CN114596257A

CN114596257A - Quantitative assessment method and device for liver reserve function based on medical image

Info

Publication number: CN114596257A
Application number: CN202210065825.6A
Authority: CN
Inventors: 薛峰; 夏强; 杨玉婷; 吴忌
Original assignee: Shanghai Jiaotong University; Renji Hospital Shanghai Jiaotong University School of Medicine
Current assignee: Shanghai Jiaotong University; Renji Hospital Shanghai Jiaotong University School of Medicine
Priority date: 2022-01-20
Filing date: 2022-01-20
Publication date: 2022-06-07

Abstract

The invention provides a quantitative assessment method, a device, electronic equipment and a computer-readable storage medium for liver reserve function based on medical images, wherein a liver portal phase image in preoperative epigastric enhancement CT of a liver cancer patient is adopted, and a machine learning analysis module is used for processing the liver portal phase image in combination with clinical biochemical detection data of the patient, so that a 15-minute retention rate (ICGR15) value of indocyanine green, namely a machine learning predicted value, can be automatically output, and a basis is provided for operation decision of a clinician, so that the quantitative assessment method has good practical application value. The use of the predictive model of the present application is time consuming, highly accurate, requires no purchase of millions of instruments and presents no risk of reagent allergy.

Description

Quantitative assessment method and device for liver reserve function based on medical image

Technical Field

The invention relates to the field of quantitative assessment of liver reserve function based on imaging omics development.

Background

To date, hepatectomy is an important approach to the curative treatment of a variety of primary and secondary liver tumors. However, despite significant advances in perioperative management and hepatectomy, a number of life-threatening complications still occur after hepatectomy, with hepatic failure (PHLF) being the most severe, among others. PHLF refers to severe liver damage caused by various reasons after hepatectomy, severe disorders or decompensations of liver synthesis, detoxification, excretion, biotransformation and other functions, and a group of clinical syndromes mainly manifested by blood coagulation dysfunction, jaundice, hepatic encephalopathy, hydrops in abdominal cavity and the like after the hepatectomy. The international liver surgery research group (ISGLS) in 2011 proposed a unified definition and severity grading criteria for PHLF, based on a generalized summary of several dozen studies on PHLF. PHLF was defined by ISGLS group as functional impairment of liver synthesis, secretion, detoxification and the like after surgery, mainly with Total Bilirubin (TB) and International Normalized Ratio (INR) as evaluation criteria, i.e. TB and INR increased and were greater than the upper limit of normal values on or after 5 days after surgery. Notably, changes in clinical symptoms and laboratory test metrics due to biliary obstruction should be eliminated. In addition, the ISGLS panel totaled PHLF severity into 3 grades, with PHLF grade a being defined as transient, small-scale deterioration of liver function, without any special treatment required in conventional treatments. Grade B PHLF is defined as a deviation from normal postoperative clinical management, with the patient's condition remaining in a controlled range and requiring treatment by non-invasive means, and grade C is defined as a patient with severe liver or multiple organ failure requiring invasive treatment for intervention. PHLF is a leading cause of death in patients after hepatectomy, and the occurrence of PHLF can prolong the hospitalization time of patients and affect the long-term prognosis of patients. The liver reserve function of a liver cancer patient is accurately evaluated before an operation, a perfect operation scheme is made, and operation decision is carefully made, so that the method has important significance for preventing PHLF.

Although various methods for evaluating liver reserve function have been developed in recent years, for example, in the past, liver reserve function scores including Child-Pugh score, MELD score and ALBI score are used as scoring systems for calculating liver reserve function, indocyanine green (ICG) clearance test is still the most widely used method at present. ICG clearance detection is a quantitative test which can effectively evaluate the effective blood flow of the liver and the function of liver cells, has the advantages of high excretion speed, no participation in chemical reaction in vivo, no entering into enterohepatic circulation, no lymphatic countercurrent, no excretion from other extrahepatic organs such as kidney and the like, has mature technology and is widely applied to surgical decision making.

However, the ICG clearance test requires expensive test equipment and special test reagents, and is difficult to popularize in primary hospitals, for example, the DDG-3300K liver function analyzer developed in japan is an expensive device for developing ICG clearance test, and only 15 minutes is required to obtain the value of ICGR15, the test method adopted by the apparatus is the PDD method, which is based on the principle that when two different light-absorbing substances exist in blood, a tissue is irradiated with two different wavelengths to obtain a pulse of transmitted light, and the concentration ratio of the two light-absorbing substances in blood can be determined, which is called pulse spectrophotometry, and is referred to as pulse photometry for short. In addition, intravenous ICG photosensitizing dyes may cause adverse reactions such as nausea, fever and allergic reactions. Therefore, there is a strong need to develop a simplified alternative tool to help clinicians quickly and safely obtain the value of ICGR15 during daily tasks.

Liver reserve function refers to the ability of the liver to compensate for normal physiological functions when it is injured or hit, and accurate assessment of liver reserve function is a necessary means for guiding surgeons to perform accurate hepatectomy. Indocyanine green (ICG) clearance rate detection is mainly used for quantitatively evaluating the liver reserve function clinically, but the ICG clearance rate detection needs to depend on special equipment and detection reagents and can also cause anaphylactic reaction.

Disclosure of Invention

The problems that equipment dependence exists in an ICG clearing test in the prior art, the detection is complex, adverse reactions are likely to occur, and popularization is not easy to realize are solved. A first aspect of the present application provides a method for quantitative assessment of hepatic reserve function based on imaging omics, comprising:

step S1: constructing a liver reserve function prediction model, which comprises the following steps:

step S1.1: acquiring the image omics characteristic data and preoperative clinical biochemical detection index data of a region of interest on preoperative epigastric enhanced CT portal medical images of a plurality of liver cancer patients; the liver cancer patient in this step is a patient who has been diagnosed as having liver cancer.

Step S1.2: screening imaging omics characteristics and clinical variables highly related to ICGR15 by using spearman correlation analysis;

step S1.3: for screened image omics characteristics and clinical variables highly related to ICGR15, acquiring the optimal parameter combination of the XGboost machine learning algorithm based on five-fold cross validation, namely completing the construction of a liver reserve function prediction model;

step S2: the image omics characteristic data and preoperative clinical biochemical detection index data of the region of interest on the preoperative epigastric enhanced CT portal phase image of the liver cancer patient are input into a liver reserve function prediction model, the liver reserve function prediction model automatically utilizes the XGboost machine learning algorithm, the currently screened CT image omics characteristic and clinical detection variable are automatically subjected to combined analysis, and the ICGR15 numerical value can be automatically output. Short time consumption and high accuracy. The liver cancer patient in this step is a patient who has been diagnosed as having liver cancer. A patient with an ICGR15 value of <10 can tolerate liver resections of more than three liver segments (i.e. the patient has more resectable liver volume and does not suffer from post-operative liver failure due to too little residual liver volume).

The advantages of this patent using portal images of the liver are as follows:

(1) liver tissues and other peripheral tissues can be well distinguished from the upper abdomen enhanced CT portal phase image, so that the labeling of the liver parenchymal region and the subsequent feature extraction result are more accurate;

(2) compared with other scanning time phases, the image features extracted from the upper abdominal enhanced CT portal image data are more standard and stable, accord with Imaging Biomarker Standardization Initiative (IBSI) imaging protocol (a recognized imaging standard), and are beneficial to popularization and use of research results in different medical centers;

(3) at present, a great deal of research is carried out by selecting enhanced CT portal vein images.

Further, step S1.1 comprises: (1) segmenting and labeling the region of interest of the preoperative epigastric enhancement CT portal medical image by means of 3D Slicer software in a manual drawing mode; (2) image processing is carried out by using a LoG filter; (3) and extracting the image omics features by using Pyradiomics.

Further, the image omics features are selected from one or more of liver tissue first-order features, shape features, size features, gray level co-occurrence matrix features, gray level size area matrix features, gray level running length matrix features, adjacent gray level difference matrix features and gray level dependency matrix features.

Further, step S1.2 comprises: (1) calculating an ICC value in each group and among groups based on all the obtained image omics characteristics; (2) selecting the image omics characteristics with the ICC numerical value between the groups and the ICC numerical value in the groups larger than 0.75, further evaluating the correlation between any two image omics characteristics, and if the correlation | Rho | between the two image omics characteristics is larger than 0.9, excluding any one of the two image omics characteristics; (3) evaluating correlations between ICGR15 and all parameters including pre-operative clinical biochemical detection indicators and the screened iconomics features of (2) using a spearman correlation analysis; (4) screening the image omics characteristics of | Rho | 0.3 and the clinical biochemical detection indexes of | Rho | 0.3 to obtain the image omics characteristics and clinical variables highly related to ICGR 15.

Further, the imaging omics-based quantitative assessment method for hepatic reserve function further comprises the step S3: when the input data is abnormal, sending an abnormal early warning, and tracing the abnormal input data according to the data type; and when the running state is abnormal, outputting an abnormal value and analyzing the abnormal reason.

A second aspect of the present application provides a liver reserve function quantification and assessment device based on imaging omics, comprising: the system comprises a data acquisition module, a data screening module, a model construction module, an input module, an analysis module and an output module;

the data acquisition module is used for acquiring the image omics characteristic data and preoperative clinical biochemical detection index data of the region of interest on the preoperative epigastric enhanced CT portal medical image of a plurality of liver cancer patients;

the data screening module is used for screening the imaging group characteristics and clinical variables highly related to ICGR15 by adopting the spearman correlation analysis;

the model construction module is used for acquiring the optimal parameter combination of the XGboost machine learning algorithm based on five-fold cross validation for the screened image omics characteristics and clinical variables highly related to ICGR15, namely completing construction of a liver reserve function prediction model;

the input module is used for inputting the image omics characteristic data and preoperative clinical biochemical detection index data of the region of interest on the preoperative epigastric enhanced CT portal phase image of the liver cancer patient into the liver storage function prediction model;

the output module is used for outputting the ICGR15 value.

Furthermore, the data acquisition module comprises a segmentation and labeling module, an image processing module and an image feature extraction module, wherein the segmentation and labeling module is used for segmenting and labeling the region of interest of the preoperative epigastric enhanced CT portal medical image by means of 3D Slicer software in a manual drawing mode; the image processing module is used for processing images by using a LoG filter; the image feature extraction module is used for extracting the image omics features by using Pyradiomics.

Further, the data screening module comprises a calculating module, an evaluating module and a screening module, wherein the calculating module is used for calculating the ICC numerical values in the groups and among the groups based on all the obtained image omics characteristics; the evaluation module is used for selecting the image omics characteristics of which the ICC numerical values between the groups and in the groups are both larger than 0.75, further evaluating the correlation between any two image omics characteristics, and if the correlation | Rho | between the two image omics characteristics is larger than 0.9, excluding any one of the two image omics characteristics; the evaluation module is further used for evaluating the correlation between the ICGR15 and all parameters by using a spearman correlation analysis, wherein all parameters comprise preoperative clinical biochemical detection indexes and screened imaging omics characteristics; the screening module is used for screening the imaging omics characteristics of | Rho | >0.3 and the clinical biochemical detection indexes of | Rho | >0.3 to obtain the imaging omics characteristics and clinical variables highly related to ICGR 15.

A third aspect of the present application provides an electronic device comprising: a memory, a processor and a computer program stored on the memory and executable on the processor, the computer program, when executed by the processor, implementing the steps of the above medical image-based liver reserve function quantification assessment method.

A fourth aspect of the present application provides a computer-readable storage medium having stored thereon a computer program which, when being executed by a processor, implements the steps of the above-mentioned medical image-based liver reserve function quantification assessment method.

After the technical scheme is adopted, compared with the prior art, the method has the following beneficial effects:

the invention provides an artificial intelligence model for predicting ICGR15 values based on a visual omics technology and developed through machine learning for the first time, and the model highlights that CT image characteristics are innovatively used for predicting ICGR 15. The quantitative evaluation method is mainly applied to quantitative evaluation of the liver reserve function and provides basis for surgical decisions of surgeons. The ICG clearance that this patent can replace current ICG clearance basically detects, does not rely on special check out test set and detect reagent, has established the basis for future popularization and application. The use of the predictive model of the present application is time consuming, highly accurate, requires no purchase of millions of instruments and presents no risk of reagent allergy.

A pre-operation epigastric part enhanced CT liver portal phase image of a liver cancer patient is adopted, clinical biochemical detection data of the patient are combined, and after machine learning analysis processing, a 15-minute retention rate (ICGR15) numerical value (machine learning predicted value) of indocyanine green can be automatically output. The correlation R value of the predicted value and the measured value (the measured value of the equipment) can reach 0.9, the accuracy of the model is verified in an independent verification set, and the method has clinical conversion application value.

The portal vein image with the largest cross-sectional area of the liver in the patient's epigastric CT and the clinical detection data are input into an artificial intelligence system, and the system can automatically output an ICGR15 predicted value after calculation.

The machine learning method used in this patent is an extreme gradient boost model (XGBoost). In a plurality of common integrated algorithms, the XGboost model has more excellent performance and wider application, compared with the traditional gradient boosting algorithm, the XGboost model is improved in many ways, the computing speed of the XGboost model is higher, and the XGboost model is an advanced evaluator with ultrahigh performance in classification and regression.

The method comprises the steps of carrying out feature extraction on an enhanced CT portal phase image of the upper abdomen of a patient through an image omics technology, screening effective variables by adopting a statistical method in combination with clinical biochemical detection data, and then obtaining the optimal parameter combination of a machine learning algorithm by adopting a grid search (five-fold cross validation) method, so that a predicted value of ICGR15 is obtained, a basis is provided for the operation decision of a clinician, and the method has good practical application value.

Drawings

Fig. 1 is a flowchart of a method for quantitative assessment of hepatic reserve function based on imaging omics according to an embodiment of the present disclosure;

FIG. 2 is an example of a picture obtained by ROI segmentation, in which a dotted-line frame region is a marked liver parenchymal region;

FIGS. 3A-J are the correlation (scattergrams) between the CT image features incorporated in the present invention and ICGR15, the scattergrams are the correlation analyses between the features extracted from the CT image (after screening) and ICGR15, the ordinate is the image feature value, the abscissa is the value of ICGR15, and Rho is the correlation coefficient;

FIG. 4 is a graph showing the correlation between the clinical biochemical test index incorporated in the present invention and ICGR15 (a scatter plot), wherein the vertical axis shows the test values, the horizontal axis shows the values of ICGR15, and Rho shows the correlation coefficient, and the scatter plot shows the correlation analysis between the screened clinical test index and ICGR 15. A is a scatter plot of the correlation of albumin-bilirubin score (ALBI) with ICGR 15; b is a correlation scatter plot of Total Bile Acid (TBA) and ICGR 15; c is a scatter plot of the correlation of Direct Bilirubin (DBIL) with ICGR 15; d is a scatter plot of the correlation of Prealbumin (PA) with ICGR 15;

fig. 5 is a comparison graph of the results of the prediction group and the actual measurement group in an actual application example of the present invention, and the dotted box area in the graph is the marked liver parenchymal area.

Detailed Description

The advantages of the invention are further illustrated in the following description of specific embodiments in conjunction with the accompanying drawings. It is to be understood by persons skilled in the art that the following detailed description is illustrative and not restrictive, and is not to be taken as limiting the scope of the invention.

Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the exemplary embodiments below are not intended to represent all implementations consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure, as detailed in the appended claims.

The terminology used in the present disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. As used in this disclosure and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items. The word "if" as used herein may be interpreted as "at … …" or "when … …" or "in response to a determination", depending on the context.

The imaging omics-based quantitative assessment device for the hepatic reserve function in the embodiment comprises: the system comprises a data acquisition module, a data screening module, a model construction module, an input module and an output module;

the model construction module is used for acquiring the optimal parameter combination of the XGboost machine learning algorithm based on five-fold cross validation for the screened image omics characteristics and clinical variables highly related to ICGR15, namely completing construction of a liver reserve function prediction model; the model construction process includes a training process and a testing process of the model, and the training process and the testing process are conventional technical means in the field and are not described herein again.

the output module is used for outputting the ICGR15 value.

As shown in fig. 1, the method for evaluating liver reserve function quantification by using the device for evaluating liver reserve function quantification based on machine learning includes the following steps:

1. constructing a liver reserve function prediction model, which comprises the following steps:

step 1.1: acquiring the image omics characteristic data and preoperative clinical biochemical detection index data of a region of interest on preoperative epigastric enhanced CT portal medical images of a plurality of liver cancer patients;

diagnosis of hepatocellular carcinoma (HCC) is based on: all eligible patients in the study cohort received hepatectomy, and the pathological diagnosis based on excised tumor tissue specimens was the "gold standard" for confirmed diagnosis of HCC. Pathological diagnosis of HCC is in accordance with internationally recognized recommendations for pathological diagnosis.

Diagnosis of hepatic insufficiency (PHLF) after hepatectomy:

1) the parameters for acquisition and acquisition of the upper abdominal enhanced CT image are described as follows: all patients were subjected to an upper abdominal enhancement CT scan using a multi-row helical CT scanner (Discovery CT750HD, GE Healthcare, Milwaukee, USA, for the specific model and manufacturer, respectively). Before receiving CT scanning, a patient needs to receive intravenous injection of a nonionic iodine contrast agent (370 Iopamidol, Polecoconcible pharmaceutical Co., Ltd., Shanghai, China), the injection speed is 3-4ml/s, and portal vein period liver imaging is carried out 60-70 seconds after the contrast agent is injected. Enhanced CT images of all patients were acquired at 120kVp tube voltage and reconstructed at a thickness of 1.25 mm after scanning. The enhanced CT image data is acquired from the image archiving and communication system of the affiliated renji hospital of the Shanghai university of transportation medical school and then saved in DICOM format for further analysis.

2) The ROI segmentation method is described as follows: we used the upper abdomen enhanced CT portal phase image for ROI image segmentation. Since the manual segmentation is the "gold standard" for the segmentation of CT images, the resulting image results are the most reliable. Therefore, in the invention patent, in order to ensure the accuracy of ROI region segmentation, a research team adopts a manual drawing mode to perform 2D ROI labeling of liver parenchyma by means of 3D Slicer software. The 3D Slicer software is a free-source medical image analysis and visualization software platform, has various functions, is widely applied to medical, biomedical and related imaging researches, and makes an important contribution to the development of imaging omics (Radiomics) researches. The 3D Slicer software provides rich user interaction and visual interfaces, and researchers can firstly import medical images and then carry out operations such as image segmentation, reconstruction, marking point selection and measurement. At the same time, the software itself also provides a very large number of modules, such as excellent and complete interactive segmentation and reconstruction modules, in which various operations such as image resampling, cropping, filtering, etc. required by researchers can be performed. In addition, if the function module of the 3D Slicer software cannot meet various requirements of researchers, the researchers can independently select and install the plug-in according to the requirements of the researchers, hundreds of open source plug-ins are provided on a plug-in installation platform of the software for the researchers to select, and various advanced data processing is supported. If researchers still cannot find out the functions and contents required by the research on the plug-in mounting platform, the researchers can write codes by themselves to generate the plug-ins, so that individual requirements of the researchers are met, and convenience of radio research is greatly improved. The 3D Slicer software has very wide application in the field of medical images, provides certain reference for the formulation of clinical operation schemes, for example, the application in the operation planning and operation navigation based on the medical images, and carries out 3D image segmentation and image reconstruction on an operation part and the like before the operation scheme is formulated, so that a clinician can carry out operation planning and simulation on the reconstructed tissue structure, and a reliable basis is provided for accurate operation excision.

In ROI segmentation, we select the portal image with the largest cross-sectional area of the liver, in which the liver portion is delineated along the outline of the liver parenchyma, and then perform further processing analysis of the image, for example, a picture obtained by segmentation as shown in fig. 2. To ensure the accuracy and repeatability of ROI segmentation, we calculated inter-and intra-group correlation coefficients (ICC), which is commonly used to evaluate the similarity of certain quantitative attributes between individuals with defined relatives (e.g., twins, siblings, etc.), and also to evaluate the repeatability or consistency of different assays or evaluators to the same quantitative measurement. In diagnostic tests, ICC indicators are also often used to evaluate the reproducibility of the diagnosis of the same set of test results by different researchers. Typically, the ICC value is between 0 and 1. In diagnostic tests, if the ICC value is less than 0.4, we consider the diagnostic test to be less reproducible; if the ICC value is greater than 0.75, the diagnostic test is more reproducible. In the patent of the invention, CT images of 45 patients are randomly selected, and two experienced doctors respectively label ROI so as to calculate the correlation coefficient between groups. In addition, after the first ROI labeling, one of the medical imaging physicians with abundant working experience in abdominal imaging repeats the ROI labeling process at an interval of 2 weeks, thereby calculating the intra-group correlation coefficient. The ROI segmentation for the remaining 305 patients was performed by the imaging physician with a high working experience. Through the ROI segmentation process described above, we finally obtain liver tissue region labeling of all patients.

3) The imaging characteristic extraction method comprises the following steps:

the present patent uses pyraadiomics (version 3.0.1) to extract the iconographic features from each ROI (original image, wavelet filter image and laplacian gaussian (LoG) filter image). Furthermore, we used a LoG filter for image processing with sigma values set to 2 and 4 mm, respectively. We use four different band combinations (high-high HH, high-low HL, low-high LH, and low-low LL) to generate wavelet-based texture features.

We extracted 2D CT image features using the Pyradiomics software package, and most of the defined features conform to the feature definitions described in Imaging Biomarker Standardization Initiative (IBSI) Imaging protocol. The definitions of the various features can be subdivided into a number of categories, with specific information in table 1.

TABLE 1. characteristics of imaging group

Table 1.Radiomic features

Abbreviation: GLCM, gray level cooccurrence matrix, gray level co-occurrence matrix; GLDM, gray level dependency matrix; GLRLM, gray level run length matrix; GLSZM, gray level size zone matrix, gray scale area matrix; NGTDM, neighboring gray level differential matrix.

<1 > First Order Features of liver tissue (First Order Features)

This first order feature describes the distribution of voxel intensities within the image area defined by the Mask (Mask). This feature set may be subdivided into the following sub-features: (1) energy (Energy): energy is a measure of the size of the voxel values in the image, with larger values representing the sum of the squares of these values. (2) Total Energy (Total Energy): total energy is a value of the energy characteristic, measured by the voxel volume (unit: cubic millimeters). (3) Entropy (Entropy): entropy characteristics refer to the uncertainty/randomness of image values, which measures the average amount of information needed to encode an image value. (4) Minimum (Minimum) (5) 10th percentile (10th percentile) (6) 90th percentile (90th percentile) (7) Maximum (Maximum): maximum intensity of gray within ROI. (8) Mean (Mean): average intensity of gray within ROI. (9) Median (Median): moderate gray intensity within the ROI. (10) Interquartile Range (11) Range (Range): gray value range in ROI. (12) Mean Absolute Deviation (MAD)): average distance of all intensity values from the image array mean. (13) Robust Mean Absolute Deviation (rMAD): the average distance of all intensity values from the mean value calculated over a subset of the image array having a gray level between or equal to the 10th and 90th percentiles. (14) Root Mean Square (RMS): the square root of the mean of all squared intensity values, which is another measure of the size of the image values. (15) Standard Deviation (16) Skewness (Skewness): asymmetry with respect to the mean distribution. (17) Kurtosis (Kurtosis): the "peak" of the ROI measurement distribution. (18) Variance (Variance): the mean of the squared distance of each intensity value from the mean. (19) Uniformity (Uniformity): a measure of the sum of squares of each intensity value.

< 2 > Shape Features and size Features (2D)

The set of image features contains a description of the shape and size of the 2D ROI, which features are independent of the intensity distribution of the gray scale in the ROI

(1) Gridding Surface (Mesh Surface)

(2) Pixel Surface (Pixel Surface)

(3) Perimeter (Perimeter)

(4) Perimeter to Surface ratio (Perimeter to Surface ratio)

(5) Degree of Sphericity (sphere)

(6) Spherical imbalance (Spherical aberration)

(7) Maximum 2D diameter (Maximum 2D diameter)

(8) Major Axis Length (Major Axis Length)

(9) Minor Axis Length (Minor Axis Length)

(10) Elongation (Elongation)

< 3 > Gray Level Co-occurence Matrix (GLCM) characteristics

(1) Auto-correlation (Autocorrelation): measurement of texture fineness and roughness

(2) Joint Average (Joint Average): measuring average gray scale intensity

(3) Prominent clustering (Cluster research): measuring skewness and asymmetry

(4) Cluster Shade (Cluster Shade): measuring skewness and uniformity

(5) Clustering Tendency (Cluster trending): method for measuring grouping of voxels with similar gray values

(6) Contrast (Contrast): measuring local intensity variations, tending to deviate from diagonal values

(7) Correlation (Correlation):

(8) mean Difference (Difference Average): the relationship between the number of occurrences of pairs with similar intensity values and the number of occurrences of pairs with different intensity values is measured.

(9) Difference Entropy (Difference Entropy): a measure of randomness/variability of neighborhood intensity value differences.

(10) Variance of Difference (Difference Variance): measurement of heterogeneity

(11) Joint Energy (Joint Energy): measurement of homogeneity patterns in images

(12) Joint Entropy (Joint Entropy): measure of neighborhood intensity value randomness/variability

(13) Information Measure of Correlation 1(information Measure of Correlation (IMC) 1): evaluating correlations between class 1 probability distributions

(14) Information measurement of correlation 2(IMC 2): evaluating correlations between class 2 probability distributions

(15) Contrast Moment (Inverse Difference Moment (IDM)): measurement of local homogeneity in images

(16) Maximum Correlation Coefficient (MCC)): texture complexity measure

(17) Moment of contrast normalization (Inverse Difference Moment Normalized (IDMN)): measurement of local inhomogeneities in images

(18) Inverse Difference (ID)): another method for measuring local homogeneity of image

(19) Inverse Difference normalization (Inverse Difference Normalized (IDN)): another method of image local homogeneity measurement (after data normalization)

(20) Inverse Variance (Inverse Variance):

(21) maximum Probability (Maximum Probability) of occurrence of the most dominant pair with neighboring intensity values

(22) And Average (Sum Average): measuring the relationship between the number of occurrences of pairs with lower intensity values and the number of occurrences of pairs with higher intensity values

(23) And Entropy (Sum entry): the sum of the neighborhood intensity value differences.

(24) Sum of Squares (Sum of Squares): measuring the distribution of pairs of adjacent intensity levels of the mean intensity level in a GLCM

< 4 > Gray Level Size Zone Matrix (GLSZM) characteristics

(1) Small Area Emphasis (SAE): measurement of Small Area distribution

(2) Wide Area Emphasis (Large Area Emphasis (LAE)): measurement of Large Area distribution

(3) Gray Level Non-uniformity (GLN) measuring the variability of Gray Level intensity values in an image

(4) Gray Level Non-Uniformity normalization (GLNN) measuring the variability of Gray intensity values in an image (after normalization)

(5) Size-Zone Non-uniformity (SZN) measuring the variability of Size-Zone volume in images

(6) Size-Zone Non-Uniformity normalization (SZNN) measuring the variability of the Size-Zone volume in an image (after normalization)

(7) Zone Percentage (ZP)): measuring roughness of texture

(8) Gray Level Variance (GLV): measuring the change of regional Gray Level intensity

(9) Zone differentiation (Zone Variance (ZV): measurement of the difference in size and volume of zones

(10) Zone Entropy (ZE)) measuring uncertainty/randomness of Zone size and gray level distribution

(11) Low Gray Level Zone Emphasis (LGLZE) measuring the distribution of Low Gray Level Zone size zones

(12) High Gray Level Zone Emphasis (HGLZE) measuring distribution of High Gray Level Zone area

(13) Small Area Low Gray Level Emphasis (SALGLE): measuring the proportion of Small Area areas and Low Gray values jointly distributed in the image

(14) Small Area High Gray Level Emphasises (SAHGLE): measuring the ratio of Small Area and High Gray Level value distributed in image

(15) Large Area Low Gray Level Emphasis (LALGLE)) measuring the proportion of the combined distribution of Large Area and Low Gray values in an image

(16) Large Area High Gray Level Emphasis (LAHGLE): measuring the ratio of Large Area and High Gray Level in the image

< 5 > Gray Level Run Length Matrix (GLRLM) feature

(1) Short-stroke Emphasis (Short Run Emphasis (SRE)): measuring short stroke length

(2) Long stroke Emphasis (Long Run Emphasis (LRE)): measuring long stroke length

(3) Grayscale Non-uniformity (Gray Level Non-uniformity (gln)): measuring similarity of intensity values of gray levels in an image

(4) Gray Level Non-Uniformity Normalized (GLNN)): measuring the similarity of intensity values of gray levels in an image (after normalization)

(5) Run Length Non-uniformity (rln)): measuring similarity of image run lengths

(6) Run Length Non-Uniformity normalization (RLNN): measuring similarity of image run lengths (after normalization)

(7) Run Percent (RP)): measuring roughness of texture

(8) Gray Level Variance (GLV): measuring variance of gray scale intensity of travel

(9) Run Variance (RV)): measuring the variance of the stroke length

(10) Run Entropy (RE)): measuring uncertainty/randomness of run length and gray level distribution

(11) Low Gray Level Run Emphasis (LGLRE)): measuring the distribution of low grey values

(12) High Gray Level Run Emphasis (HGLRE)): measuring distribution of high gray values

(13) Short Run Low Gray Level Emphasis (SRLGLE)): combined short run length distribution for low gray scale value measurement

(14) Short Run High Gray Level Emphasis (SRHGLE): measurement of the joint distribution of Short Run lengths of High Gray values

(15) Long-stroke Low Gray Level Emphasis (LRLGLE): measurement of the Joint distribution of Long-stroke lengths for Low Gray values

(16) Long Run High Gray Level Emphasis (LRHGLE): measuring the combined distribution of Long Run lengths of High Gray Level

< 6 > Neighboring Gray Tone Difference Matrix (NGTDM) feature

Quantifying the difference between a grey value and the average grey value of its adjacent areas

< 7 > Gray Level Dependency Matrix (GLDM) characteristics

(1) Weak correlation advantage (Small dependency ions (SDE)): measurement of the distribution of the weak correlation

(2) Strongly correlated dominance (Large dependency ions (LDE)) by measuring the distribution of strong correlations

(3) Gray Level Non-uniformity (GLN) measuring the similarity of Gray Level intensity values in an image

(4) Correlation Non-uniformity (DN) by measuring image-correlated similarity

(5) Correlation Non-Uniformity normalization (DNN) by measuring image-dependent similarity (normalization)

(6) Gray Level Variance (GLV) measuring the Variance of image Gray levels

(7) Correlation variation (dependency Variance (DV)): Variance of correlation size of measured image

(8) Entropy of Dependence (Dependence Encopy (DE))

(9) Low Gray Level Emphasis (LGLE) measuring the distribution of Low Gray Level values

(10) High Gray Level Emphasis (HGLE) measuring the distribution of High Gray Level values

(11) Low-correlation Low Gray Level Emphasis (SDLGLE): measuring the joint distribution of Low-grey value weak correlation

(12) Low-correlation High-Gray Level Emphasis (SDHGLE): measurement of joint distribution of High-Gray Level weak correlation

(13) Measuring the Joint distribution of the Low Gray Level Emphasis (LDLGLE) of Low Gray Level

(14) Measuring the joint distribution of High Gray value strong correlation

Step 1.2: screening imaging omics characteristics and clinical variables highly related to ICGR15 by using spearman correlation analysis;

in each preoperative epigastric enhancement CT portal phase image (model development cohort), the present patent extracted a total of 660 Radiomics features from each annotated liver ROI, but not all CT image features were reproducible, so to evaluate the intra-and inter-observer reproducibility of the imagery omics features, we first calculated intra-and inter-group ICC values. The invention patent only selects the CT image characteristics with ICC value greater than 0.75 between groups and the ICC value between groups for further analysis. Next, in order to screen out the effective imaging omics (Radiomics) features required by the present invention, we used Spearman correlation analysis (correlation coefficient Rho) to evaluate the correlation between any two CT image features. An absolute value of Rho (| Rho |) >0.9 indicates that two Radiomics features are highly correlated. If the correlation | Rho | between two Radiomics features >0.9, we exclude either feature.

Finally, as shown in FIGS. 3-4, we evaluated the correlation between ICGR15 and all parameters (including the selected Radiomics signature and the collected clinical parameters described above) using Spearman correlation analysis. Since | Rho | ≦ 0.3 generally represents no correlation between the two variables, the final | Rho | >0.3 variable will be used in the development of the prediction model.

The invention patent uses SPSS (version 22.0, SPSS, Inc., Chicago, IL) and GraphPad Prism (version 8.0) software to carry out statistical analysis on all data.

To evaluate the correlation between ICGR15 and CT imaging characteristics, clinical parameters, a Spearman correlation analysis was used (correlation coefficient is Rho), and the variable, | Rho | >0.3 was selected. All significance tests were two-sided tests, with a P value less than 0.05 indicating significant statistical significance.

Step 1.3: for screened image omics characteristics and clinical variables highly related to ICGR15, acquiring the optimal parameter combination of the XGboost machine learning algorithm based on five-fold cross validation, namely completing the construction of a liver reserve function prediction model;

illustratively, the code for the model building process is as follows:

1. guide warehouse

#windows

pip install xgboost # install xgboost library

pip install-update xgboost # update xgboost library

from sklearn.preprocessing import StandardScaler

import xgboost as xgb

from xgboost import XGBRegressor as XGBR

from sklearn.ensemble import RandomForestRegressor as RFR

from sklearn.linear_model import LinearRegression as LinearR

from sklearn.datasets import load_boston

from sklearn.model_selection import KFold,cross_val_score as CVS

from sklearn.metrics import mean_squared_error as MSE

import pandas as pd

import numpy as np

import matplotlib.pyplot as plt

from time import time

import datetime

from sklearn.preprocessing import StandardScaler

from sklearn.model_selection import cross_val_score

from sklearn.model_selection import train_test_split

from sklearn import metrics

from sklearn.metrics import r2_score

2. Importing data

dataset＝pd.read_csv('E:\\modelconstruction.csv')

print(dataset.head())

X＝dataset.iloc[:,1:15].values

y＝dataset['ICG'].values

X＝pd.DataFrame(X)

y＝pd.DataFrame(y)

X

X.shape

y.shape

X.head()

3. Modeling

X_train,X_test,y_train,y_test＝train_test_split(X,y,test_size＝0.3,random_state＝50)

# standardization

sc＝StandardScaler()

X_train＝sc.fit_transform(X_train)

X_test＝sc.transform(X_test)

def plot_learning_curve(estimator,title,X,y,

ax is None, # selects subgraph

The longitudinal coordinate of ylim is set as the range of value

cv-None, # cross validation

Thread to be used by None # is set

):

clf＝XGBR(n_estimators＝10,max_depth＝7,subsample＝0.8,learning_rate＝0.3)

x_train＝X_train

x_test＝X_test

clf.fit(x_train,y_train)

4. Model prediction (test set)

train_predict＝clf.predict(x_train)

test_predict＝clf.predict(x_test)

5. Evaluation of model Effect Using r2

print('The r2 of model is:',metrics.r2_score(y_train,train_predict))

print('The r2 of the model is:',metrics.r2_score(y_test,test_predict))

6. Feature importance assessment

clf.feature_importances_

7. Grid search related parameters

# grid search with n _ estimator determined

The # imports trellis reference function from sklern library

from sklearn.model_selection import GridSearchCV

Parameter range defined by #

learning_rate＝[0.01,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9]

subsample＝[0.4,0.5,0.6,0.7,0.8,0.9]

colsample_bytree＝[0.5,0.6,0.7,0.8,0.9]

max_depth＝[5,6,7,8,9]

parameters＝{'learning_rate':learning_rate,

'subsample':subsample,

'colsample_bytree':colsample_bytree,

'max_depth':max_depth

}

model＝XGBR()

# grid search

clf＝GridSearchCV(model,parameters,cv＝5,scoring＝'r2',verbose＝1,n_jobs＝-1)

clf＝clf.fit(X_train,y_train)

Optimal parameter combination after # trellis search

clf.best_params_

Step 2: the image omics characteristic data and preoperative clinical biochemical detection index data of the region of interest on the preoperative epigastric enhanced CT portal phase image of the liver cancer patient are input into a liver reserve function prediction model, and the ICGR15 numerical value can be automatically output.

FIG. 5 is a graph comparing the results of the prediction group and the actual measurement group.

1. Prediction group:

(1) inputting portal phase image with largest liver cross section area in upper abdomen enhanced CT to liver reserve function prediction model

After the enhanced CT portal image of the upper abdomen of a patient is input into an artificial intelligent system, the system automatically adopts a feature extractor module in the radiomics to extract features, and a plurality of effective CT image features are automatically selected.

(2) Inputting clinical detection data into liver reserve function prediction model

The preoperative clinical detection data of the patient are input into a manual intelligent system, and the system automatically screens and selects a plurality of effective clinical indexes.

(3) Automatic output ICGR15 predicted value of liver reserve function prediction model

And (4) processing and analyzing the screened CT image characteristic numerical values and clinical data by an artificial intelligence system, and outputting an ICGR15 predicted value by the system.

In practice, the ICGR15 value can be obtained only in 15-30 seconds when the prediction model of the application is used.

2. And (3) actual measurement group:

using DDK300 liver function analyzer manufactured in Japan, it took 15 minutes to obtain ICGR15 value.

As can be seen from the comparison between the prediction group and the actual measurement group, the prediction model of the application has the advantages of short time consumption, high accuracy, no purchase of millions of instruments and no reagent allergy risk.

Another embodiment of the present application may be an electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the above-mentioned method for quantitative assessment of liver reserve function based on medical images. The electronic device includes, but is not limited to, a user device, a network device, or a device formed by integrating a user device and a network device through a network.

Another embodiment of the present application may be a computer-readable storage medium having a computer program stored thereon, which when executed by a processor implements the steps in the above-mentioned medical image-based liver reserve function quantification assessment method. The computer readable storage medium may be a tangible device that can hold and store the instructions for use by the instruction execution device. The computer-readable program instructions may be downloaded from a computer-readable storage medium to a respective computing/processing device, or to an external computer or external storage device over a network, such as the internet, a local area network, a wide area network, and/or a wireless network, to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other devices to produce a computer-implemented process such that the instructions which execute on the computer, other programmable data processing apparatus, or other devices implement the functions/acts specified in the flowchart block or blocks.

It should be noted that the embodiments of the present invention have been described in terms of preferred embodiments, and not by way of limitation, and that those skilled in the art can make modifications and variations of the embodiments described above without departing from the spirit of the invention.

Claims

1. A quantitative assessment method for liver reserve function based on imaging omics is characterized by comprising the following steps:

step S1.1: acquiring the image omics characteristic data and preoperative clinical biochemical detection index data of a region of interest on preoperative epigastric enhanced CT portal medical images of a plurality of liver cancer patients;

step S2: the image omics characteristic data and preoperative clinical biochemical detection index data of the region of interest on the preoperative epigastric enhanced CT portal phase image of the liver cancer patient are input into a liver reserve function prediction model, and the ICGR15 numerical value can be automatically output.

2. The medical image-based quantitative assessment method for liver reserve function according to claim 1, wherein the step S1.1 comprises: (1) segmenting and labeling the region of interest of the preoperative epigastric enhancement CT portal medical image by means of 3D Slicer software in a manual drawing mode; (2) image processing is carried out by using a LoG filter; (3) and extracting the image omics features by using Pyradiomics.

3. The medical image-based quantitative assessment method of hepatic reserve function according to claim 2, wherein the image omics features are selected from one or more of liver tissue first order features, shape features, size features, gray level co-occurrence matrix features, gray level size region matrix features, gray level run length matrix features, adjacent gray level difference matrix features, and gray level dependency matrix features.

4. The medical image-based quantitative assessment method for liver reserve function according to claim 1, wherein the step S1.2 comprises: (1) calculating an ICC value in each group and among groups based on all the obtained image omics characteristics; (2) selecting the image omics characteristics with the ICC numerical value between the groups and the ICC numerical value in the groups larger than 0.75, further evaluating the correlation between any two image omics characteristics, and if the correlation | Rho | between the two image omics characteristics is larger than 0.9, excluding any one of the two image omics characteristics; (3) evaluating correlations between ICGR15 and all parameters including pre-operative clinical biochemical detection indicators and the screened iconomics features of (2) using a spearman correlation analysis; (4) screening the image omics characteristics of | Rho | 0.3 and the clinical biochemical detection indexes of | Rho | 0.3 to obtain the image omics characteristics and clinical variables highly related to ICGR 15.

5. The medical image-based quantitative assessment method for liver reserve function according to any one of claims 1 to 4, further comprising the step S3: when the input data is abnormal, sending an abnormal early warning, and tracing the abnormal input data according to the data type; and when the running state is abnormal, outputting an abnormal value and analyzing the abnormal reason.

6. A liver reserve function quantification assessment device based on imaging omics, characterized by comprising: the system comprises a data acquisition module, a data screening module, a model construction module, an input module and an output module;

the input module is used for inputting the image omics characteristic data of the interested region on the preoperative epigastric enhanced CT portal phase image of the liver cancer patient and the preoperative clinical biochemical detection index data into the liver reserve function prediction model,

the output module is used for outputting the ICGR15 value.

7. The imaging omics-based quantitative assessment device for liver reserve function according to claim 6, wherein the data acquisition module comprises a segmentation labeling module, an image processing module and an image feature extraction module, and the segmentation labeling module is used for performing segmentation and labeling of a region of interest on the preoperative epigastric enhanced CT portal medical image by means of a 3D Slicer software in a manual delineation manner; the image processing module is used for processing images by using a LoG filter; the image feature extraction module is used for extracting the image omics features by using Pyradiomics.

8. The imaging omics-based quantitative assessment device for liver reserve function according to claim 6, characterized in that the data screening module comprises a calculation module, an assessment module and a screening module, the calculation module is used for calculating the ICC values within and between groups based on all the obtained imaging omics characteristics; the evaluation module is used for selecting the image omics characteristics of which the ICC numerical values between the groups and in the groups are both larger than 0.75, further evaluating the correlation between any two image omics characteristics, and if the correlation | Rho | between the two image omics characteristics is larger than 0.9, excluding any one of the two image omics characteristics; the evaluation module is further used for evaluating the correlation between the ICGR15 and all parameters by using a spearman correlation analysis, wherein all parameters comprise preoperative clinical biochemical detection indexes and screened imaging omics characteristics; the screening module is used for screening the imaging omics characteristics of | Rho | >0.3 and the clinical biochemical detection indexes of | Rho | >0.3 to obtain the imaging omics characteristics and clinical variables highly related to ICGR 15.

9. An electronic device, comprising: memory, a processor and a computer program stored on the memory and executable on the processor, the computer program, when executed by the processor, implementing the steps of the medical image-based liver reserve function quantification assessment method according to any one of claims 1-5.

10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method for medical image-based quantitative assessment of liver reserve according to any one of claims 1 to 5.