Prediagnosis information processing system based on large model intrahepatic bile duct stone cancerization

Abstract

The invention discloses a pre-diagnosis information processing system based on large model intrahepatic bile duct stone cancerization, which belongs to the field of intrahepatic bile duct stone cancerization, and comprises a data acquisition module for acquiring character data of a patient to form a structured data packet, and a pre-diagnosis information processing module for processing pre-diagnosis information based on an unstructured text input to the patient, the NLP analysis module is used for establishing a sequential clinical basic file and is used for collecting multi-source data of a patient; the multi-modal receiving module is used for forming a complete file in combination with a clinical basic file; the intelligent diagnosis and analysis module is used for receiving the complete file and performing sequential analysis; the grading and decision-making module is used for automatically matching preset action suggestions according to parameters output by the intelligent diagnosis and analysis module; the multimodal data is fused, the dynamic time sequence analysis function is performed, and the pre-diagnosis function focuses on pre-diagnosis of intrahepatic bile duct stone cancerization.

Description

Technical Field

[0001] This invention belongs to the field of intrahepatic bile duct stone carcinoma, specifically, it relates to a pre-diagnosis information processing system based on a large model of intrahepatic bile duct stone carcinoma. Background Technology

[0002] Intrahepatic bile duct stones (Hepatolithiasis) is a common biliary tract disease in Asia. Its most serious complication is intrahepatic cholangiocarcinoma (H-CCA). H-CCA has an insidious onset, is highly aggressive, and has an extremely poor prognosis, with a low five-year survival rate. Therefore, early and accurate cancer risk screening and stratified management for patients with intrahepatic bile duct stones are crucial for improving patient prognosis and securing optimal surgical opportunities.

[0003] Currently, the clinical diagnosis of H-CCA mainly relies on the following methods and their limitations:

[0004] Imaging examinations, such as ultrasound, CT, and MRI, are used. However, in the early stages, the imaging manifestations of H-CCA are often similar to inflammatory lesions such as cholangitis and thickening of the bile duct wall, making them difficult to distinguish. The diagnostic specificity and sensitivity are not ideal, and the diagnosis is highly dependent on the personal experience of the radiologist.

[0005] Serum tumor marker detection, such as cancer antigen 19-9 (CA19-9), can also significantly increase under benign biliary obstruction conditions such as cholangitis, leading to insufficient specificity and a high risk of false positives. Detection values ​​at a single time point cannot effectively reflect the dynamic evolution of the disease.

[0006] Clinical comprehensive judgment: Doctors make a comprehensive judgment based on the patient's medical history, symptoms, imaging, and laboratory results. This method lacks unified and quantifiable standards, and the accuracy and consistency of the assessment results are greatly affected by the physician's experience level, making it difficult to popularize in primary hospitals or among inexperienced doctors, resulting in inconsistent levels of patient management.

[0007] Meanwhile, existing prediagnosis systems are mostly general models for multiple diseases, lacking a deep understanding and modeling of the specific pathophysiological evolution process from intrahepatic bile duct stones to H-CCA. They cannot accurately capture its unique risk factors. Even if the model is trained by implanting a large amount of case data, the assessment result is often a general, single-dimensional "high / medium / low" risk label, which cannot accurately judge the prediagnosis result. Summary of the Invention

[0008] To address the above shortcomings, this invention provides a pre-diagnosis information processing system for intrahepatic bile duct stone carcinoma based on a large model, capable of fusing multimodal data and performing dynamic time-series analysis, comprising the following modules:

[0009] A data acquisition module is used to collect patients' textual data and form structured data packages. The data acquisition module includes a medical history collection unit and a geographic information unit.

[0010] An NLP analysis module for time-series clinical baseline records is constructed based on unstructured text input from patients.

[0011] A multimodal receiving module used to collect multi-source patient data and combine it with basic clinical records to form a complete medical record;

[0012] The intelligent diagnostic analysis module for receiving complete archives and performing sequential analysis includes a spatial structure quantification subunit for image analysis, a chronic inflammation burden score subunit for time-series analysis and generating dynamic trends, and a malignant lesion index subunit for final prediction.

[0013] The system automatically matches the pre-defined action recommendations to the classification and decision-making modules based on the parameters output by the chronic inflammation burden score subunit and the malignancy index subunit.

[0014] Furthermore, the NLP analysis module adopts the RoBERTa-wwm-ext-large model and establishes temporal relationships for the collected text.

[0015] Furthermore, the multimodal receiving module includes an OCR subunit and an image data subunit. The information collected by the two subunits is used to combine with basic clinical records to form a complete record.

[0016] The OCR subunit is used to recognize and extract text information from image files uploaded by patients and output it to the NLP analysis module.

[0017] The image data subunit is used to identify the complete set of DICOM image files uploaded by the patient and then store them in the image server.

[0018] Furthermore, the spatial structure quantization subunit is used to retrieve DICOM image files stored in the image server, reconstruct them into 3D data volumes, preprocess the 3D data volumes, and then analyze them using the built-in Swin-UNETR model.

[0019] The chronic inflammation burden scoring subunit uses the built-in XGBoost gradient boosting tree model to score the chronic inflammation burden of patients based on the complete patient profile.

[0020] The malignant lesion index unit uses a built-in multimodal fusion model to score the malignant lesion index based on the complete archives in three dimensions: imaging, time series, and clinical presentation.

[0021] Furthermore, the preset action suggestions include the following three ranges:

[0022] When a patient has both a high chronic inflammation burden score and a low malignancy index score, they are assessed as having a high-risk precancerous lesion status. It is recommended that they make an appointment with a specialist for diagnosis and undergo preventive resection.

[0023] When both a high chronic inflammation burden score and a high malignancy index score are present, the patient is assessed as being highly suspected of having cancer and is advised to seek medical attention as soon as possible.

[0024] A score of low chronic inflammatory burden indicates a low-risk status, and regular follow-up examinations are recommended.

[0025] Compared with the prior art, the present invention has the following advantages:

[0026] We constructed a deep learning model specifically designed for the clinical scenario of intrahepatic bile duct carcinoma (H-CCA). By accurately modeling the pathophysiological process unique to this disease, we significantly improved the accuracy, sensitivity, and specificity of early cancer risk identification.

[0027] By incorporating time-series data of key patient indicators (such as CA19-9) into the model, the changing trends, rates, and accelerations of the indicators can be automatically analyzed, capturing dynamic evolution patterns that are more predictive than single static values.

[0028] By combining the Chronic Inflammatory Burden Score (CIBS) with the Malignant Lesion Index (MSI), nine quadrants are formed, generating standardized recommendations covering the entire process of "low-risk monitoring," "medium-risk management," and "high-risk intervention." This provides high-level decision support for physicians, effectively bridging the experience gap and demonstrating extremely high clinical application value. Detailed Implementation

[0029] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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] Example 1

[0031] This embodiment provides a pre-diagnosis information processing system for intrahepatic bile duct stone carcinoma based on a large model, including the following modules:

[0032] The data acquisition module outputs a structured data package containing medical history duration, inflammation frequency score, and geographic risk weight. The data acquisition module includes a medical history collection unit and a geographic information unit.

[0033] The medical history collection unit collects patients' medical history through a built-in guided questionnaire, including the filling in of key time points (i.e. the year of the first diagnosis of stones, and this module automatically calculates the duration of the medical history based on the current time) and the filling in of inflammation quantification (i.e. the average number of hospitalizations per year / month due to right upper quadrant pain and fever during the medical history period). Key time points (such as the time of each hospitalization) are used as variables of core risk, while inflammation quantification is used as an indicator for calculating frequency and severity.

[0034] The geographic information unit collects external environmental information from patients through a built-in guided questionnaire, including life history (including but not limited to whether there is a habit of eating raw freshwater fish and shrimp that is associated with the risk of liver fluke infection) and geographic information (the geographic location filled in by the patient is cross-compared with the built-in "Liver Fluke Disease Epidemic Area Database" and "H-CCA High Incidence Area Database"). The comparison results are used to generate geographic risk weights.

[0035] The NLP analysis module transforms unstructured text (fragmented, colloquial descriptions) input by patients into a well-structured timeline of disease evolution containing key nodes. It employs a RoBERTa model finely tuned on a Chinese medical corpus to analyze patient-input text or uploaded old medical record photos (collected and transmitted by the multimodal receiving module). For text extraction, it not only identifies "abdominal pain" and "fever," but also focuses on extracting signal words strongly correlated with intrahepatic bile duct stones, such as "liver lobe atrophy," "bile duct wall thickening," and "progressive jaundice." Finally, based on the identified time adverbs and event descriptions in the text, it automatically constructs and outputs a structured, time-series clinical baseline profile, such as "Time: 2021, Event: Intermittent abdominal pain - Time: 2022, Event: Weight loss of 5kg."

[0036] This multimodal receiving module is used to collect multi-source patient data and combine it with basic clinical records to form a complete patient profile. It includes an OCR subunit and an image data subunit.

[0037] The OCR subunit is used to recognize and extract text information from image files uploaded by patients (such as lab reports and imaging reports), and output it to the NLP analysis module. The NLP analysis module then performs named entity recognition (such as CA19-9, ALP, GGT, and diagnostic conclusions), relation extraction (binding indicators with values ​​and dates), and structured output (generating standard JSON) on the text information in the image files. After the JSON is generated, key indicators (especially CA19-9 and CEA), i.e., the rate of change and acceleration of CA19-9, can be automatically calculated by querying the patient's historical records again, and plotted into a visual curve for patients and doctors to view.

[0038] The image data subunit is used to identify the complete set of DICOM image files uploaded by the patient, store them in the hospital's internal image server (PACS system), and label the patient's ID.

[0039] The intelligent diagnostic analysis module, used to receive complete files and perform sequential analysis, includes a spatial structure quantification subunit, a chronic inflammation burden scoring subunit, and a malignant lesion index subunit:

[0040] The spatial structure quantification subunit uses the Swin-UNETR model. It retrieves a complete set of DICOM image sequences from the PACS system, which are stored in the multimodal receiver module. Then, it uses Pydicom to read these DICOM files and reconstructs several 2D slices into a 3D data volume (3DVolume / Tensor) in memory. Subsequently, it performs preprocessing such as window width and window level adjustment and voxel resampling to form a standardized 3D data volume. Then, it is analyzed by the built-in Swin-UNETR model to quantify spatial structure indicators such as liver lobe atrophy rate and bile duct dilatation, which are used as key parameters for subsequent CIBS scoring.

[0041] The Chronic Inflammatory Burden Score (CIBS) subunit uses the XGBoost gradient boosting tree model to score (0-100) based on the duration of medical history, frequency of inflammation, long-term ALP / GGT levels, and degree of bile duct dilatation (automatically calculated by the Swin-UNETR model, such as 1.5cm in the widest diameter of the left intrahepatic bile duct, which is 1.5cm; this precise value is the degree of bile duct dilatation). CIBS score represents the cumulative pressure and degree of cancer in the patient's biliary system due to long-term stone obstruction and inflammation.

[0042] The Malignancy Index (MSI) sub-unit employs a cross-modal Transformer model, scoring the Malignancy Index after inputting data from three dimensions (A, B, and C).

[0043] A. Image dimensions: such as image features obtained from spatial structure quantification sub-units (e.g., degree of bile duct dilation);

[0044] B. Time series dimension: such as the rate of change of CA19-9;

[0045] C. Clinical dimension: such as signal words like "bile duct wall thickening";

[0046] It should be noted that the cross-modal Transformer model used in this scoring process uses a complex "attention mechanism" to allow data from different modalities to interact deeply, reference each other, and enhance information within the model. At the same time, it requires a large number of complete data packages (including images, long-term laboratory reports, and clinical history) of patients with intrahepatic bile duct stones for training. During the training process, the model will automatically learn how to assign the most appropriate weights to different features (such as a certain texture feature of the image or a certain growth pattern of CA19-9) to provide a quantitative assessment of the "probability of malignancy" of new cases. The score directly reflects the intensity and consistency of the "malignant lesion" captured by the model from the three dimensions of imaging, time series, and clinical data.

[0047] Specifically, static data streams (such as weight loss, a certain texture feature of an image, etc.) are processed by a multilayer perceptron (MLP), while time-series data streams (such as a certain growth pattern of CA19-9) are processed by a recurrent neural network (RNN) to determine whether CA19-9 is continuously rising, falling, or stable, the speed of the rise, and the upward trend (accelerating or slowing down). Therefore, the combination of the two is fed into the decision layer of the cross-modal Transformer model to calculate the final MSI score of 0-100.

[0048] The grading and decision-making module automatically matches pre-defined action recommendations based on the scores output by the Chronic Inflammatory Burden Score (CIBS) and Malignancy Index (MSI) sub-units.

[0049] The principle is to use a built-in two-dimensional decision matrix (which requires consensus from an expert panel) to divide CIBS and MSI into three risk levels: "low," "medium," and "high," based on their score ranges, forming a 3x3 decision matrix.

[0050] The Chronic Inflammatory Burden Score (CIBS) is classified as follows:

[0051] Low levels (<40): suggest a lower degree of chronic inflammation and structural damage to the intrahepatic biliary system;

[0052] Moderate levels (40 to 70): suggest significant, long-term accumulation of chronic inflammation and pathophysiological changes;

[0053] High level (>70): suggests the presence of severe, diffuse chronic inflammation, cholestasis, and liver parenchymal atrophy, constituting a high-grade precancerous lesion background;

[0054] The Malignancy Index (MSI) classification is as follows:

[0055] Low suspicion (<20): No statistically significant evidence of malignancy was detected in imaging, laboratory reports, or clinical findings.

[0056] Moderately suspicious (20-50): Non-specific malignant signs were detected, requiring differential diagnosis from severe inflammatory response or benign hyperplasia;

[0057] Highly suspicious (>50): Detected multidimensional and strongly correlated evidence of malignant tumors, suggesting a high probability of cancerous transformation.

[0058] The final result is divided into three different zones, displayed in green, yellow, and red respectively, and automatically matched with preset action suggestions:

[0059] Green indicates low-risk monitoring (Category 1):

[0060] CIBS < 40 and MSI < 20 are considered to be in the double low range, and clinical assessment indicates that the risk of intrahepatic bile duct stone-related complications and cancer is at a low level. The recommended action is to have regular follow-up examinations (once every 11 months), which should include: abdominal ultrasound, liver function tests (including ALP, GGT) and serum tumor markers (CA19-9, CEA).

[0061] Yellow indicates medium-risk monitoring (Category III):

[0062] (1) CIBS: 40-70 and MSI<20, which is in the middle to low range. Clinical assessment indicates a significant chronic inflammatory background, which is a moderate risk factor for H-CCA. However, there is currently no clear evidence of malignancy. The main clinical challenge is to control and reverse the chronic pathological process. The action recommendation is to consult a hepatobiliary surgeon or gastroenterologist to assess the necessity of initiating active medical intervention, such as oral choleretic drugs (e.g., ursodeoxycholic acid) or endoscopic retrograde cholangiopancreatography (ERCP) for biliary drainage or stone removal to relieve cholestasis. At the same time, the follow-up examination frequency is once every 6 months.

[0063] (2) CIBS (40 and MSI: 20-50, which is in the low to medium range, clinical assessment shows that the background of chronic inflammation is not prominent, but there are non-specific malignant signs. The main clinical contradiction lies in the qualitative diagnosis of these signs and the nature of the suspicious signals. The action recommendation is for the patient to undergo high-resolution imaging examination within 3 months, with the upper abdominal enhanced magnetic resonance imaging (MRI) combined with magnetic resonance cholangiopancreatography (MRCP) as the first choice. At the same time, the re-examination frequency is once every 6 months, and serum tumor markers need to be re-examined to assess their short-term dynamic changes.

[0064] (3) CIBS: 40-70 and MSI: 20-50, which is in the middle range. The clinical assessment shows that the patient has a moderate background of chronic inflammation and moderate suspected malignancy. The clinical decision-making is highly complex and the risk is significantly increased. The action recommendation is to submit this case to a multidisciplinary team (MDT) for consultation and develop an individualized comprehensive management plan that includes close monitoring (every 3-6 months), detailed imaging examination and possible endoscopic / surgical intervention.

[0065] Red indicates high-risk monitoring (two categories):

[0066] (1) CIBS>70 and MSI<20, which is the high-low range, clinical assessment indicates that the patient is in a high-grade precancerous lesion state. Although there is no direct evidence of cancer, the risk of cancer is extremely high. The action recommendation is to consult a hepatobiliary surgeon to comprehensively assess the indications and risks of prophylactic liver resection. The purpose of the surgery is to eradicate lesions including stones, sclerosing cholangitis and atrophic liver segments, thereby eliminating the pathological basis for cancer.

[0067] (2) If CIBS is any value and MSI>50, which is any high range, the clinical assessment is highly suspicious of cancer. The weight of MSI score is decisive at this stage. The action recommendation is to go to a medical center with the ability to diagnose and treat hepatobiliary tumors immediately for a comprehensive evaluation for the purpose of obtaining pathological diagnosis and radical surgery. At the same time, the system automatically marks this case as an "emergency".

[0068] It should be noted that, in addition to the categories mentioned above, there are two other categories:

[0069] If CIBS > 70 and MSI is between 20 and 50, it falls within the high school range.

[0070] CIBS < 40 and MSI > 50, which indicates a low to high range;

[0071] Because these cases are rare in clinical practice, but in order to be responsible for patients, they should all be strictly evaluated. Therefore, they are all classified as "red for high-risk monitoring", which can be defined as high-low intervals and any high intervals.

[0072] Thus, the entire system has formed a complete closed loop from data collection, refinement, in-depth analysis to decision support.

[0073] The workflow of the aforementioned pre-diagnosis information processing system is illustrated in the following examples.

[0074] Example 2

[0075] Mr. Li, 42 years old, entered his personal information through the data collection module:

[0076] Guided questionnaire:

[0077] Year of first diagnosis: 2021 (the platform automatically calculates the duration of the medical history, approximately 4 years);

[0078] History of cholangitis attacks: None, no history of abdominal pain, fever, or jaundice;

[0079] Epidemiological history: No history of consuming raw freshwater fish and shrimp, and resides in an area not endemic for liver fluke disease;

[0080] The following text was entered: During a company physical examination, small stones were found in my liver. I don't feel any discomfort. I would like to ask if it's serious.

[0081] The patient also uploaded the following lab results: CA19-9: 12 U / mL (normal range), ALP: 75 U / L (normal range), GGT: 30 U / L (normal range). An abdominal ultrasound was also uploaded, showing no dilation of intrahepatic bile ducts, and several strong echogenic foci, the largest approximately 0.5 cm, with posterior acoustic shadowing, visible in the left lobe of the liver.

[0082] The pre-diagnosis information processing system processes the above information:

[0083] S1. The data collection module identified: short medical history (4 years), completely asymptomatic, no epidemiological risk factors, and the NLP analysis module confirmed that the patient had no complaints of discomfort.

[0084] S2, the multimodal receiving module (OCR subunit) extracts and structures the test report data, generates standard JSON, and confirms that all key indicators are within the normal range;

[0085] S3-1, Spatial Structure Quantification Subunit: Based on the ultrasound findings of no bile duct dilation and no liver lobe atrophy, the liver morphology was determined to be normal.

[0086] S3-2, Chronic Inflammatory Burden Scoring Subunit: Based on the input medical history duration (4 years), inflammation frequency (0), low GGT / ALP levels, and bile duct dilatation of 0, the CIBS score is 15 (low level).

[0087] S3-3, the Malignancy Index subunit, is based on the absence of malignant clinical manifestations (weight loss, etc.), low and stable CA19-9 levels, and no suspicious signs on imaging, with an MSI score of 5 (low suspiciousness).

[0088] S4, the grading and decision-making module automatically matches pre-defined action recommendations based on CIBS and MSI scores:

[0089] CIBS (40 and MSI < 20) indicates a double low range, displayed in green. A routine annual check-up is recommended. The check-up should include: abdominal ultrasound, liver function tests (including ALP and GGT), and serum tumor markers (CA19-9, CEA). The system will automatically create a health record and mark Mr. Li's ID, and will automatically send a check-up reminder after 11 months (based on the registered mobile phone number).

[0090] Example 3

[0091] Ms. Li, 58 years old, entered her personal information through the data collection module:

[0092] Guided questionnaire:

[0093] Year of first diagnosis: 2014 (the platform automatically calculates the duration of the medical history, approximately 10 years);

[0094] History of cholangitis attacks: Over the past 5 years, there have been 1-2 episodes of right upper quadrant abdominal distension and pain per year on average, without fever, which resolve spontaneously;

[0095] Epidemiological history: No history of consuming raw freshwater fish and shrimp, and resides in an area not endemic for liver fluke disease;

[0096] Enter the text yourself: I've had kidney stones for many years, and they hurt occasionally. During my most recent physical exam, the doctor said that my liver function indicators, GGT and CA19-9, were slightly higher than normal.

[0097] The patient also uploaded the following lab results: CA19-9: 65 U / mL (mildly elevated), ALP: 140 U / L (mildly elevated), GGT: 180 U / L (moderately elevated). Simultaneously, an enhanced abdominal CT scan was uploaded: the intrahepatic bile ducts in the left lobe of the liver showed mild beaded dilatation, with slightly thickened and enhanced duct walls. The left lobe of the liver was slightly smaller than the right lobe, and no obvious space-occupying lesions were observed.

[0098] The pre-diagnosis information processing system processes the above information:

[0099] S1. The data collection module identified that the patient had a short medical history (10 years) and intermittent symptoms, which constituted a medium-risk background. The NLP analysis module captured the keyword "high indicators".

[0100] S2, the multimodal receiving module (OCR subunit) extracted and structured the test report data, generated standard JSON, and confirmed that ALP and GGT were continuously elevated to a mild to moderate level, and CA19-9 was higher than the normal value but not to a level that strongly suggests malignancy.

[0101] S3-1, the spatial structure quantification subunit automatically measures and confirms mild dilation of the left hepatic duct (average diameter 0.8cm) on CT images based on the Swin-UNETR model, and calculates the left hepatic lobe volume / right hepatic lobe volume ratio as 0.55, indicating the presence of a mild atrophy-proliferation complex;

[0102] S3-2, Chronic Inflammatory Burden Scoring Subunit: Based on a long medical history, moderate inflammation frequency, persistently elevated GGT / ALP, and quantified bile duct dilation and liver atrophy indicators, the CIBS score was 58 (moderate level).

[0103] S3-3, the Malignant Lesion Index subunit, is based on CA19-9 being higher than normal but not reaching a level highly suggestive of malignancy, lacking clear clinical manifestations of malignancy, and the presence of mild atrophic-hyperplastic complex on imaging, with an MSI score of 35 (moderately suspicious).

[0104] S4, the grading and decision-making module automatically matches pre-defined action recommendations based on CIBS and MSI scores:

[0105] A CIBS score of 40-70 and an MSI score of 20-50, which falls within the mid-range and is indicated by a yellow color, suggests the presence of both a moderate background of chronic inflammation and moderately suspicious malignant signs. It is recommended that this case be referred to a multidisciplinary team (MDT) for consultation, and an individualized comprehensive management plan should be developed that includes close monitoring (every 3-6 months), detailed imaging examinations, and possible endoscopic / surgical interventions.

[0106] Example 4

[0107] Mr. Zhao, 65 years old, entered his personal information through the data collection module:

[0108] Guided questionnaire:

[0109] Year of first diagnosis: 1999 (the platform automatically calculates the duration of the medical history, approximately 25 years);

[0110] History of cholangitis: recurrent right upper quadrant pain, high fever, chills, and 4 hospitalizations in the past 10 years;

[0111] Epidemiological history: Early history included a habit of eating raw freshwater fish slices;

[0112] Enter the following text: I have lost almost 20 pounds in the past six months, have no appetite, my skin and eyes have always been yellow, and my urine tastes like strong tea.

[0113] The patient also uploaded the following lab reports: CA19-9: 152 U / mL (December 2023 lab report), CA19-9: 980 U / mL (April 2024 lab report, indicating a sharp increase compared to 2023), ALP: 450 U / L (mildly elevated), GGT: 510 U / L, total bilirubin: 180 μmol / L. The patient also uploaded MRCP imaging: severe atrophy of the left lobe of the liver, with huge cystic dilatations of the intrahepatic bile ducts filled with stones and sediment-like echoes. An irregular soft tissue mass of approximately 4 cm in size with delayed enhancement was visible at the confluence of the dilated left hepatic ducts. Enlarged lymph nodes were also observed in the porta hepatis.

[0114] The pre-diagnosis information processing system processes the above information:

[0115] S1. The data acquisition module identified: a very long medical history (25 years), a history of frequent severe cholangitis attacks, and epidemiological high-risk factors. The NLP analysis module captured classic malignant tumor symptoms such as "weight loss" and "persistent jaundice".

[0116] S2, the multimodal receiving module (OCR subunit) extracts and structures the test report data, generates standard JSON, and time-series data shows that CA19-9 has increased exponentially within 4 months (this is a very strong malignant signal), and liver function indicators also show severe cholestasis.

[0117] S3-1, Spatial Structure Quantification Subunit: Based on the Swin-UNETR model, it was confirmed that the left liver was severely atrophied (volume ratio < 0.3) and the bile ducts were severely dilated. The analysis showed that the lesion area had typical malignant tumor characteristics such as high heterogeneity, irregular margins, and rich blood supply.

[0118] S3-2, the chronic inflammation burden scoring subunit, based on a 25-year medical history, frequent inflammation, high GGT / ALP, and changes in spatial structure determined by the spatial structure quantification subunit, scored CIBS 91 (high level).

[0119] S3-3, the Malignancy Index subunit, is assigned an MSI score of 95 (highly suspicious) based on strong clinical malignancy characteristics, a rapid growth rate of CA19-9 (captured by RNN), and highly malignant radiomics features.

[0120] S4, the grading and decision-making module automatically matches pre-defined action recommendations based on CIBS and MSI scores:

[0121] CIBS: 91 (any value) and MSI>50, which is an arbitrary high range and displayed in red, indicates a high suspicion of cancerous changes. The system has detected multidimensional and strongly correlated evidence of malignant tumors. The action recommendation is to advise the patient to go to a medical center immediately for a comprehensive evaluation to obtain a pathological diagnosis and radical surgery, and to mark it as an "emergency".

[0122] It should be noted that the structure described in this invention can be implemented in many different forms and is not limited to the embodiments described. Any equivalent transformations made by those skilled in the art based on the content of this specification, or direct or indirect applications in other related technical fields, such as the loading and unloading of other items, are included within the protection scope of this invention.

Claims

1. A pre-diagnosis information processing system for intrahepatic bile duct stone carcinoma based on a large model, characterized in that, Includes the following modules: A data acquisition module is used to collect patients' textual data and form structured data packages. The data acquisition module includes a medical history collection unit and a geographic information unit. An NLP analysis module for time-series clinical baseline records is constructed based on unstructured text input from patients. A multimodal receiving module used to collect multi-source patient data and combine it with basic clinical records to form a complete medical record; The intelligent diagnostic analysis module for receiving complete archives and performing sequential analysis includes a spatial structure quantification subunit for image analysis, a chronic inflammation burden score subunit for time-series analysis and generating dynamic trends, and a malignant lesion index subunit for final prediction. The system automatically matches the pre-defined action recommendations to the classification and decision-making modules based on the parameters output by the chronic inflammation burden score subunit and the malignancy index subunit.

2. The pre-diagnosis information processing system for intrahepatic bile duct stone carcinoma based on a large model as described in claim 1, characterized in that: The NLP analysis module uses the RoBERTa-wwm-ext-large model and establishes temporal relationships for the collected text.

3. The pre-diagnosis information processing system for intrahepatic bile duct stone carcinoma based on a large model as described in claim 2, characterized in that: The multimodal receiving module includes an OCR subunit and an image data subunit. The information collected by the two subunits is combined with the basic clinical records to form a complete record. The OCR subunit is used to recognize and extract text information from image files uploaded by patients and output it to the NLP analysis module. The image data subunit is used to identify the complete set of DICOM image files uploaded by the patient and then store them in the image server.

4. The pre-diagnosis information processing system for intrahepatic bile duct stone carcinoma based on a large model as described in claim 3, characterized in that: The spatial structure quantization subunit is used to retrieve DICOM image files stored in the image server, reconstruct them into 3D data volumes, preprocess the 3D data volumes, and then analyze them using the built-in Swin-UNETR model. The chronic inflammation burden scoring subunit uses the built-in XGBoost gradient boosting tree model to score the chronic inflammation burden of patients based on the complete patient profile. The malignant lesion index unit uses a built-in multimodal fusion model to score the malignant lesion index based on the complete archives in three dimensions: imaging, time series, and clinical presentation.

5. The pre-diagnosis information processing system for intrahepatic bile duct stone carcinoma based on a large model as described in claim 1, characterized in that: The proposed pre-defined actions include the following three ranges: When a patient has both a high chronic inflammation burden score and a low malignancy index score, they are assessed as having a high-risk precancerous lesion status. It is recommended that they make an appointment with a specialist for diagnosis and undergo preventive resection. When both a high chronic inflammation burden score and a high malignancy index score are present, the patient is assessed as being highly suspected of having cancer and is advised to seek medical attention as soon as possible. A score of low chronic inflammatory burden indicates a low-risk status, and regular follow-up examinations are recommended.