Early intelligent screening and risk assessment system for liver cancer based on multi-omics data

Abstract

The invention discloses a liver cancer early intelligent screening and risk assessment system based on multi-omics data, and relates to the technical field of liver cancer screening. The system comprises an information collection module used for collecting multi-omics data of a patient and constructing a feature data set of the patient; the intelligent screening module is responsible for constructing and training a liver cancer early risk assessment model; the risk assessment module is used for inputting the constructed patient feature data set into the trained risk assessment model; the probability checking module compares the calculated risk probability value with a preset risk threshold value; the grade dividing module divides the liver cancer risk of the patient into corresponding risk grades according to the checked risk probability value; and the tracing follow-up module is used for performing continuous follow-up visit on the patient. According to the method, the technical problem that multi-dimensional information cannot be deeply fused in early screening of liver diseases related to alcoholism in the prior art is solved by establishing the risk assessment model of people suffering from long-term alcoholism.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of liver cancer screening, and in particular to a liver cancer early intelligent screening and risk assessment system based on multi-omics data. BACKGROUND

[0002] Liver cancer is a malignant tumor originated from uncontrolled proliferation of malignant cells in the liver, and the most common type is hepatocellular carcinoma. Because the liver has strong compensatory function, early liver cancer often has no obvious symptoms, resulting in most patients being diagnosed at an advanced stage, missing the best treatment opportunity. The treatment means for advanced liver cancer is limited, and the curative effect is poor, and the five-year survival rate of patients is extremely low. Therefore, early screening is crucial. Early liver cancer can significantly improve the five-year survival rate to 60%-80% through radical treatment methods such as surgical resection and local ablation, and even achieve clinical cure. Because some people maintain the habit of long-term alcohol abuse, regular screening of such groups can detect lesions in the early stage before symptoms appear, and therefore early screening can effectively reduce mortality and win valuable treatment time window for patients.

[0003] Existing liver cancer screening systems mostly rely on traditional methods such as liver ultrasound and alpha-fetoprotein detection, which have value in universal screening, but for early screening of alcohol-related liver diseases, the existing mode often treats imaging and serological indicators in isolation and fails to deeply integrate multi-dimensional information, which has certain defects.

[0004] Therefore, there is an urgent need to protect a liver cancer early intelligent screening and risk assessment system based on multi-omics data. SUMMARY

[0005] The application provides a liver cancer early intelligent screening and risk assessment system based on multi-omics data, which solves the technical problem in the prior art that multi-dimensional information cannot be deeply integrated for early screening of alcohol-related liver diseases by establishing a risk assessment model for long-term alcohol abuse groups.

[0006] To solve the above technical problems, the application provides the following technical solutions: The application provides a liver cancer early intelligent screening and risk assessment system based on multi-omics data, which includes: An information collection module for collecting multi-omics data of patients and constructing a patient feature data set; An intelligent screening module responsible for constructing and training a liver cancer early risk assessment model and continuously optimizing the model through an optimization algorithm; A risk assessment module for inputting the constructed patient feature data set into the trained risk assessment model to calculate the early liver cancer risk probability value of the patient; The probability test module: the accuracy of the output risk probability value is verified by comparing the calculated risk probability value with the preset risk threshold; The grade division module: according to the tested risk probability value, the liver cancer risk of the patient is divided into a corresponding risk grade; The trace follow-up module: used for continuous follow-up of the patient, and recording the time sequence change data of the health condition of the patient.

[0007] The technical scheme provided by the present application has at least the following beneficial effects: When constructing the risk assessment model, the long-term alcoholic patient is trained through the characteristic data of the long-term alcoholic patient, so that a characteristic risk assessment model conforming to the early liver cancer of the long-term alcoholic patient can be constructed, and the accurate judgment of the early liver cancer risk of the long-term alcoholic patient is increased by taking the specific features closely related to the evolution of alcoholic liver disease to liver cancer as the benchmark.

[0008] The probability verification module of the present application compares and optimizes the probability value output by the risk assessment model, which can more accurately reflect the risk value of the early liver cancer of the patient, and further improve the accuracy of screening.

[0009] The risk division module of the present application can divide the early liver cancer risk of the patient, and different trace periods are taken according to different grades, so that the patient data can be updated, and the overall operation efficiency of the system can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0010] In order to more clearly illustrate the technical scheme in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0011] Figure 1 is the flowchart of the early intelligent screening and risk assessment system for liver cancer based on multi-omics data provided by the embodiments of the present application. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical scheme and advantages of the present application more clear, the following will further describe the embodiments of the present application in combination with the drawings.

[0013] EMBODIMENT Early intelligent screening and risk assessment system for liver cancer based on multi-omics data.

[0014] Please refer to Figure 1 is the early intelligent screening and risk assessment system for liver cancer based on multi-omics data provided by the embodiments of the present application.

[0015] I. Information Collection Module Multi-omics data included abdominal ultrasound imaging data and cfDNA whole-genome sequencing data; Abdominal ultrasound image data is preprocessed to extract features. The specific steps are as follows: an initial three-dimensional geometric model of the liver and its internal vascular system is generated, and an interactive three-dimensional digital twin model of the liver is generated by surface optimization of the three-dimensional geometric model. It should be noted that the preprocessing process is as follows: a trained deep neural network segmentation model is used to perform semantic segmentation on the preprocessed image data to obtain the voxel mask of the liver parenchyma. Then, a vascular enhancement filtering algorithm and a region growing method are used to extract the vascular tree structure of the intrahepatic portal vein and hepatic vein from the enhanced scan image data. At the same time, a moving cube algorithm is used to reconstruct the three-dimensional mesh of the liver parenchyma voxel mask and the vascular tree structure to generate the corresponding surface geometry model. The surface optimization steps of the model are as follows: the generated surface mesh is smoothed by Laplacian and the mesh is simplified. A diffuse texture map is generated based on the gray values ​​of the original image. A normal map is synthesized to enhance the visual perception of surface details.

[0016] The information collection module anchors and integrates feature information extracted from genomics and radiomics into a three-dimensional liver digital twin in the form of visual elements, including color coding, highlight points, and texture mapping. cfDNA whole genome sequencing data were analyzed using bioinformatics processes to detect somatic mutations, copy number variations, and methylation characteristics for feature extraction. The feature dataset is generated by mapping abdominal ultrasound imaging data and cfDNA whole genome sequencing data to the specific spatial location of a three-dimensional liver mesh model using a hemodynamic model, and then rendering the three-dimensional liver mesh model in a virtual reality environment and displaying it as a fusion of visual elements.

[0017] It should be noted that the hemodynamic model mapping steps are as follows: based on the three-dimensional model of the liver vascular system, the blood flow distribution is simulated, the cfDNA features with high mutation allele frequencies are mapped to high blood supply areas, and according to the physiological functional characteristics of different liver regions, the mutations of metabolism-related genes are mapped to metabolically active liver segments. At the same time, the genomic instability features are spatially correlated with the heterogeneous regions of the imaging.

[0018] II. Intelligent Screening Module The risk assessment model extracts specific features closely related to the progression of alcoholic liver disease to liver cancer from genomic, radiomic, and clinical data of long-term alcohol drinkers, and selects a deep neural network as the basic architecture of the model, using feature data from a cohort of long-term alcohol abusers to fine-tune the model. It should be noted that the risk assessment model is fine-tuned using a differential learning rate, in which the learning rate of the model close to the output layer is higher than that of the model close to the input layer.

[0019] The risk assessment model is optimized by a loss function and a parameter optimization strategy. The loss function uses Focal Loss as the loss function of the model, and the parameter optimization strategy uses the Bayesian optimization algorithm to automatically search and determine the key parameters of the model.

[0020] It should be noted that the Focal Loss function is used to alleviate the imbalance between liver cancer patients and non-liver cancer patients in the long-term alcohol abuse patient cohort.

[0021] The specificity features include: genomic data related to alcohol metabolism and alcoholic liver damage gene mutations and methylation markers; quantitative texture and morphological features from imageomics data that can characterize alcoholic hepatitis, liver fibrosis, and malignant regeneration nodules; composite scores based on liver function indicators and serum marker dynamic changes from clinical data.

[0022] III. Risk assessment module Before the feature data set is input, it is transformed using the StandardScaler standardizer to match the feature list used during training of the risk assessment model. It should be noted that before the feature data set is input, the system will check whether the current input patient feature data set matches the feature list used during model training. If there are missing features, the mean difference calculated from the training set is used to supplement the input data.

[0023] The calculation process of the cancer risk probability value is as follows: the predict_proba method of the risk assessment model is called to obtain two arrays containing liver cancer probability values from the model output, and one of the arrays is extracted as the final early liver cancer risk probability value. The array structure is [prob_class_0, prob_class_1], where prob_class_0 represents the probability of belonging to the non-liver cancer category, and prob_class_1 represents the probability of belonging to the early liver cancer category.

[0024] The risk assessment module generates and dynamically renders a risk heat map on the corresponding anatomical position of the three-dimensional liver digital twin through the virtual reality visualization engine, combining the final risk probability value with the image space features. It should be noted that the image space feature data includes the three-dimensional spatial coordinates, volume, and morphological description information of multiple suspected lesion regions segmented from medical images.

[0025] The generation step of the risk heat map is: based on the preset weight distribution algorithm, the calculated overall risk probability value is decomposed and associated to multiple suspected lesion regions, the local risk intensity value of each region is calculated, the corresponding color attribute and transparency attribute of each region are calculated, and a semi-transparent heat texture cover layer is generated and updated in real time on the corresponding anatomical position of the three-dimensional liver digital twin.

[0026] It should be noted that the risk intensity value is visually represented by color mapping and region transparency.

[0027] IV. Probability testing module The preset verification strategy includes a credibility evaluation strategy and a feature consistency verification strategy; The credibility evaluation strategy is obtained by calculating the credibility score of the Softmax layer of the risk assessment model when outputting the risk probability value; The feature consistency verification strategy extracts the top K features that contribute most to the current risk probability value through the model explainability engine, and compares the top K features with the known early liver cancer key biomarkers in the medical knowledge base to obtain the comparison result; The probability testing module calculates the comprehensive reliability weight based on the comparison result of the credibility evaluation strategy and the feature consistency verification strategy, and modifies the original risk probability value according to the comprehensive reliability weight to obtain the final risk probability value; The comprehensive weight is calculated using the weighted geometric mean formula, which is as follows: ; In the formula, represents the degree of bias towards credibility; represents the degree of bias towards feature consistency; represents the credibility evaluation score; represents the feature consistency verification score; represents the comprehensive reliability weight; The original risk probability value is modified using the reliability weighted shrinkage method, which is as follows: ; In the formula, represents the final risk probability value; represents the original risk probability value; represents the preset prior probability, representing the default risk level when the information is completely unreliable; represents the comprehensive reliability weight.

[0028] V. Grade division module The grade division module compares the final risk probability value with a plurality of predefined risk probability thresholds, and divides the patient into a corresponding preliminary risk grade according to the comparison result. The risk threshold includes two thresholds T1 and T2, and the preliminary risk level includes three risk levels of low, medium and high. When the risk probability value is less than T1, it is classified as a “low risk” level; when T1 ≤ risk probability value < T2, it is classified as a “medium risk” level; and when the risk probability value is greater than or equal to T2, it is classified as a “high risk” level. It should be noted that if the patient is a hepatitis B virus carrier or a severe cirrhosis patient, the classification level will automatically increase by one.

[0029] The preliminary risk level is fused by weighting the concentration of serological markers. According to the comprehensive score after weighting fusion, the final risk level is determined.

[0030] It should be noted that the serological markers include one of alpha-fetoprotein, alpha-fetoprotein heterogeneity ratio, and abnormal prothrombin.

[0031] The concentration of serological markers is converted into a marker risk score by a linear function, and the comprehensive risk score is calculated by weighted summation, as follows: ; In the formula, indicates the comprehensive risk score, indicates the weight of the preliminary risk level; indicates the threshold of the preliminary risk level; indicates the weight of the serological marker concentration; indicates the score of the serological marker concentration, which is obtained by linear interpolation; wherein: when the threshold of the preliminary risk level is classified as a low level, the weight of the serological marker is increased and the weight of the preliminary risk level is decreased; when the threshold of the preliminary risk level is classified as a medium level and a high level, the weight of the preliminary risk level is increased and the weight of the serological marker is decreased.

[0032] It should be noted that after calculating the comprehensive risk score, it is compared with the threshold of the preliminary risk level to obtain the final risk level Six, follow-up module The follow-up plan automatically initiates a data collection request within a preset follow-up time period, obtains the time series multi-omics data of the patient from the hospital information system, and associates and stores the time series multi-omics data with the system historical prediction data to construct the personal health time evolution file of the patient. It should be noted that when the change trend indicates that the risk level has changed significantly, a re-evaluation instruction is automatically sent to the level classification module to trigger dynamic adjustment of the patient's risk level.

[0033] The follow-up time period includes three periods of 12 months, 6 months and 3 months.

[0034] It should be noted that the follow-up cycle of the high-risk level patient is 3 months, the follow-up cycle of the medium-risk level patient is 6 months, and the follow-up cycle of the low-risk level patient is 12 months.

[0035] Moreover, it should be noted that the present application can be provided as a method, an apparatus, or a computer program product. Therefore, the embodiments of the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the embodiments of the present application can take the form of a computer program product on one or more computer-usable storage media (including disks, diskettes, hard disk drives, CD-ROMs, and so on) embodying computer-readable program code.

[0036] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the method, terminal device (system), and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, so that the instructions executed by the computer or other programmable data processing terminal device produce an apparatus for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in the flowcharts and / or block diagrams.

[0037] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing terminal device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including instruction apparatus, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in the flowcharts and / or block diagrams. These computer program instructions can also be loaded into a computer or other programmable data processing terminal device, so that a series of operation steps are performed on the computer or other programmable terminal device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable terminal device provide a process for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in the flowcharts and / or block diagrams.

[0038] It is also noted that, as used herein, the terms "first", "second", and the like, merely designate a relationship or order of one entity or action to another, and do not necessarily indicate that any such entity or action have to exist or occur in any given order. The terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0039] Finally, it is noted that the foregoing description is of a preferred embodiment of the application, and that numerous changes and modifications can be made thereto without departing from the spirit and scope of the application as set forth in the appended claims.

Claims

1. A smart early screening and risk assessment system for liver cancer based on multi-omics data, characterized in that, include: Information collection module: used to collect multi-omics data from patients and construct a characteristic dataset of patients; Intelligent screening module: responsible for building and training an early risk assessment model for liver cancer, and continuously optimizing the model through algorithm optimization; Risk assessment module: This module is used to input the constructed patient feature dataset into the trained risk assessment model and calculate the patient's early liver cancer risk probability value. Probability verification module: Verifies the accuracy of the output risk probability value by comparing the calculated risk probability value with a preset risk threshold; Risk classification module: Based on the risk probability value after testing, the patient's liver cancer risk is classified into corresponding risk levels; The follow-up module is used to continuously follow up with patients and record the time-series changes in their health status.

2. The intelligent early screening and risk assessment system for liver cancer based on multi-omics data as described in claim 1, characterized in that, The information collection module is used to collect multi-omics data from patients and construct a feature dataset for patients, wherein: The multi-omics data includes abdominal ultrasound imaging data and cfDNA whole genome sequencing data; The abdominal ultrasound image data is preprocessed to extract features. The specific steps are as follows: an initial three-dimensional geometric model of the liver and its internal vascular system is generated, and an interactive three-dimensional digital twin model of the liver is generated by surface optimization of the three-dimensional geometric model. The information collection module anchors and integrates feature information extracted from genomics and radiomics into a three-dimensional liver digital twin in the form of visual elements, including color coding, highlight points, and texture mapping. The whole genome sequencing data of cfDNA were analyzed using a bioinformatics process to detect somatic mutations, copy number variations, and methylation characteristics for feature extraction. The feature dataset is generated by mapping abdominal ultrasound imaging data and cfDNA whole genome sequencing data to the specific spatial location of a three-dimensional liver mesh model using a hemodynamic model, and then rendering the three-dimensional liver mesh model in a virtual reality environment and displaying it as a fusion of visual elements.

3. The intelligent early screening and risk assessment system for liver cancer based on multi-omics data as described in claim 1, characterized in that, The intelligent screening module is responsible for constructing and training an early-stage liver cancer risk assessment model, and continuously optimizing the model through an optimization algorithm, wherein: The risk assessment model extracts specific features closely related to the progression of alcoholic liver disease to liver cancer from genomic, radiomics and clinical data of long-term drinkers, and selects a deep neural network as the basic architecture of the model, and uses feature data from a cohort of long-term alcohol abusers to fine-tune the model. The risk assessment model is optimized through a loss function and a parameter optimization strategy. The loss function uses FocalLoss as the model's loss function, and the parameter tuning strategy uses a Bayesian optimization algorithm to automatically search for and determine the model's key parameters.

4. The intelligent early screening and risk assessment system for liver cancer based on multi-omics data as described in claim 3, characterized in that, The risk assessment model uses specific features extracted from a cohort of long-term alcoholics as input, wherein: The specific features include: gene mutations and methylation markers related to alcohol metabolism and alcoholic liver injury from genomic data; quantitative texture and morphological features from radiomics data that characterize alcoholic hepatitis, liver fibrosis, and malignant transformation of regenerating nodules; and composite scores from clinical data constructed based on dynamic changes in liver function indicators and serum markers.

5. The intelligent early screening and risk assessment system for liver cancer based on multi-omics data as described in claim 1, characterized in that, The risk assessment module is used to input the constructed patient feature dataset into the trained risk assessment model to calculate the patient's early-stage liver cancer risk probability value, wherein: Before inputting the feature dataset, it is transformed using the StandardScaler normalizer to make it completely match the feature list used during the training of the risk assessment model. The calculation process of the cancer risk probability value is as follows: by calling the predict_proba method of the risk assessment model, two arrays containing liver cancer probability values ​​are obtained from the model output, and one of the arrays is extracted as the final early liver cancer risk probability value. The array has the structure [prob_class_0, prob_class_1], where prob_class_0 represents the probability of belonging to the non-liver cancer category and prob_class_1 represents the probability of belonging to the early-stage liver cancer category.

6. The intelligent early screening and risk assessment system for liver cancer based on multi-omics data as described in claim 5, characterized in that, The risk assessment module is used to input the constructed patient feature dataset into the trained risk assessment model to calculate the patient's early-stage liver cancer risk probability value, wherein: The risk assessment module uses a virtual reality visualization engine to combine the final risk probability value with image spatial features to generate and dynamically render a risk heat map at the corresponding anatomical location of the three-dimensional liver digital twin. The steps for generating the risk heatmap are as follows: the calculated overall risk probability value is decomposed and associated with multiple suspected lesion areas based on a preset weight allocation algorithm, the local risk intensity value of each area is calculated, the corresponding color attribute and transparency attribute are calculated for each area, and a semi-transparent thermal texture overlay layer is generated and updated in real time at the anatomical position corresponding to the three-dimensional liver digital twin.

7. The intelligent early screening and risk assessment system for liver cancer based on multi-omics data as described in claim 1, characterized in that, The probability verification module verifies the accuracy of the output risk probability value by comparing the calculated risk probability value with a preset verification strategy, wherein: The preset verification strategy includes a credibility assessment strategy and a feature consistency verification strategy; The credibility assessment strategy is obtained by calculating the credibility score of the Softmax layer of the risk assessment model when it outputs a risk probability value. The feature consistency test strategy extracts the top K features that contribute the most to the current risk probability value through the model interpretability engine, and compares the top K features with the known key biomarkers for early-stage liver cancer in the medical knowledge base. The probability testing module calculates the comprehensive reliability weight based on the comparison results between the credibility assessment strategy and the feature consistency testing strategy, and corrects the original risk probability value according to the comprehensive reliability weight to obtain the final risk probability value. The overall weight is calculated using a weighted geometric mean formula, as follows: ; In the formula, This indicates the degree of emphasis placed on credibility; This represents the degree of emphasis placed on feature consistency; This indicates the credibility assessment score; Indicates the feature consistency test score; Indicates the overall reliability weight; The original risk probability value is corrected using a reliability-weighted contraction method, as shown in the following formula: ； In the formula, This represents the final risk probability value; This represents the original risk probability value; This represents the preset prior probability, indicating the default risk level when the information is completely unreliable; This represents the overall reliability weight.

8. The intelligent early screening and risk assessment system for liver cancer based on multi-omics data as described in claim 1, characterized in that, The risk classification module categorizes a patient's liver cancer risk into corresponding risk levels based on the tested risk probability values, wherein: The risk classification module compares the final risk probability value with multiple predefined risk probability thresholds and classifies patients into corresponding preliminary risk levels based on the comparison results. The risk thresholds include two thresholds, T1 and T2. The preliminary risk levels include three risk levels: low, medium, and high. When the risk probability value < T1, it is classified as "low risk". When T1 ≤ risk probability value < T2, it is classified as "medium risk". When the risk probability value ≥ T2, it is classified as "high risk". The preliminary risk level is determined by weighting and fusing it with the concentration of serological markers, and the final risk level is determined based on the comprehensive score after weighted fusion.

9. The intelligent early screening and risk assessment system for liver cancer based on multi-omics data as described in claim 8, characterized in that, The preliminary risk level is determined by weighted fusion with serological marker concentrations, and the final risk level is determined based on the comprehensive score after weighted fusion, wherein: The concentrations of the serological markers are converted into marker risk scores using a linear function, and a weighted summation method is used to calculate the comprehensive risk score, as shown in the following formula: ； In the formula, Indicates the overall risk score, The weights indicating the initial risk level; The threshold indicating the initial risk level; The weight of the serological marker concentration is represented by ; the fraction of the serological marker concentration is represented by linear interpolation; where: when the initial risk level threshold is classified as low, the weight of the serological marker is increased and the weight of the initial risk level is decreased; when the initial risk level threshold is classified as medium or high, the weight of the initial risk level is increased and the weight of the serological marker is decreased.

10. The intelligent early screening and risk assessment system for liver cancer based on multi-omics data as described in claim 1, characterized in that, The follow-up module is used to develop personalized follow-up plans based on the patient's risk level and record time-series changes in their health status, wherein: The follow-up plan automatically initiates data collection requests within a preset follow-up period, obtains the patient's time-series multi-omics data from the hospital information system, and associates and stores the time-series multi-omics data with the system's historical prediction data to construct the patient's personal health time-series evolution profile. The follow-up period includes three cycles: 12 months, 6 months, and 3 months.

Images