CN-122023923-A - Machine learning method-based acute necrotizing pancreatitis severe prediction model construction method

Abstract

The invention discloses a machine learning method-based acute necrotic pancreatitis severe prediction model construction method which comprises the following steps of S1, collecting patient data based on inclusion standards and exclusion standards, S2, preprocessing the data and selecting the characteristics, S3, obtaining an optimal differential diagnosis prediction model by combining radiological characteristics and a plurality of machine learning algorithms, S4, obtaining evaluation indexes through prediction performance comparison, and S5, constructing a pancreas model, a pancreas surrounding model and a combination model, and producing ROC curve interpretation results. The present invention develops and validates a machine learning model that distinguishes heavy and moderate ANP (i.e., ANSP and ANMSP). These models will be based on radiological features extracted from CECT images of the parenchymal portal vein of the pancreas, peripancreatic necrosis, and combinations thereof. By evaluating the diagnostic performance of these models, early identification of disease severity is achieved and support is provided for treatment-related clinical decisions.

Inventors

• XIAO BO
• FENG YUE
• HU XIHAO
• JIANG ZHIQIONG

Assignees

• 重庆市璧山区人民医院

Dates

Publication Date

20260512

Application Date

20260203

Claims (5)

1. 1. The method for constructing the acute necrotizing pancreatitis severe prediction model based on the machine learning method is characterized by comprising the following steps of: Step S1, collecting patient data for later use based on inclusion criteria and exclusion criteria; step S2, preprocessing the patient data in the step 1 and selecting the characteristics to obtain preprocessed patient data for later use; The method comprises the steps of performing feature extraction analysis on a portal vein CECT image by adopting 3D-Slicer software, manually delineating a region of interest (ROI) to comprise two independent components, wherein the first part is the whole pancreas parenchyma and comprises a necrosis region but does not comprise bile ducts and blood vessels, the second part is peripancreatic necrosis accumulation corresponding to an anatomical layer, automatically extracting radiological features of a volume of interest by using a PyRadiomics Python software package, discretizing the radiological features into 10 intervals by using an equal-width box dividing method and an equal-frequency box dividing method, randomly rearranging the discretized features based on LightGBM algorithm, and deducing an optimal feature subset by 10-fold cross-validation iteration, wherein the optimal feature subset is preprocessed patient data, and constructing a classification model by adopting optimal radiological features and a 10-fold cross-validation frame; Step S3, acquiring a differential diagnosis prediction model by combining the radiological features of the patient data preprocessed in the step 2 and a forest learning algorithm, and respectively constructing pancreas, pancreas circumference and combined image histology prediction models by adopting the forest learning algorithm and combining 10-fold cross validation based on the pancreas parenchyma, pancreas circumference necrosis and the optimal image histology features jointly screened by the two areas; s4, obtaining an evaluation index by comparing the prediction performance of the differential diagnosis prediction model in the step 3; And S5, constructing a pancreas model, a pancreas surrounding model and a combined model, and producing ROC curve interpretation results.
2. 2. The machine learning method-based acute necrotizing pancreatitis severe prediction model construction method according to claim 1, wherein the inclusion criteria in the step S1 include (1) hospitalization and CECT examination within 7 days of ANP incidence, (2) complete laboratory data, medical record and imaging examination data, (3) age not less than 18 years, exclusion criteria include (1) hospitalization 7 days after ANP incidence, (2) hospitalization within 7 days of ANP incidence but not CECT examination, (3) history of chronic pancreatitis, pancreatic malignancy or pancreatic surgery, (4) gestation period or age less than 18 years, (5) incomplete and no portal vein phase axial image, poor image quality leading to impossibility, absence of critical clinical data, hospitalization due to other acute abdominal diseases, severe group features persistent organ dysfunction >48 hours, improved Marshall score not less than 2 score, and meets any two criteria including 24 score not less than 3 score, glasw-Imrie score not less than 3 score or less than 48 score and transient score not less than 48 score for severe organ disorder not less than 48 degrees.
3. 3. The method for constructing the machine learning method-based acute necrotizing pancreatitis severe prediction model according to claim 1, wherein the specific operation of the step S3 is that three prediction models including a pancreatic model, a peripancreatic model and a joint model are finally established by analyzing the portal CECT data by applying a forest learning algorithm; random forest RF, mathematical principle, integrating multiple decision trees: The modeling steps are that ① Bootstrap samples are used for creating 200 decision trees; ② Selecting optimal characteristics and segmentation points during node splitting: a. the base purity is minimized: wherein t represents the current node in the decision book and p (k/t) represents a scalar; b. Information gain maximization: Wherein Δi represents the information gain, I (parent) represents the index of the unrepeace of the parent node, Σ child represents the sum of all child nodes, N child represents the number of samples of the child node, N parent represents the number of samples of the parent node, N child /N parent represents the sample duty cycle of the child node, and I (child) represents the index of the unrepeace of the child node; ③ Maximum depth=8, leaf node minimum number of samples=5.
4. 4. The method for constructing the machine learning-based acute necrotizing pancreatitis severe prediction model according to claim 1, wherein the specific operation of the step S4 is that 1520, 1520 and 3040 radiological features are extracted from the pancreatic parenchyma, peripancreatic necrosis and joint regions, respectively, that the pancreas group 1089 features, the pancreas Zhou Zu features and the joint group 2190 features show good consistency after consistency test, that 620, 515 and 1135 features in each model have significant differences among different severity groups, and that the optimal feature subset finally determined by LightGBM and cross-validation contains 10 pancreas features, 9 peripancreatic features and 14 joint features.
5. 5. The method for constructing a machine learning method-based acute necrotizing pancreatitis severe prediction model according to claim 1, wherein the specific operation of step S5 is to distinguish the severity of the disease using five machine learning algorithms and an optimal feature subset extracted from portal CECT data based on a radiological development model.

Description

Machine learning method-based acute necrotizing pancreatitis severe prediction model construction method Technical Field The invention relates to the technical field of disease diagnosis, in particular to a machine learning method-based acute necrotizing pancreatitis severe prediction model construction method. Background Acute Pancreatitis (AP) is a common acute abdominal disorder characterized by local and systemic inflammatory responses. The clinical course of the disease varies from self-limiting light to moderately severe AP. Worldwide, the incidence rate of AP year is about 33.74/10 ten thousand people, and the death rate is about 1.16/10 ten thousand people. In addition, the incidence of the disease tends to rise year by year. The atlanta classification revised in 2012 classifies AP into two categories, interstitial oedema pancreatitis and necrotizing pancreatitis (ANP) according to morphology and pathology. ANP (morphological type) is a more severe type, often involving multiple organ systems with more severe clinical manifestations, leading to higher mortality and poorer prognosis. About 20% of AP patients progress to moderate to severe acute pancreatitis (MSAP) or Severe Acute Pancreatitis (SAP). SAP is the most critical type, characterized by high mortality (20% -40%) and poor prognosis. Although the terms ANP (morphology) and SAP (clinical) are often used interchangeably in the literature, their definition and clinical manifestations are not exactly the same. To establish a clearer distinction based on disease severity in ANP populations, the present invention introduces the new terms "acute necrotizing moderate pancreatitis (ANMSP)" and "Acute Necrotizing Severe Pancreatitis (ANSP)". The term integrates imaging and clinical features, providing a more accurate framework for classification and management. Contrast Enhanced Computed Tomography (CECT) is the primary imaging examination means to evaluate the morphological features of necrotizing pancreatitis. CECT has a broader range of applications than Magnetic Resonance Imaging (MRI) in diagnosing Acute Pancreatitis (AP) and assessing its severity. This may be due to the increased popularity of CT devices, faster scan times, and easier interpretation of their image results by clinicians. CECT can clearly show necrotic areas (peripancreatic necrosis only, pancreatic necrosis only, or both) by highlighting the substantial enhancement of the difference from peripancreatic angiograms, thereby comprehensively assessing the severity of ANP and the extent of surrounding tissue involvement. Furthermore, the radiology concept was first proposed by Lambin et al 2012, which refers to a technique for high-throughput extraction and analysis of a large number of high-quality quantitative image features from medical images. In the early stages of AP, changes in pancreatic morphology in some patients (especially pancreatic necrosis cases) may not be apparent in imaging examinations, resulting in underestimation of disease severity. As a non-invasive method, radiology is able to capture microscopic heterogeneity of lesions that could not be detected by conventional imaging examinations at an early stage. By quantitatively analyzing these features, it establishes a key link between imaging findings and clinical practice, thereby assisting in the selection of therapeutic regimens. In recent years, radiology has been mainly used for diagnosis of pancreatic tumors and identification of different types of pancreatitis. However, there is currently little research focused on the assessment of the severity of ANP, and no research has been done to explore early differential diagnosis by combining the radiological features of pancreatic parenchyma with peripancreatic necrosis. Disclosure of Invention The invention aims to provide a machine learning method-based acute necrotizing pancreatitis severe prediction model construction method. The present invention develops and validates a machine learning model that distinguishes heavy and moderate ANP (i.e., ANSP and ANMSP). These models will be based on radiological features extracted from CECT images of the parenchymal portal vein of the pancreas, peripancreatic necrosis, and combinations thereof. By evaluating the diagnostic performance of these models, early identification of disease severity is achieved and support is provided for treatment-related clinical decisions. The purpose of the invention is realized in the following way: A machine learning method-based acute necrotizing pancreatitis severe prediction model construction method comprises the following steps: Step S1, collecting patient data for later use based on inclusion criteria and exclusion criteria; step S2, preprocessing the patient data in the step 1 and selecting the characteristics to obtain preprocessed patient data for later use; The method comprises the steps of carrying out feature extraction analysis on a CECT image in a portal vein period by adopting