CN112017791B - System for determining prognosis condition of liver cancer patient based on artificial neural network model - Google Patents
System for determining prognosis condition of liver cancer patient based on artificial neural network model Download PDFInfo
- Publication number
- CN112017791B CN112017791B CN202010337089.6A CN202010337089A CN112017791B CN 112017791 B CN112017791 B CN 112017791B CN 202010337089 A CN202010337089 A CN 202010337089A CN 112017791 B CN112017791 B CN 112017791B
- Authority
- CN
- China
- Prior art keywords
- liver cancer
- input
- output
- detection module
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/80—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for detecting, monitoring or modelling epidemics or pandemics, e.g. flu
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Public Health (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Databases & Information Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Pathology (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Investigating Or Analysing Biological Materials (AREA)
Abstract
The invention provides a device for judging prognosis condition of a liver cancer patient based on an artificial neural network model, which comprises the following components: (1) An input module configured to input tumor characteristic information of a patient to be diagnosed with liver cancer; (2) The prognosis judging module is arranged to comprise an artificial neural network model, and the artificial neural network model is used for calculating the prognosis condition of the liver cancer patient to be judged based on the input information; (3) An output module configured to output the resulting prognosis. The invention also provides a method for grouping prognosis conditions of liver cancer patients by using the device.
Description
Technical Field
The invention relates to the field of medicine, in particular to a system for judging the progress of a liver cancer patient, and more particularly relates to an artificial neural network model-based personalized system for judging the survival of the liver cancer patient without progress.
Background
Statistics indicate that liver cancer-induced death accounts for the fourth of all tumors worldwide. Although the discovery of various early monitoring indexes and regular image examination enable early diagnosis of liver cancer, the therapeutic level of liver cancer is gradually improved, but the number of life loss caused by 2016 liver cancer has been increased from the third position in 2006 to the second position in all tumors due to population growth and aging, and research predicts that the number is continuously increased by 69.6% compared with the current number in 2040 years. The new onset and death cases of liver cancer in China account for over 50 percent of the whole world, and most of the cases are caused by HBV infection.
Various treatment modalities have been developed for liver cancer of different stages, but recurrence and metastasis of the treated patients remain a serious clinical problem. Research reports show that the liver transplantation of the liver cancer patient still has 8-20% of recurrence rate, and the median survival time of the patient after the recurrence is only 8.7 months; the rate of early-stage relapse (within 1 year) in the liver of a patient after liver cancer resection is 19-38%, the rate of late-stage relapse (more than 1 year) is 41-70%, and the 3-year survival rate of the patient after relapse is less than 35%; radio frequency therapy is used as a replacement therapy for unresectable liver cancer, and the 3-year distant recurrence rate of patients after treatment is 63.3%. It can be determined that disease progression in liver cancer patients remains a significant cause of their death.
The current models for predicting recurrence of patients with liver cancer mostly focus on resection or liver transplantation, and for other treatment modes, the models for recurrence or progression of liver cancer are few. In addition, for the whole liver cancer population, the current prognosis prediction is mainly the final outcome, namely the prediction of Overall Survival (OS), for example, BCLC staging, TNM staging, etc., which are staging systems for evaluating the Overall Survival of liver cancer patients.
Indeed, progression-Free Survival (PFS) is more often used as a surrogate for overall Survival in many solid tumors. PFS has clear advantages over OS: on the one hand, results for PFS were obtained earlier within the same follow-up time; on the other hand, the evaluation of the outcome by the PFS is not influenced by the risk of competitive death, and is more objective and accurate. However, few predictive studies are currently conducted on the progression-free survival of the entire population of liver cancer patients.
An Artificial Neural Network (ANN), which is a kind of machine learning, is essentially a mathematical model driven based on a biological Neural system (similar to brain neuron processing information). ANN has found numerous applications in medical decision-making, with the advantage that statistical analysis of complex relationships, linear, logical or non-linear, can be performed. At present, an early warning model which can carry out non-progress survival on liver cancer patients by utilizing the technical advantages of an artificial neural network is not available in the field.
Disclosure of Invention
The invention aims to solve the technical problem that an early warning method capable of judging the non-progress survival of a liver cancer patient, particularly an HBV-HCC (Hepatitis B Virus-related Hepatocellular Carcinoma) patient in 1 year is established by identifying influencing factors related to the non-progress survival of the liver cancer patient and constructing an ANN model by utilizing the influencing factors, so that a high-risk group which is possible to have the disease progress is accurately identified, and early intervention is realized to reduce the death rate of the patient.
In view of the above technical problems, the present invention aims to provide a device for determining prognosis of a patient with liver cancer based on an artificial neural network model, which can be used to determine prognosis of the patient with liver cancer; in addition, the invention also aims to provide a method for grouping prognosis situations of liver cancer patients by adopting the device, namely, dividing the liver cancer patients into high, middle or low risks on the basis of determining the prognosis situations so as to take further treatment measures to the patients in time.
The technical scheme of the invention is as follows:
in one aspect, the present invention provides an apparatus for determining prognosis of a patient with liver cancer based on an artificial neural network model, the apparatus comprising:
(1) An input module configured to input information of a patient to be diagnosed with liver cancer as follows:
the number of tumors is 1 when the number is more than or equal to 2 and 0 when the number is less than 2;
the size of the tumor is 1 when the size is more than or equal to 5cm and 0 when the size is less than 5 cm;
for the portal cancer embolus, "1" is input when combined and "0" is input when not combined;
AFP, inputting 1 when the concentration is more than or equal to 400ng/mL, and inputting 0 when the concentration is less than 400 ng/mL;
white blood cell count, input at 10 9 Number of counts/L;
inputting a ratio of the neutrophil lymphocyte;
gamma-GGT, inputting the numerical value in U/L;
ALP, inputting a numerical value in U/L;
albumin, the value in g/L is input;
CD 4T cell count, entering values in counts per μ L;
creatinine, inputting a value in mu mol/L;
(2) The prognosis judging module is set to comprise an artificial neural network model, the artificial neural network model is used for calculating the prognosis condition of the patient with the liver cancer to be judged based on input information, the artificial neural network model is of a multilayer perceptron structure and is constructed by Mathemica 11.1.1 software, the input layer is 11 input neurons, the output layer is 2 output neurons and comprises 3 hidden layers, wherein the 1 st hidden layer and the 3 rd hidden layer adopt Ramp as activation functions, the 2 nd hidden layer adopts an ArcTan function as an activation function, and the output layer adopts a SoftMax function as an activation function;
(3) An output module configured to output the resulting prognosis.
Preferably, the artificial neural network model is built by the following process:
obtaining information of tumor number, tumor size, portal cancer thrombus, AFP, leucocyte count, neutrophil lymphocyte proportion, gamma-GGT, ALP, albumin, CD 4T cell count and creatinine of a liver cancer patient, using the information as input neurons of an input layer, and using the information as output neurons of an output layer according to the number of months of non-progress survival and whether the non-progress survival occurs within 1 year, training the artificial neural network, and the training comprises the following steps: the input propagates from the first layer of neurons to each layer of neurons until an output is generated; comparing the output value with the expected output value, if there is a difference between the two values, generating an error signal, and then using a Back Propagation (BP) method to change the weight of the inter-neuron connection to reduce the overall error of the network; until the error between the value of the output and the value of the desired output decreases to a minimum.
Specifically, the process of establishing the artificial neural network of the present invention is described in the "detailed description" section and is illustrated in fig. 1.
Preferably, the prognosis condition output by the output module is the survival probability of no progress within 1 year of the patient with liver cancer to be distinguished. Preferably, the liver cancer patient is hepatitis B virus-related liver cancer, preferably primary liver cancer.
Preferably, the apparatus further comprises:
(4) A detection module, the detection module comprising:
a tumor number detection module;
a tumor size detection module;
a portal cancer embolus detection module;
an AFP detection module;
a leukocyte detection module;
a neutrophil lymphocyte proportion detection module;
a gamma-GGT detection module;
an ALP detection module;
an albumin detection module;
a CD 4T cell detection module;
and a creatinine detection module.
The device for judging the prognosis condition of the liver cancer patient based on the artificial neural network model can be computer equipment. The prognostic discrimination module can be included in the memory of the computer device; the computer device further comprises a processor capable of running a computer program of the artificial neural network model.
In another aspect, the present invention also provides a method for prognosis grouping of a liver cancer patient using the device, the method comprising:
(1) Obtaining the information of the following indexes of the liver cancer patient to be grouped:
the number of tumors; tumor size; portal cancer emboli; AFP; counting white blood cells; the neutrophil lymphocyte fraction; gamma-GGT; ALP; albumin; CD 4T cell count; creatinine;
(2) Inputting the information obtained in the step (1) into an input module of the device:
the number of tumors is input to be 1 when the number is more than or equal to 2 and is input to be 0 when the number is less than 2;
the size of the tumor is 1 when the size is more than or equal to 5cm and 0 when the size is less than 5 cm;
for the portal cancer embolus, "1" is input when combined and "0" is input when not combined;
AFP, inputting 1 when the concentration is more than or equal to 400ng/mL, and inputting 0 when the concentration is less than 400 ng/mL;
white blood cell count, input at 10 9 Number of counts/L;
inputting the proportion of the neutrophil lymphocyte and the numerical value;
gamma-GGT, inputting the numerical value in U/L;
ALP, inputting a numerical value in U/L;
albumin, input in the value in g/L;
CD 4T cell count, input number in counts/μ L;
creatinine, input value in mu mol/L;
(3) And calculating the prognosis condition of the liver cancer patient to be distinguished by using the artificial neural network model included in the prognosis distinguishing module, and outputting the prognosis condition from an output module.
Preferably, the prognosis is the survival probability of no progress within 1 year of the patient with liver cancer to be discriminated. Preferably, the liver cancer patient is hepatitis B virus-related liver cancer, preferably primary liver cancer.
Preferably, the method further comprises:
(4) Determining the liver cancer patients with the obtained probability of less than or equal to 32.8 percent as a high risk group with poor prognosis, determining the liver cancer patients with the obtained probability of less than 74.6 percent but more than 32.8 percent as a common middle risk group with good prognosis, and determining the liver cancer patients with the obtained probability of more than or equal to 74.6 percent as a low risk group with good prognosis.
The grouping method provided by the invention is not used for diagnosing diseases, but can provide intermediate results or auxiliary information for clinical monitoring and treatment of liver cancer patients.
2890 hepatitis B related primary liver cancer patients admitted to a Beijing Di Tan hospital affiliated to the university of capital medical science are included in the invention and are randomly grouped, wherein 1480 patients are training sets, and 637 patients are verification sets. After independent influence factors influencing 1-year progress of the liver cancer patient are analyzed by adopting Cox multifactor regression, an artificial neural network model is constructed. The area under the ROC curve, the C-index and the calibration curve are adopted to evaluate the discrimination and the calibration of the model, so that the ANN model disclosed by the invention has good individualized prediction performance and is beneficial to evaluating the progress of HCC patients in clinical practice.
Therefore, the artificial neural network prediction model suitable for the individual patients is constructed by applying the machine learning method for the first time, and the probability of 1-year progression-free survival of HCC patients can be calculated. This model tool can form a simple, easy-to-operate calculator on a website that integrates the oncology features of HCC patients: tumor size, number, portal cancer emboli, AFP, creatinine, liver function (ALB, γ -GGT, and ALP), inflammation indices (NLR and WBC), and immune indices (CD 4T cell count).
The C index of the ANN prediction model is more than 0.7 in the training set and the verification set, and the result is reliable. In addition, the ANN model constructed by the invention is compared with models such as BCLC, TNM, okuda, CUPI, CLIP, JIS, ALBI and the like, and the AUC value and the C-index of the ANN model are higher than those of other models (p is less than 0.0001), and the same is true in sub-layers of different genders, different ages, different AFP levels, different child grades and different treatment modes. These demonstrate that the ANN model of the present invention has better clinical utility than other models.
The invention proves that the information of the 11 indexes of the liver cancer patient is synthesized, the trained ANN model is adopted, the liver cancer patient can be effectively judged for prognosis, and the patients are grouped based on the prognosis, so that useful intermediate results or auxiliary information is provided for the monitoring and treatment mode selection of the liver cancer patient, the monitoring and the treatment mode selection of the liver cancer patient are realized by clinicians according to the risk level, and the method has important significance for improving the treatment effect of the patient, saving social resources and the like.
Drawings
Embodiments of the invention are described in detail below with reference to the attached drawing figures, wherein:
FIG. 1 shows a grouping scheme of liver cancer patients.
Fig. 2 shows survival over time as tracked into cohort liver cancer patients, where 2A: progression-free survival rate; 2B: overall survival rate.
FIG. 3 shows an artificial neural network model established by the present invention for predicting 1-year progression-free survival of a liver cancer patient.
Fig. 4 shows a calibration curve for predicting the survival probability of 1 year without progress of the training set and the validation set of liver cancer patients by using the artificial neural network model established by the invention, wherein 4A: training a set; 4B: and (5) verifying the set.
Fig. 5 shows a survival analysis comparison of 1-year progression-free survival of patients in high, medium and low risk groups differentiated by the artificial neural network model established in the present invention, wherein 5A: training a set; 5B: and (5) verifying the set.
Fig. 6 shows K-M survival curves for high CD4 patients compared to low CD4 patients, where 6A: PFS comparisons of two groups of patients in the training set; 6B: comparing PFS of two groups of patients in a verification set; 6C: PFS comparison of two groups of patients in training set data 1; 6D: PFS comparisons of two groups of patients in training set data 2; 6E: PFS comparisons of two groups of patients in training set data 3.
Detailed Description
The invention is illustrated below with reference to specific examples. It will be understood by those skilled in the art that these examples are for illustrative purposes only and are not intended to limit the scope of the present invention in any way.
The experimental procedures in the following examples are conventional unless otherwise specified. The reagent materials and the like used in the following examples are commercially available products unless otherwise specified.
In this study:
data analysis used SPSS 21.0 (IBM Corp, armonk, NY, united States).
Quantitative data comparison was performed using either the t-test or the Mann-Whitney U-test.
Qualitative data comparison using Fisher's exact or chi 2 And (6) checking.
COX one-factor and multifactorial analysis (forward, maximum partial likelihood estimation likelihood ratio test) screened risk factors affecting the progression-free survival of liver cancer patients for 1 year.
The ANN model is constructed by adopting Mathemica 11.1.1 software.
The finally established model is compared with the following prognosis models respectively in the training set and the verification set and in different subgroups:
Tumor-Node-Metastasis (TNM), see Minagawa M, ikai I, matsuyama Y, et al.Staging of hepatocellular Carcinoma: assessment of the Japanese TNM and AJCC/UICC TNM systems in a family of 13, 772 patents in Japan [ J ]. Annals of surgery,2007, 245 (6): 909;
barcelona client Cancer (BCLC), see Forner A, reig M E, de Lope C R, et al Current protocols for standing and traffic: the BCLC update and future promoters [ C]//Seminars in liver disease.
Thieme Medical Publishers，2010，30(01)：061-074；
Okuda, see Okuda K, obata H, nakajima Y, et al, prognosis of Primary heparin cartinosoma [ J ]. Hepatology,1984,4 (S1): 3S-6S;
cancer of the Liver Italian Program (CLIP), see Cancer of the Liver Italian Program (CLIP) investors.A new qualitative system for hepatocellular carriers: a retrospecific study of 435 properties [ J ]. Hepatology,1998, 28 (3): 751-755;
a Chinese University diagnostic Index (CUPI), see Leung T W T, tang A M Y, zee B, et al.Construction of the Chinese University diagnostic Index for a heterocyclic ring and complex with the TNM regulating system, the Okuda regulating system, and the Cancer of the Liver Italian Program regulating system: a study based on 926 substrates, [ J ]. Cancer,2002, 94 (6): 1760-1769;
japan Integrated Staging (JIS), see Kudo M, chung H, osaki Y.lignotic standing system for a heterocyclic carbon (CLIP score): its value and limitations, and a porosity for a new stabilizing system, the Japan Integrated stabilizing Score (JIS Score) [ J ]. Journal of gastroenterology,2003, 38 (3): 207-215;
albumin-bilirubin (ALBI) fractionation, see Pinato D J, sharma R, allara E, et al, the ALBI grades programs object aspect fire regenerative evaluation of nuclear BCLC stage of a nuclear reactor [ J ]. Journal of nuclear, 2017, 66 (2): 338-346.
The discrimination of the model was tested using the area under the ROC curve (AUC) and the consistency index (C-index). To check the calibration of the model, the Hosmer-Lemeshow test was applied and a calibration curve was drawn. Data analysis used R language version 3.3.2, rms, survival, rmda package. All tests with p < 0.05 considered statistically different.
Example 1Baseline characterization and survival analysis of patients enrolled in the study
The retrospective continuous inclusion of 2890 patients who were first diagnosed with hepatitis b-related primary liver cancer who were admitted to the Beijing Ditan hospital affiliated with the university of capital medicine from month 1 to month 2016 at month 12 in 2008. The study was approved by the ethics committee of the altar hospital.
Inclusion criteria were: (1) patients with primary liver cancer; (2) age 18-75 years; (3) positive for hepatitis B surface antigen > 6 months. Excluding: (1) cholangiocarcinoma patients (n = 196); (2) metastatic liver cancer patients (n = 85); (3) merging other types of tumors (n = 59); (4) patients who were not visited (n = 139); (5) patients with other chronic liver diseases (n = 172); (6) incomplete clinical data (n = 122). Finally, 2117 patients were enrolled. The liver cancer diagnosis standard meets the standard of the Asia-Tai liver cancer clinical guideline.
All patients were tested periodically every three months for CT, nuclear magnetic, ultrasound or serum AFP. When the AFP in the serum of a patient is increased or new intrahepatic nodules are discovered by ultrasound, dynamic CT or MRI is further adopted to determine whether the liver cancer progresses. With reference to mRECIST criteria, progression is defined as a sum of enhanced lesion diameters at the arterial phase in liver cancer patients increasing by > 20% or new lesions appearing. The occurrence of vascular metastasis or extrahepatic spread is also considered to be progression. "progression free survival" is defined as progression from baseline to the onset of the patient. "survival time" is defined as from patient enrollment to death or to 2018.12.31, first arrival.
The final included 2117 patients were randomly grouped, 70% of patients were training set (n = 1480) and 30% of patients were validation set (n = 637). In the training set, 434 (29.3%) patients developed disease progression within 1 year, of which 174 (40.1%) died within 1 year and 310 (71.4%) died within 3 years after progression; 784 of the 1-year progression-free patients, only 173 (22.1%) died within 3 years. In the validation set, 181 (28.4%) patients developed disease progression within 1 year, of which 93 (51.4%) died within 1 year and 134 (74.0%) died within 3 years after progression; of the 327 patients who survived no progress for 1 year, only 82 (25.1%) died within 3 years (see fig. 1). Of all patients with hepatitis B-associated liver cancer, 1848 (87.3%) had antiviral therapy, 1388 (65.6%) had reached serological conversion, and 1148 (54.2%) had reached < 500IU/ml HBV. Most patients receive topical treatment such as TACE or RFA, the remaining 9.3% receive curative resection treatment and 22.4% receive palliative treatment.
The baseline characteristic comparison is shown in table 1.
The median survival time for all populations was 26.2 months (95% CI. The median survival time for the training set was 27.2 months (95% ci. There were no statistical differences between the two groups.
Example 2Screening of progression-free survival independent influence factors of liver cancer patients and construction of ANN model
In reviewing the literature and summarizing the clinical experience, the factors found to be relevant to the survival of patients with liver cancer mainly include: tumor characteristics, concentrated in tumor size, infiltration, vascular metastasis, satellite nodules, etc.; liver function index: including albumin, total bilirubin, glutamyl transpeptidase (r-GGT), etc. However, these indicators rarely take into account the effects of immune function and inflammatory indicators on the death of patients with liver cancer. Clinical data for the patients was therefore collected throughout the study, including:
(1) Demographic characteristics: age, sex, smoking history, drinking history, family history of liver cancer;
(2) Merging the medical histories: diabetes, hypertension, coronary heart disease, hyperlipidemia, and liver cirrhosis;
(3) Hepatitis B related features: HBeAg, HBV-DNA, antiviral therapy;
(4) Laboratory indexes are as follows: blood routine: white blood cells, neutrophil percentage, lymphocyte percentage, neutrophil proportion, platelets; liver function: ALT, AST, TBIL, r-GGT, albumin, ALP, LDH, cholinesterase; renal function: creatinine; blood lipid level: triglycerides, total cholesterol; blood coagulation function: prothrombin activity, international normalized ratio; inflammation index: c-reactive protein;
(5) Tumor-related characteristics: tumor size, number, portal cancer embolus, distant metastasis, AFP;
(6) Immunological indexes: absolute count of T cells, CD 4T cells, CD 8T cells;
(7) The treatment mode comprises the following steps: excision, minimally invasive, conservative;
(8) Grading indexes: child classification and BCLC classification.
As shown in Table 2, independent influencing factors for predicting the progression-free survival of the liver cancer patients are obtained by carrying out COX single-factor and multi-factor regression analysis on the indexes.
TABLE 2 Single and Multi-factor analysis for predicting progression-free survival in liver cancer patients
The analysis gave the following results, with the risk ratios (HR) adjusted:
(1) Risk factor for 1-year progression-free survival of liver cancer patients
Combine smoking history (HR =1.188, 95% ci 1.017-1.386, p = 0.029);
tumor number greater than or equal to 2 (HR =2.357, 95% CI 1.993-2.788, p < 0.0001);
tumor size ≥ 5cm (HR =1.243, 95% CI 1.066-1.449, p = 0.001);
amalgamation of portal cancer emboli (HR =1.776, 95% ci 1.5-2.104, p < 0.001);
white blood cell count (HR =1.053, 95% CI 1.02-1.087, p = 0.001), 10 9 Per liter;
neutrophil lymphocyte fraction (HR =1.031, 95% ci 1.012-1.051, p = 0.002);
gamma-glutamyl transpeptidase (gamma-GGT) (HR =1.002, 95% ci 1.001-1.002, p < 0.0001), U/L;
alkaline phosphatase (ALP) (HR =1.001, 95% CI 1.001-1.002, p < 0.0001), U/L;
AFP≥400ng/mL(HR＝1.723，95％CI 1.456-2.026，p＜0.0001)。
(2) Protective factor for 1-year progression-free survival of liver cancer patients
Albumin (HR =0.971, 95% CI 0.959-0.984, p < 0.0001), g/L;
CD 4T cell counts (HR =0.998, 95% CI 0.998-0.998, p < 0.0001), counts/. Mu.L.
According to clinical significance, other factors, namely creatinine (Cr), except the smoking history, are selected as indexes to be used for constructing the ANN model.
The multi-layered sensor (MLP) is the most common ANN structure, and its basic components include an input layer, a hidden layer, an output layer (Amato F, lo pez a,
-Méndez E M，et al.Artificial neural networks in medical diagnosis[J].2013, 11 (2): 47-58). The invention constructs an ANN model of a multilayer perceptron structure, wherein an input layer is clinical or biochemical parameters and has 11 input neurons (the 11 factors), and an output layer is a corresponding prognosis outcome and has 2 output neurons (the number of months of non-progressive survival and whether the non-progressive survival occurs for 1 year).
As shown in fig. 3, neurons are connected by weighted links, and 3 hidden layers are added after many times of debugging and testing to improve the performance of MLP. The ANN may learn from each case data, connecting each input to a corresponding output by changing the weight of the connections between neurons. When applied, the input metrics (individual metrics for each patient) will propagate from the first layer of neurons to each layer of neurons until an output is generated; an adaptive process is then performed and the output value is compared to the expected output value (actual number of months and 1 year of progression free survival for the patient, probability ≧ 0.5 occurrence and < 0.5 non-occurrence). During the learning process, the error between the value of the actual output of the ANN and the desired output value decreases until a minimum value is reached (i.e., convergence of the network). After the two training processes, the ANN may generate an output (prognosis) from the new input data as a reasoning process based on the knowledge accumulated during the training process. In the invention, the first hidden layer and the third hidden layer adopt Ramp as an activation function, the second hidden layer adopts an ArcTan function as an activation function, and the activation function of the final output layer is a SoftMax function.
Example 3Discrimination, calibration, and comparison with other models for ANN models
The ANN model constructed in example 2 was used to predict 1-year progression-free survival of patients in the training and validation sets.
As a result, in the training set, the area under the ROC curve (AUC) of the probability that ANN predicts 1-year PFS in liver cancer patients was 0.866 (95% CI 0.848-0.884); c-index was 0.782 (95% CI 0.767-0.797). In the validation set, ANN predicted the area under the ROC curve (AUC) of 1-year PFS in liver cancer patients to be 0.730 (95% CI 0.690-0.770); c-index was 0.704 (95% CI 0.675-0.732) (Table 3).
TABLE 3 ROC Curve lower area and C index comparison of ANN model with other models to predict 1-year progression-free survival probability for hepatoma patients
The calibration curves for the predicted 1-year PFS probability and the corresponding actual observed probability are shown in fig. 4A, 4B, indicating that the ANN model has a better fit in the training and validation set for the 1-year PFS.
Based on the probability of progression-free survival of liver cancer patients in the ANN model, all patients were classified into three risk layers of progression according to the upper quartile 32.8% and the lower quartile 74.6%: strata1, low risk; strata2, moderate risk; strata3, high risk. In the training set, with reference to strata1, the risk ratios (HR) for progression-free survival for strata2, 3 were 5.25 (95% CI 3.73-7.38 p < 0.0001) and 26.42 (95% CI 18.74-37.25 p < 0.0001), respectively (FIG. 5A. In the validation set, using strata1 as a reference, the HR for progression-free survival of strata2, 3 was 1.90 (95% ci 1.35-2.66 p < 0.0001) and 6.13 (95% ci 4.28-8.79 p < 0.0001) (fig. 5B. The ANN model constructed by the present invention was demonstrated to significantly distinguish all patients from different risk of progression, whether in the training set or the validation set.
Taking the median of the CD 4T cell counts of 445/μ l as cut-off value, dividing all patients into two groups, one group of CD 4T cells is larger than or equal to 445/μ l, and one group of CD 4T cells is smaller than 445/μ l. It was found from the K-M survival curves that patients in both the training and validation sets with the high CD4 group had significantly higher PFS than patients in the low CD4 group (p < 0.05) (fig. 6A to 6B). However, in the high, medium and low risk subgroup of ANN differentiation, clinical benefit of high CD4 was found to be present only in low and medium risk patients, and in data 3, where the risk of disease progression was high, there was no difference in survival between high CD4 and low CD4 patients (fig. 6C to 6E).
In addition, when the ANN model constructed by the invention is compared with BCLC, TNM, okuda, CUPI, CLIP, JIS and ALBI, the AUC value and the C-index of the ANN model are higher than those of other models (p is less than 0.0001). Further, comparisons of AUC and C-index were also made in these sub-layers for different gender, different age, different AFP levels, different child classifications, different treatment modalities, all finding the ANN model to be higher than the other models, with the results shown in tables 4 and 5.
The above description of the specific embodiments of the present invention is not intended to limit the present invention, and those skilled in the art may make various changes and modifications according to the present invention without departing from the spirit of the present invention, which is defined by the scope of the appended claims.
Claims (8)
1. An apparatus for discriminating prognosis of a patient with liver cancer based on an artificial neural network model, the apparatus comprising:
(1) An input module configured to input information of a patient with liver cancer to be discriminated as follows:
the number of tumors is input to be 1 when the number is more than or equal to 2 and is input to be 0 when the number is less than 2;
the size of the tumor is 1 when the size is larger than or equal to 5cm and 0 when the size is smaller than 5 cm;
for the portal cancer embolus, "1" is input when combined, and "0" is input when not combined;
AFP, inputting '1' when the concentration is more than or equal to 400ng/mL, and inputting '0' when the concentration is less than 400 ng/mL;
white blood cell count, input at 10 9 Number of counts/L;
inputting the proportion of the neutrophil lymphocyte and the numerical value;
gamma-GGT, inputting the numerical value in U/L;
ALP, inputting a numerical value in U/L;
albumin, the value in g/L is input;
CD 4T cell count, input number in counts/μ L;
creatinine, inputting a value in mu mol/L;
(2) The prognosis judging module is set to comprise an artificial neural network model, the artificial neural network model is used for calculating the prognosis condition of the patient with the liver cancer to be judged based on input information, the artificial neural network model is of a multilayer perceptron structure and is constructed by Mathemica 11.1.1 software, the input layer is 11 input neurons, the output layer is 2 output neurons and comprises 3 hidden layers, wherein the 1 st hidden layer and the 3 rd hidden layer adopt Ramp as activation functions, the 2 nd hidden layer adopts an ArcTan function as an activation function, and the output layer adopts a SoftMax function as an activation function; wherein the artificial neural network model is established by the following process:
obtaining information of tumor number, tumor size, portal cancer thrombus, AFP, leukocyte count, neutrophil lymphocyte proportion, gamma-GGT, ALP, albumin, CD 4T cell count and creatinine of a liver cancer patient, using the information as input neurons of an input layer, and using the information as output neurons of an output layer according to the number of months of progression-free survival and whether the progression-free survival occurs within 1 year, and training the artificial neural network, wherein the information comprises the following steps: the input propagates from the first layer of neurons to each layer of neurons until an output is generated; comparing the output value with the expected output value, if there is a difference between the two values, generating an error signal, and then using a Back Propagation (BP) method to change the weight of the connection between neurons to reduce the overall error of the network; until the error between the value of the output and the value of the desired output decreases to a minimum;
(3) An output module configured to output the resulting prognosis.
2. The apparatus of claim 1, wherein the prognosis output by the output module is survival probability of no progress within 1 year of the patient with liver cancer to be determined.
3. The device of claim 2, wherein the liver cancer patient is hepatitis B virus-related liver cancer.
4. The device of claim 2, wherein the liver cancer patient is a primary liver cancer.
5. The apparatus of any one of claims 1 to 4, further comprising:
(4) A detection module, the detection module comprising:
a tumor number detection module;
a tumor size detection module;
a portal cancer embolus detection module;
an AFP detection module;
a leukocyte detection module;
a neutrophil lymphocyte proportion detection module;
a gamma-GGT detection module;
an ALP detection module;
an albumin detection module;
a CD 4T cell detection module;
and a creatinine detection module.
6. The apparatus of any one of claims 1 to 4, wherein the apparatus is a computer device.
7. The apparatus of claim 6, wherein the prognostic discrimination module is included in a memory of the computer device.
8. The apparatus of claim 7, wherein the computer device further comprises a processor capable of running a computer program of the artificial neural network model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010337089.6A CN112017791B (en) | 2020-04-24 | 2020-04-24 | System for determining prognosis condition of liver cancer patient based on artificial neural network model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010337089.6A CN112017791B (en) | 2020-04-24 | 2020-04-24 | System for determining prognosis condition of liver cancer patient based on artificial neural network model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112017791A CN112017791A (en) | 2020-12-01 |
| CN112017791B true CN112017791B (en) | 2022-10-11 |
Family
ID=73506538
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010337089.6A Active CN112017791B (en) | 2020-04-24 | 2020-04-24 | System for determining prognosis condition of liver cancer patient based on artificial neural network model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112017791B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113517023B (en) * | 2021-05-18 | 2023-04-25 | 柳州市人民医院 | Liver cancer prognosis marker factor related to sex and screening method |
| CN114155956B (en) * | 2021-12-02 | 2022-09-20 | 首都医科大学附属北京地坛医院 | System and method for predicting blood vessel invasion probability of primary liver cancer patient incapable of being resected by surgery |
| CN114141371B (en) * | 2021-12-02 | 2022-09-20 | 首都医科大学附属北京地坛医院 | System for predicting prognosis of liver cancer associated with tumor-related anemia patient caused by hepatitis B virus infection |
| CN114252612B (en) * | 2021-12-10 | 2023-11-14 | 四川大学华西医院 | Liver cancer metastasis prediction system and method based on nerve cell adhesion molecules |
| CN115985497B (en) * | 2022-12-16 | 2023-08-08 | 首都医科大学附属北京地坛医院 | System for predicting prognosis of patient with non-operative treatment primary liver cancer based on platelet/spleen aspect ratio value |
| CN117352169A (en) * | 2023-09-08 | 2024-01-05 | 四川互慧软件有限公司 | Snake wound rehabilitation evaluation method and device based on neural network model and electronic equipment |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR102190299B1 (en) * | 2017-02-02 | 2020-12-11 | 사회복지법인 삼성생명공익재단 | Method, device and program for predicting the prognosis of gastric cancer using artificial neural networks |
| CN107180154A (en) * | 2017-02-08 | 2017-09-19 | 冯德昭 | The method and system of patients with orthotopic liver transplantation prognosis situation packet is carried out to the patients with hepatocellular carcinoma of single tumour |
| CN108565024B (en) * | 2018-03-19 | 2020-05-05 | 首都医科大学附属北京地坛医院 | System for determining recurrence risk within 1 year after single HBV (hepatitis B Virus) related primary small liver cancer operation |
-
2020
- 2020-04-24 CN CN202010337089.6A patent/CN112017791B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN112017791A (en) | 2020-12-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112017791B (en) | System for determining prognosis condition of liver cancer patient based on artificial neural network model | |
| EP2844131B1 (en) | Methods and systems of evaluating a risk of a gastrointestinal cancer | |
| CN107209184B (en) | Marker combinations for diagnosing multiple infections and methods of use thereof | |
| Jie et al. | Low platelet to white blood cell ratio indicates poor prognosis for acute-on-chronic liver failure | |
| Qu et al. | Correlation analysis of hemoglobin-to-red blood cell distribution width ratio and frailty in elderly patients with coronary heart disease | |
| CA2741117A1 (en) | Biomarkers for heart failure | |
| EP2664922A1 (en) | Method for prediction of prognosis of sepsis | |
| Resino et al. | An artificial neural network improves the non-invasive diagnosis of significant fibrosis in HIV/HCV coinfected patients | |
| CN105849566B (en) | Biomarkers for kidney disease | |
| Yu et al. | Alpha-hydroxybutyrate dehydrogenase as a biomarker for predicting systemic lupus erythematosus with liver injury | |
| CN116287220B (en) | Molecular biomarkers and assay methods for rapid diagnosis of kawasaki disease | |
| Awasthi et al. | Monocyte HLADR and immune dysregulation index as biomarkers for COVID-19 severity and mortality | |
| Faro et al. | Polygenic risk scores and risk stratification in deep vein thrombosis | |
| Ma et al. | A novel explainable online calculator for contrast-induced AKI in diabetics: a multi-centre validation and prospective evaluation study | |
| Deng et al. | Aortic dissection detection risk score and D-dimer for acute aortic syndromes in the chinese population: exploration of optimal thresholds and integrated diagnostic value | |
| Gu et al. | A nomogram incorporating red blood cell indices to predict post-stroke cognitive impairment in the intracerebral hemorrhage population | |
| CN113969312A (en) | Marker for predicting susceptibility of ulcerative colitis patient to golimumab drug | |
| Acedera et al. | Age, comorbidities, c-reactive protein and procalcitonin as predictors of severity in confirmed COVID-19 patients in the Philippines | |
| Du et al. | A diagnostic model for Kawasaki disease based on immune cell characterization from blood samples | |
| Radmanić et al. | Antiviral treatment significantly reduces the levels of CXCL9, CXCL10 and CXCL11 in chronic hepatitis C | |
| CN116773825B (en) | Blood biomarkers and methods for diagnosing acute Kawasaki disease | |
| Salama et al. | Serum cholinesterase level as a biomarker in detecting liver injury in Patients with chronic hepatitis c | |
| CN113699235B (en) | Application of immunogenic cell death related gene in head and neck squamous cell carcinoma survival prognosis and radiotherapy responsiveness | |
| Fuse et al. | Development of machine learning models for predicting unfavorable functional outcomes in patients with chronic subdural hematomas | |
| Sarı | The Value of Indirect Biomarkers in the Monitoring of Hepatitis C Virus Infection at the Mogadishu Hospital, Somalia |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |