and converts them into Z-scores; S4: Calculate the comprehensive lipid accumulation score. The system calls the preset weighted formula derived from large-sample principal component analysis to calculate the corresponding lipid accumulation score. S5: Joint assessment and risk output. The lipid accumulation score obtained in S4 is used as the core input feature. Combined with the patient's clinical baseline risk factors, it is input into a pre-trained regression prediction model to calculate the final probability of the patient developing diabetes after acute pancreatitis and generate a corresponding risk warning stratification report.

2. The method for predicting post-acute pancreatitis diabetes based on a multimodal lipid accumulation score according to claim 1, characterized in that, In path A, S2 specifically includes: using a pre-trained deep learning AI model to automatically segment abdominal CT images, extracting the area of ​​visceral adipose tissue from the second to the third lumbar vertebra, performing three-dimensional liver segmentation to calculate the average liver attenuation value, and simultaneously obtaining the values ​​of triglycerides and low-density lipoprotein in the blood.

3. The method for predicting post-acute pancreatitis diabetes based on a multimodal lipid accumulation score according to claim 1, characterized in that, In path B, S2 specifically includes: obtaining the patient's body mass index, waist circumference, gamma-glutamyl transferase, triglycerides, and total cholesterol; and using established clinical formulas to calculate the fatty liver index and lipid accumulation product based on body mass index, waist circumference, gamma-glutamyl transferase, and triglycerides, respectively.

4. The method for predicting post-acute pancreatitis diabetes based on a multimodal lipid accumulation score according to claim 2, characterized in that, S4 specifically includes: calculating the imaging lipid accumulation score using the following formula: The imaging lipid accumulation score = 0.541 × Z-score of visceral adipose tissue area − 0.630 × Z-score of liver attenuation value + 0.452 × Z-score of triglycerides + 0.324 × Z-score of low-density lipoprotein.

5. The method for predicting post-acute pancreatitis diabetes based on a multimodal lipid accumulation score according to claim 3, characterized in that, S4 specifically includes: calculating the estimated lipid accumulation score using the following formula: Estimated lipid accumulation score = 0.203 × Z-score of total cholesterol + 0.686 × Z-score of fatty liver index + 0.699 × Z-score of lipid accumulation product.

6. The method for predicting post-acute pancreatitis diabetes based on a multimodal lipid accumulation score according to claim 4, characterized in that, The pre-trained deep learning AI model is based on the nnU-Net framework; the three-dimensional liver segmentation includes calculating the average liver attenuation value for the liver region.

7. The method for predicting post-acute pancreatitis diabetes based on a multimodal lipid accumulation score according to claim 1, characterized in that, The clinical baseline risk factors described in S5 include a history of recurrent pancreatitis and baseline glycated hemoglobin levels.

Images