Feature processing method and equipment for assisting disease diagnosis

Abstract

The application discloses a feature processing method and equipment for assisting disease diagnosis, so as to reduce the limitation of some diseases relying on specific medical equipment. The feature processing method for assisting disease diagnosis comprises the following steps: obtaining a sample set containing M blood samples, obtaining N1 candidate feature values of each blood sample in the M blood samples respectively; based on the N1 candidate feature values of each blood sample, a sub-sample set containing abnormal blood samples and control blood samples for a target feature is screened out from the sample set; through a sequential forward selection algorithm and at least one machine learning algorithm, based on the N1 candidate feature values of each abnormal blood sample and the N1 candidate feature values of each control blood sample in the sub-sample set, correlation analysis is performed on the N1 candidate features and the target feature, and one or more associated features of the target feature are screened out from the N1 candidate features.

Description

Technical Field

[0001] The present invention relates to the fields of medical and artificial intelligence technology, and in particular to a feature processing method and device for assisting disease diagnosis. Background Art

[0002] In the field of medical technology, blood samples are often tested for multiple characteristic index values ​​(feature values), and specific diseases are diagnosed based on the test results. For example, the red blood cells, hemoglobin, etc. in the blood of the diagnosed subject are tested, and anemia is diagnosed based on the test results. For example, in the current medical field, the diagnosis of chronic kidney disease-mineral and bone disorder (CKD-MBD) is mainly carried out by evaluating the indicator values ​​of calcium, phosphorus and PTH in the blood to determine whether the diagnosed subject has complications of chronic kidney disease.

[0003] However, characteristics that are highly correlated with the diagnosis of some diseases require special medical equipment / systems for more accurate detection. For example, the detection of calcium, phosphorus and PTH mentioned above requires special medical equipment. However, medical conditions vary from region to region. In some areas and hospitals with limited medical conditions, without special medical equipment / systems that can detect specific characteristics, it is easy to delay the diagnosis and treatment of the disease of the patient to be diagnosed. If an epidemic disease is encountered, it will seriously affect the diagnosis and treatment of the epidemic disease among the public, posing a serious safety risk to society. Therefore, how to reduce the limitations of disease diagnosis is an issue that needs to be addressed. Summary of the Invention

[0004] The present invention provides a feature processing method and device for assisting disease diagnosis, which at least to a certain extent reduces the limitation of some disease diagnosis relying on specific medical equipment, and to a certain extent reduces the limitation of some disease diagnosis.

[0005] In a first aspect, an embodiment of the present invention provides a feature processing method for assisting disease diagnosis, the feature processing method for assisting disease diagnosis comprising:

[0006] Obtain a sample set comprising M blood samples, and obtain N1 candidate feature values ​​for each of the M blood samples; wherein the candidate feature values ​​are obtained by detecting corresponding candidate features on the blood samples, each candidate feature value in the same blood sample corresponds to one of the N1 candidate features, and different candidate feature values ​​in the same blood sample correspond to different candidate features; wherein M is an integer greater than 1, and N1 is an integer greater than 1;

[0007] Based on the N1 candidate feature values ​​of each of the blood samples, a subsample set for a target feature is screened out from the sample set; the subsample set includes abnormal blood samples and control blood samples, the abnormal blood samples include blood samples with abnormal target feature values ​​associated with a target disease among the M blood samples, and the control blood samples include some or all blood samples among the M blood samples except the abnormal blood samples, and the target feature values ​​are obtained by detecting the target feature on the corresponding blood samples;

[0008] performing a correlation analysis on the N1 candidate feature values ​​of each abnormal blood sample and the N1 candidate feature values ​​of each control blood sample in the subsample set using a sequential forward selection algorithm, and screening a set of associated candidate features for the target feature from the N1 candidate features based on the result of the correlation analysis, wherein the set of associated candidate features includes candidate features among the N1 candidate features whose correlation with the target feature is greater than a threshold value;

[0009] The candidate features in the associated candidate feature set are cross-validated by at least one machine learning algorithm, and at least one associated feature of the target feature is screened out from the associated candidate feature set based on the result of the cross-validation; the associated features are used to diagnose the target disease in the subject to be diagnosed based on the characteristic values ​​of the associated features in the blood of the subject to be diagnosed.

[0010] In the embodiment of the present application, by constructing a sample set and using a sequential forward selection algorithm and a machine learning algorithm, the correlation relationship between N1 candidate features of the blood sample and the target features associated with the target disease is analyzed multiple times, and the associated features with a high degree of correlation with the target features are screened out from the NI candidate features. Then, when diagnosing the target disease of the subject to be diagnosed, a medical device that can detect the associated features is used to detect the characteristic values ​​of the associated features in the blood of the subject to be diagnosed, and then the characteristic values ​​of the associated features are used to assist in the diagnosis of the target disease. In the diagnosis of the target disease of the subject to be diagnosed, there is no need to detect the characteristic values ​​of the target features in their blood, and the diagnosis of the target disease does not require specific medical equipment to detect the target features, thereby reducing the dependence of the diagnosis of the target disease on the aforementioned specific medical equipment, thereby improving the diversity and flexibility of the methods for diagnosing the target disease.

[0011] In a second aspect, an embodiment of the present invention provides a feature processing device for assisting disease diagnosis, the device comprising a processor and a memory, the memory being used to store programs executable by the processor, the processor being used to read the programs in the memory and execute the steps of the feature processing method for assisting disease diagnosis described in the first aspect.

[0012] In a third aspect, an embodiment of the present invention further provides a feature processing device for assisting disease diagnosis, the feature processing device comprising:

[0013] a first data processing unit, configured to obtain a sample set comprising M blood samples, and obtain N1 candidate feature values ​​for each of the M blood samples; wherein the candidate feature values ​​are obtained by detecting corresponding candidate features on the blood samples, each candidate feature value in the same blood sample corresponds to one of the N1 candidate features, and different candidate feature values ​​in the same blood sample correspond to different candidate features; wherein M is an integer greater than 1, and N1 is an integer greater than 1;

[0014] a second data processing unit, configured to screen out a subsample set for a target feature from the sample set based on the N1 candidate feature values ​​of each of the blood samples; the subsample set comprising abnormal blood samples and control blood samples, the abnormal blood samples comprising blood samples having abnormal target feature values ​​associated with a target disease among the M blood samples, the control blood samples comprising some or all of the blood samples other than the abnormal blood samples among the M blood samples, the target feature values ​​being obtained by detecting the target feature on the corresponding blood samples;

[0015] a first feature processing unit, configured to perform a correlation analysis on the N1 candidate feature values ​​of each abnormal blood sample and the N1 candidate feature values ​​of each control blood sample in the subsample set using a sequential forward selection algorithm, and to screen out, from the N1 candidate features, a set of associated candidate features for the target feature based on the result of the correlation analysis, wherein the set of associated candidate features includes candidate features among the N1 candidate features whose correlation with the target feature is greater than a threshold value;

[0016] The second feature processing unit is used to cross-validate the candidate features in the associated candidate feature set through at least one machine learning algorithm, and screen out at least one associated feature of the target feature from the associated candidate feature set based on the result of the cross-validation; the associated features are used to diagnose the target disease in the subject to be diagnosed based on the feature values ​​of the associated features in the blood of the subject to be diagnosed.

[0017] In a fourth aspect, an embodiment of the present invention further provides a computer storage medium having a computer program stored thereon, which, when executed by a processor, is used to implement the steps of a feature processing method for assisting disease diagnosis as described in the first aspect above.

[0018] These and other aspects of the present application will be more clearly understood in the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 A flow chart of a feature processing method for assisting disease diagnosis provided by an embodiment of the present invention;

[0021] Figure 2 A flowchart of a feature screening operation provided by an embodiment of the present invention;

[0022] Figure 3 A flowchart of screening a candidate feature set for association provided by an embodiment of the present invention;

[0023] Figure 4 A schematic diagram of an implementation effect provided by an embodiment of the present invention;

[0024] Figure 5 A schematic diagram of a feature processing device for assisting disease diagnosis provided by an embodiment of the present invention;

[0025] Figure 6 A schematic diagram of a feature processing device for assisting disease diagnosis provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0026] To make the objectives, technical solutions, and advantages of the present invention more apparent, the present invention will be further described in detail below with reference to the accompanying drawings. It is apparent that the embodiments described are only a portion of the embodiments of the present invention, rather than all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort are intended to fall within the scope of protection of the present invention.

[0027] In embodiments of the present invention, the term "and / or" describes the association relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally indicates that the associated objects are in an "or" relationship.

[0028] The application scenarios described in the embodiments of the present application are used to more clearly illustrate the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided by the embodiments of the present application. Those skilled in the art can know that, with the emergence of new application scenarios, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems. In the description of the present application, unless otherwise specified, the meaning of "a plurality of" is two or more.

[0029] First, some technical terms related to the embodiments of the present application are described:

[0030] Target disease: The target disease in the embodiments of the present application refers to a disease that can be diagnosed based on blood pressure assistance. The type of target disease is not limited, and those skilled in the art can set it according to actual needs, such as but not limited to chronic kidney disease, thyroid, anemia, etc. The following content of the embodiments of the present application only takes chronic kidney disease as a specific example of the target disease for illustration.

[0031] Target feature: that is, a feature with a high degree of association with the diagnosis of the target disease. For example, when the target disease is chronic kidney disease, the target feature can be but is not limited to at least one chemical substance in calcium, phosphorus, and parathyroid hormone PTH.

[0032] Candidate features: features whose index values / feature values ​​can be detected from the blood. In the embodiment of the present application, candidate features are possible features that can be used to replace target features to assist in the diagnosis of target diseases; for example, when the target disease is chronic kidney disease, and the target feature is at least one of calcium, phosphorus, and parathyroid hormone (PTH), candidate features may include, but are not limited to, KT / V (dialysis adequacy), GH (serum growth hormone), Glucose (blood glucose), PreHemoUA (serum uric acid before hemodialysis), TotalBA (whole blood eosinophils), GOT (aspartate aminotransferase), Beta2Microglobulin (β2 microglobulin), HDL (high-density lipoprotein) and LD (lactate dehydrogenase), eosinophil count, eosinophil rate, PreHem oUrea (urea nitrogen before hemodialysis), PostHemoUA (serum uric acid after hemodialysis), basophil count, LDL (low-density lipoprotein), neutrophil rate, monocyte count, cystatin C, MCV (mean corpuscular volume); PostHemoCreatinine (creatinine after hemodialysis), AP (prothrombin time), Globulin (serum globulin), CDBRate (white-globulin ratio), TBil (total bilirubin) and DDimerTest (D-dimer test), PreHemoCreatinine (creatinine before hemodialysis), MeanHemoglobin (mean hemoglobin), MonocyteCount (monocyte count), Glucose (glucose), TG (triglyceride), Iron (iron) among multiple ones.

[0033] Associated features: candidate features that are screened out from candidate features and have a high degree of correlation with the target features; the associated features are used to diagnose the target disease in the subject to be diagnosed based on the characteristic values ​​of the associated features in the blood of the subject to be diagnosed.

[0034] An embodiment of the present application provides a feature processing method and device for assisting disease diagnosis. In the method, a sample set comprising M blood samples is first obtained, and N1 candidate feature values ​​of each blood sample in the M blood samples are obtained respectively; then, based on the N1 candidate feature values ​​of each blood sample, a subsample set for a target feature is screened out from the sample set; the subsample set comprises abnormal blood samples and control blood samples; through a sequential forward selection algorithm and at least one machine learning (ML) algorithm, correlation analysis and cross-validation are performed on the N1 candidate features and the target feature based on the N1 candidate feature values ​​of each abnormal blood sample and the N1 candidate feature values ​​of each control blood sample in the subsample set, and finally one or more associated features of the target feature are screened out from the N1 candidate features; the associated features are used to diagnose the target disease in a subject to be diagnosed based on the feature values ​​of the associated features in the blood of the subject to be diagnosed.

[0035] The following is a detailed description of a feature processing method for assisting disease diagnosis provided in an embodiment of the present application; see Figure 1 The feature processing method for assisting disease diagnosis provided in the embodiment of the present application mainly includes the following steps S110 to S140:

[0036] Step S110, obtaining a sample set including M blood samples, and obtaining N1 candidate feature values ​​for each of the M blood samples respectively; wherein the candidate feature values ​​are obtained by detecting the corresponding candidate features of the blood samples, each candidate feature value in the same blood sample corresponds to one of the N1 candidate features, and different candidate feature values ​​in the same blood sample correspond to different candidate features; M is an integer greater than 1, and N1 is an integer greater than 1.

[0037] The specific values ​​of M and N1 are not limited in the embodiments of this application, and those skilled in the art may set them according to actual needs. As an example, considering the actual situation of blood sample testing, in the embodiments of this application, all features that can detect characteristic values ​​from blood can be used as candidate features based on the status of existing medical equipment, thereby determining the values ​​of M and N1 based on actual clinical conditions.

[0038] As an embodiment, the aforementioned M blood samples can be collected from M patients / living bodies. Considering that when blood samples are collected from patients, the blood samples are generally marked with the patient's name, provincial ID number, address, telephone number and other personal attribute information or identity information, therefore, in the embodiment of the present application, before step S110, the personal attribute information and / or identity information that is strongly related to the patient can be removed from the M blood samples obtained.

[0039] As an embodiment, considering improving the accuracy and processing efficiency of feature processing, in the embodiment of the present application, all features whose characteristic values ​​can be detected from the blood can be first used as basic features (hereinafter it is assumed that the number of basic features is N2), and then a feature filtering operation is performed on the N2 basic features. In the feature filtering operation, based on the basic characteristic values ​​of each basic feature of each blood sample, the basic features with abnormal characteristic values ​​can be filtered out from the N2 basic features, and the filtered basic features can be determined as the aforementioned N1 candidate features.

[0040] Step S120: Based on the N1 candidate feature values ​​of each blood sample, a subsample set for the target feature is screened out from the sample set; the subsample set includes abnormal blood samples and control blood samples, the abnormal blood samples include blood samples with abnormal target feature values ​​associated with the target disease among the M blood samples, and the control blood samples include some or all blood samples among the M blood samples except the abnormal blood samples, and the target feature value is obtained by detecting the target feature on the corresponding blood sample.

[0041] Step S130, using a sequential forward selection algorithm, based on the N1 candidate feature values ​​of each abnormal blood sample and the N1 candidate feature values ​​of each control blood sample in the subsample set, performing a correlation analysis on the N1 candidate features and the target feature, and screening out an associated candidate feature set for the target feature from the N1 candidate features according to the result of the correlation analysis, wherein the associated candidate feature set includes candidate features among the N1 candidate features whose correlation degree with the target feature is greater than a threshold.

[0042] Step S140, cross-validating the candidate features in the associated candidate feature set through at least one machine learning algorithm, and screening out at least one associated feature of the target feature from the associated candidate feature set based on the result of the cross-validation; the associated feature is used to diagnose the target disease in the subject to be diagnosed based on the characteristic value of the associated feature in the blood of the subject to be diagnosed.

[0043] As an embodiment, in step S330, sub-cross-validation can be performed on the candidate features in the associated candidate feature set through each machine learning algorithm in the at least one machine learning algorithm, and a sub-associated feature set of the target feature can be screened out from the associated candidate feature set based on the result of the sub-cross-validation; and then the candidate features in each screened sub-associated feature set can be verified, and at least one associated feature can be screened out from each sub-associated feature set.

[0044] As an embodiment, the at least one machine learning algorithm includes one or more of a KNN algorithm, a linear regression algorithm, a logistic regression algorithm, and a support vector machine algorithm; wherein, those skilled in the art may also choose to use an ML algorithm other than the KNN algorithm, the linear regression algorithm, the logistic regression algorithm, and the support vector machine algorithm for processing according to actual business needs; wherein:

[0045] When the aforementioned at least one machine learning algorithm includes the KNN algorithm, the candidate features in the associated candidate feature set can be used as sample data for processing by the KNN algorithm, and the candidate features in the associated candidate feature set can be processed using the existing KNN algorithm to screen out the at least one associated feature from the associated feature set; when the at least one machine learning algorithm includes one or more of a linear regression algorithm, a logistic regression algorithm, and a support vector machine algorithm, the processing of the aforementioned KNN algorithm can be referred to, and one or more of the existing linear regression algorithm, logistic regression algorithm, and support vector machine algorithm can be used to screen out the corresponding associated features from the associated feature set.

[0046] As an embodiment, in step S110, N2 basic feature values ​​of each blood sample in the M blood samples can be obtained; the N2 basic feature values ​​of each blood sample are cleared to obtain N1 candidate feature values ​​of each blood sample; the basic feature values ​​are obtained by detecting the corresponding basic features of the blood samples, each basic feature value in the same blood sample corresponds to one basic feature among the N2 basic features, and different basic feature values ​​in the same blood sample correspond to different basic features, and N2 is an integer greater than or equal to N1; wherein, the specific numerical value of the aforementioned N1 is not limited in the embodiment of the present application, and those skilled in the art can set it according to actual needs, and all features whose feature values ​​can be detected in the blood clinically can be used as basic features, the total number of all basic features that can be known clinically can be used as N2, or the total number of some basic features among all basic features that can be known clinically can be used as N2.

[0047] As an embodiment, the step of performing data cleaning on the N2 basic feature values ​​of each blood sample to obtain N1 candidate feature values ​​of each blood sample includes:

[0048] The N2 basic features are respectively used as target basic features, and a feature screening operation is performed on the target basic features to determine N1 candidate features from the N2 basic features; the feature values ​​corresponding to the N1 candidate features in each blood sample are determined as the N1 feature values ​​of each blood sample; wherein:

[0049] See Figure 2The feature screening operation includes the following steps S210 to S230:

[0050] Step S210, obtaining a target basic characteristic value of each blood sample, wherein the target basic characteristic value is obtained by detecting the target basic characteristic of the blood sample;

[0051] Step S220, determining the total number of outliers in all the obtained target basic feature values;

[0052] Step S230: If the total number of outliers is less than a first number threshold, the target basic feature is determined as one of the N1 candidate features; otherwise, the target basic feature is determined as a basic feature other than the N1 candidate features.

[0053] Among them, there is no limit on the first quantity threshold, and technical personnel in this field can set it based on actual needs, such as it can be but not limited to using the integer value of the product of N1 and the preset ratio as the first quantity threshold; there is no limit on the aforementioned preset ratio, and technical personnel in this field can set it according to actual needs, such as it can be but not limited to setting it to 15%, 20%, 25% or 30%, etc.

[0054] See Figure 3 As an embodiment, the process of screening out the associated candidate feature set in step S130 may include but is not limited to steps S310 to S330.

[0055] Step S310 , determining an F-Test score of each of the N1 candidate features based on the N1 candidate feature values ​​of each abnormal blood sample and the N1 candidate feature values ​​of each control blood sample in the subsample set.

[0056] Step S320 , sorting the N1 candidate features in descending order of F-Test scores to obtain a candidate feature sequence.

[0057] Step S330: performing N1 iterations on the N1 candidate features according to the arrangement order of the N1 candidate features in the candidate feature sequence to obtain the associated candidate feature set; wherein the i+1th iteration in the N1 iterations includes the following steps S3301 to S3303:

[0058] Step S3301: determining the candidate feature ranked at the i+1th position in the candidate feature sequence as the candidate feature to be processed;

[0059] Step S3302: determining the classification contribution of the candidate feature to be processed to the target feature based on the candidate feature added to the associated candidate feature set in the i-th iteration operation;

[0060] Step S3303: If the determined classification contribution is greater than the classification contribution determined in the i-th iteration operation, the candidate feature to be processed is determined to be added to the associated candidate feature set; wherein:

[0061] The i is a value greater than 0 and less than N1; and when the i is 1, in the i-th iteration operation, the candidate feature ranked at the first position in the candidate feature sequence is directly added to the associated candidate feature set.

[0062] As an embodiment, the classification contribution is the area under the receiver operating characteristic curve (ROC curve) AUC.

[0063] Among them, the ROC curve is an indicator for evaluating binary classifiers. The horizontal axis of the ROC curve is the false positive rate (False Positive Rare), as shown in the following formula (1); the vertical axis of the ROC curve is the true positive rate (FPTure Positive Rar), as shown in the following formula (2):

[0064] Formula (1)

[0065] Formula (2)

[0066] In the above formulas (1) and (2), FP is the number of N negative samples predicted as positive samples by the classifier, N is the number of true negative samples, TN is the number of N negative samples predicted as negative samples by the classifier, TP is the number of P positive samples predicted as positive samples by the classifier, FN is the number of P positive samples predicted as negative samples by the allocator, and P is the number of true positive samples.

[0067] AUC (Area Under Curve) is defined as the area under the ROC curve, and the value of this area (AUC) will not be greater than 1. The AUC value is used as the evaluation criterion in the embodiments of the present application because the ROC curve often cannot clearly indicate which classifier is more effective. As a numerical value, the classifier with a larger AUC is more effective. The larger the AUC, the more likely the classifier is to rank the true positive samples in the front, and the better the classification performance, so as to more accurately know the classification results of the classifier for each feature. That is, in the process of steps S310 to S330 above, the F-Test score of each candidate feature in the N1 candidate features is first calculated, and the N1 candidate features are sorted in descending order according to the F-Test score to obtain a candidate feature sequence. In the embodiment of the present application, the F-Test score is used to measure the inter-class difference of different types of features in the N1 candidate features divided by the intra-class difference of the same type of features. Therefore, all candidate features in the candidate feature sequence after descending order are sorted according to their classification contribution (i.e., the discriminative power for classification); then, in the embodiment of the present application, the first ranked feature ranked in the first position in the candidate feature sequence can be introduced as a seed into the associated candidate feature set (i.e., the feature pool described below), and classification is performed under leave-one-out cross-validation. After that, the area under the ROC curve (AUC) is calculated and recorded to measure the classification performance of the candidate feature; then, the second-ranked candidate feature is added to the feature pool, and the process is repeated again to obtain a new AUC. If the new AUC is greater than the old AUC, the second-ranked candidate feature is retained in the feature pool; otherwise, it is discarded. Similarly, the process is repeated for all the following candidate features to obtain the final selected feature pool.

[0068] As an embodiment, the following content takes chronic kidney disease as a specific example of the target disease to further illustrate the process of step S110 to step S140:

[0069] When the target disease is chronic kidney disease, the target characteristic may include, but is not limited to, at least one chemical substance selected from calcium, phosphorus, and parathyroid hormone (PTH).

[0070] As an embodiment, the target features include calcium, phosphorus and parathyroid hormone (PTH) at the same time; the abnormal blood sample includes part or all of the blood samples with abnormal calcium, abnormal phosphorus and abnormal PTH in the M blood samples; and the control blood sample includes part or all of the blood samples in the M blood samples except the blood samples with abnormal calcium, abnormal phosphorus and abnormal PTH.

[0071] The following is a further explanation of a feature processing method for assisting disease diagnosis provided in an embodiment of the present application, taking a specific implementation process as an example. In this example, M is 116, N1 is 66, and the target features include a first target feature (calcium), a second target feature (phosphorus), a third target feature (PTH), and a fourth target feature (calcium, phosphorus, and PTH).

[0072] In step S110 , 116 blood samples are collected, and 66 candidate feature values ​​of each of the 116 blood samples are obtained;

[0073] Further in step S120, a first subsample set (CKDMBD-Cal), a second subsample set (CKDMBD-Phos), a third subsample set (CKDMBD-PTH), and a fourth subsample set (CKDMBD-Any) are screened from the 116 blood samples for the first target feature, the second target feature, the third target feature, and the fourth target feature, respectively, where:

[0074] Blood samples with characteristic values ​​of calcium greater than 2.5 or less than 2.1 were considered abnormal blood samples in CKDMBD-Cal; blood samples with characteristic values ​​of phosphorus greater than 1.45 or less than 0.78 were considered abnormal blood samples in CKDMBD-Phos; blood samples with characteristic values ​​of PTH greater than 600 or less than 100 were considered abnormal blood samples in CKDMBD-PTH; blood samples belonging to at least one of CKDMBD-Cal, CKDMBD-Phos, and CKDMBD-PTH were considered abnormal blood samples in CKDMBD-Any. The final CKDMBD-Cal included 57 abnormal blood samples (blood samples with abnormal calcium) and 59 control blood samples; CKDMBD-Phos included 96 abnormal blood samples (blood samples with abnormal phosphorus) and 20 control blood samples; CKDMBD-PTH included 30 abnormal blood samples (blood samples with abnormal PTH) and 86 control blood samples; CKDMBD-Any included 108 abnormal blood samples (i.e., blood samples with abnormal characteristic values ​​of at least one of calcium, phosphorus and PTH) and 8 control blood samples.

[0075] After step S130 and step S140, CKDMBD-Cal, under the sequential forward feature selection method based on leave-one-out cross-validation, and the KNN algorithm, linear regression algorithm, logistic regression algorithm and support vector regression algorithm, screened out 7, 9, 8 and 17 associated features, respectively, with a maximum AUC of 0.62, 1500.684, 0.687 and 0.992, respectively (e.g. Figure 4For CKDMBD-Phos, the four ML algorithms screened 4, 15, 11 and 10 associated features with the largest AUC of 0.673, 0.798, 0.784 and 0.772, respectively; for CKDMBD-PTH, the four ML algorithms screened 6, 8, 10 and 8 associated features with the largest AUC of 0.748, 0.792, 0.787 and 0.80, respectively; for CKDMBD-Any, the four ML algorithms screened 7, 13, 8 and 11 associated features with the largest AUC of 0.799, 0.950, 0.920 and 0.940, respectively, which were 155.

[0076] Among the four ML learning methods, the commonality and correlation of the associated features screened out by the four ML learning methods are as follows: the four ML learning methods jointly cover 27, 23, 18 and 22 associated features in the CKDMBD-Cal, CKDMBD-Phos, CKDMBD-PTH and CKDMBD-Any scenarios; among these associated features, at least two ML learning methods screen out 9, 12, 6 and 8 associated features; specifically, for CKDMBD-Cal, KT / V (dialysis adequacy) (4 counts), GH (serum growth hormone) (4), Glucose (blood glucose) (3), PostHemoUA (post-hemodialysis serum uric acid) (2), TotalBA (total blood eosinophil count) (2), GOT (aspartate aminotransferase) (2), Beta2Microglobulin (beta2 microglobulin) (2), HDL (high-density lipoprotein) (2), and LD (lactate dehydrogenase) (2) are screened out as the associated features of the first target feature (calcium); for CKDMBD-Phos, eosinophil count (4), eosinophil rate (3), PreHemoUrea (pre-hemodialysis urea nitrogen) (3), GOT (aspartate aminotransferase) (3), PreHemoUA (pre-hemodialysis serum uric acid) (2), basophil count (2), LDL (low-density lipoprotein) (2), neutrophil rate (2), PostHemoUA (post-hemodialysis serum uric acid) (2), monocyte count (2), Cystatin C (2), and MCV (mean corpuscular volume) (2) are screened out as the associated features of the second target feature (phosphorus); for CKDMBD-PTH, PostHemoCreatinine (post-hemodialysis creatinine) (4), AP (prothrombin time) (4), Globulin (serum globulin) (4), CDBRate (white globin ratio) (3), TBil (total bilirubin) (3), and DDimerTest (D-dimer test) (2) are screened out as the associated features of the third target feature (PTH); for CKDMBD-Any, PreHemoCreatinine (pre-hemodialysis creatinine) (4), MeanHemoglobin (mean hemoglobin) (4), LDL (low-density lipoprotein) (4), MonocyteCount (monocyte count) (3), PreHemoUA (pre-hemodialysis serum uric acid) (3), Glucose (blood glucose) (3), TG (triglycerides) (2), and Iron (iron) (2) are screened out as the associated features of the fourth target feature (calcium, phosphorus and PTH).

[0077] See Figure 5Based on the same inventive concept, the application further provides a feature processing device 500 for assisting disease diagnosis, which comprises:

[0078] a first data processing unit 510, configured to obtain a sample set containing M blood samples, and obtain N1 candidate feature values of each blood sample in the M blood samples respectively; wherein the candidate feature values are obtained by detecting the blood samples for corresponding candidate features, each candidate feature value in the same blood sample corresponds to one candidate feature of N1 candidate features, and different candidate feature values in the same blood sample correspond to different candidate features; M is an integer greater than 1, and N1 is an integer greater than 1;

[0079] a second data processing unit 520, configured to filter a sub-sample set for a target feature from the sample set based on the N1 candidate feature values of each blood sample; the sub-sample set contains abnormal blood samples and control blood samples, the abnormal blood samples include blood samples with abnormal target feature values associated with a target disease in the M blood samples, and the control blood samples include part or all of the blood samples other than the abnormal blood samples in the M blood samples, the target feature value is obtained by detecting the corresponding blood sample for the target feature;

[0080] a first feature processing unit 530, configured to perform correlation analysis on the N1 candidate features and the target feature based on the N1 candidate feature values of each abnormal blood sample and the N1 candidate feature values of each control blood sample in the sub-sample set by a sequential forward selection algorithm, and filter a set of associated candidate features for the target feature from the N1 candidate features according to the result of the correlation analysis, the set of associated candidate features contains candidate features in the N1 candidate features whose correlation with the target feature is greater than a threshold value;

[0081] a second feature processing unit 540, configured to perform cross-validation on the candidate features in the set of associated candidate features by at least one machine learning algorithm, and filter at least one associated feature of the target feature from the set of associated candidate features based on the result of the cross-validation; the associated feature is used to diagnose the target disease based on the feature value of the associated feature in the blood of a to-be-diagnosed object in diagnosing the target disease of the to-be-diagnosed object.

[0082] As an embodiment, the first data processing unit 510 is specifically configured to:

[0083] Obtaining N2 basic characteristic values ​​for each of the M blood samples; the basic characteristic values ​​are obtained by detecting corresponding basic characteristics of the blood samples, each basic characteristic value in the same blood sample corresponds to one basic characteristic of the N2 basic characteristics, and different basic characteristic values ​​in the same blood sample correspond to different basic characteristics, where N2 is an integer greater than or equal to N1;

[0084] Data clarification is performed on the N2 basic eigenvalues ​​of each blood sample to obtain N1 candidate eigenvalues ​​of each blood sample.

[0085] As an embodiment, the first data processing unit 510 is specifically configured to:

[0086] Using the N2 basic features as target basic features respectively, and performing feature screening operations on the target basic features to determine N1 candidate features from the N2 basic features;

[0087] Determine the feature values ​​corresponding to the N1 candidate features in each blood sample as the N1 feature values ​​of each blood sample; wherein:

[0088] The feature screening operation includes:

[0089] Obtaining a target basic characteristic value of each of the blood samples, wherein the target basic characteristic value is obtained by detecting the target basic characteristic of the blood sample;

[0090] Determining the total number of outliers in all the obtained target basic feature values;

[0091] If the total number of outliers is less than a first number threshold, the target basic feature is determined as one of the N1 candidate features; otherwise, the target basic feature is determined as a basic feature other than the N1 candidate features.

[0092] As an embodiment, the target disease includes chronic kidney disease, and the target characteristic includes at least one chemical substance selected from calcium, phosphorus, and parathyroid hormone (PTH).

[0093] As an embodiment, the target characteristics include calcium, phosphorus and parathyroid hormone (PTH);

[0094] The abnormal blood sample includes part or all of the abnormal calcium blood sample, abnormal phosphorus blood sample and abnormal PTH blood sample in the M blood samples;

[0095] The control blood sample includes part or all of the blood samples in the M blood samples except the blood sample with abnormal calcium, the blood sample with abnormal phosphorus, and the blood sample with abnormal PTH.

[0096] As an embodiment, the first feature processing unit 530 is specifically configured to:

[0097] Determining an F-Test score for each of the N1 candidate features based on the N1 candidate feature values ​​of each abnormal blood sample and the N1 candidate feature values ​​of each control blood sample in the subsample set;

[0098] Sort the N1 candidate features in descending order of F-Test scores to obtain a candidate feature sequence;

[0099] According to the arrangement order of the N1 candidate features in the candidate feature sequence, N1 iteration operations are performed on the N1 candidate features to obtain the associated candidate feature set, wherein the i+1th iteration operation in the N1 iteration operations includes:

[0100] Determine the candidate feature ranked at the i+1th position in the candidate feature sequence as the candidate feature to be processed;

[0101] Determining, based on the candidate features added to the associated candidate feature set in the i-th iteration operation, the classification contribution of the candidate features to be processed to the target feature;

[0102] If the determined classification contribution is greater than the classification contribution determined in the i-th iterative operation, then the candidate feature to be processed is determined to be added to the associated candidate feature set; wherein:

[0103] The i is a value greater than 0 and less than N1; and when the i is 1, in the i-th iteration operation, the candidate feature ranked at the first position in the candidate feature sequence is directly added to the associated candidate feature set.

[0104] As an embodiment, the classification contribution is the area under the receiver operating characteristic (ROC) curve AUC.

[0105] As an embodiment, the second feature processing unit 540 is specifically configured to:

[0106] Performing sub-cross validation on the candidate features in the associated candidate feature set using each of the at least one machine learning algorithm, and screening out a sub-associated feature set of the target feature from the associated candidate feature set based on the results of the sub-cross validation;

[0107] The candidate features in each of the screened sub-association feature sets are verified, and the at least one association feature is screened out from each of the sub-association feature sets.

[0108] As an embodiment, the at least one machine learning algorithm includes one or more of a KNN algorithm, a linear regression algorithm, a logistic regression algorithm, and a support vector machine algorithm.

[0109] Based on the same inventive concept, the embodiment of the present invention also provides a feature processing device for assisting disease diagnosis, such as Figure 6 As shown, the device includes a processor 600 and a memory 610, wherein the memory 610 is used to store programs executable by the processor 600, and the processor 600 is used to read the programs in the memory 610 and execute the steps of any one of the feature processing methods for assisting disease diagnosis and any one of the methods of the embodiments described above, which will not be repeated here.

[0110] Based on the same inventive concept, an embodiment of the present application provides a computer device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor, characterized in that when the processor executes the program, it is applicable to the steps of any one of the aforementioned feature processing methods for assisting disease diagnosis.

[0111] Based on the same inventive concept, an embodiment of the present application also provides a computer-readable storage medium, which stores computer instructions. When the computer instructions are run on a computer, the computer executes the steps of any of the aforementioned feature processing methods for assisting disease diagnosis.

[0112] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage and optical storage) containing computer-usable program code.

[0113] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that performs the functions specified in one or more boxes.

[0114] These computer program instructions may also be stored in a computer readable memory 610 that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory 610 produce an article of manufacture including an instruction device that implements the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0115] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A feature processing method for assisting disease diagnosis, characterized in that: The auxiliary diagnosis method of the disease includes: Obtain a sample set comprising M blood samples, and obtain N1 candidate feature values ​​for each of the M blood samples; wherein the candidate feature values ​​are obtained by detecting corresponding candidate features on the blood samples, each candidate feature value in the same blood sample corresponds to one of the N1 candidate features, and different candidate feature values ​​in the same blood sample correspond to different candidate features; wherein M is an integer greater than 1, and N1 is an integer greater than 1; Based on the N1 candidate feature values ​​of each of the blood samples, a subsample set for a target feature is screened out from the sample set; the subsample set includes abnormal blood samples and control blood samples, the abnormal blood samples include blood samples with abnormal target feature values ​​associated with a target disease among the M blood samples, and the control blood samples include some or all blood samples among the M blood samples except the abnormal blood samples, and the target feature values ​​are obtained by detecting the target feature on the corresponding blood samples; performing a correlation analysis on the N1 candidate feature values ​​of each abnormal blood sample and the N1 candidate feature values ​​of each control blood sample in the subsample set using a sequential forward selection algorithm, and screening a set of associated candidate features for the target feature from the N1 candidate features based on the result of the correlation analysis, wherein the set of associated candidate features includes candidate features among the N1 candidate features whose correlation with the target feature is greater than a threshold value; The candidate features in the associated candidate feature set are cross-validated by at least one machine learning algorithm, and at least one associated feature of the target feature is screened out from the associated candidate feature set based on the result of the cross-validation; the associated features are used to diagnose the target disease in the subject to be diagnosed based on the characteristic values ​​of the associated features in the blood of the subject to be diagnosed.

2. The feature processing method for assisting disease diagnosis according to claim 1, characterized in that: The step of respectively obtaining N1 candidate feature values ​​of each of the M blood samples includes: Obtaining N2 basic characteristic values ​​for each of the M blood samples; the basic characteristic values ​​are obtained by detecting corresponding basic characteristics of the blood samples, each basic characteristic value in the same blood sample corresponds to one basic characteristic of the N2 basic characteristics, and different basic characteristic values ​​in the same blood sample correspond to different basic characteristics, where N2 is an integer greater than or equal to N1; Data cleaning is performed on the N2 basic eigenvalues ​​of each blood sample to obtain N1 candidate eigenvalues ​​of each blood sample.

3. The feature processing method for assisting disease diagnosis according to claim 2, characterized in that: The step of performing data cleaning on the N2 basic eigenvalues ​​of each blood sample to obtain N1 candidate eigenvalues ​​of each blood sample includes: Using the N2 basic features as target basic features respectively, and performing feature screening operations on the target basic features to determine N1 candidate features from the N2 basic features; Determine the feature values ​​corresponding to the N1 candidate features in each blood sample as the N1 feature values ​​of each blood sample; wherein: The feature screening operation includes: Obtaining a target basic characteristic value of each of the blood samples, wherein the target basic characteristic value is obtained by detecting the target basic characteristic of the blood sample; Determining the total number of outliers in all the obtained target basic feature values; If the total number of outliers is less than a first number threshold, the target basic feature is determined as one of the N1 candidate features; otherwise, the target basic feature is determined as a basic feature other than the N1 candidate features.

4. The feature processing method for assisting disease diagnosis according to claim 1, characterized in that: The target disease includes chronic kidney disease, and the target characteristic includes at least one chemical substance selected from calcium, phosphorus, and parathyroid hormone (PTH).

5. The feature processing method for assisting disease diagnosis according to claim 1, characterized in that: The target characteristics also include calcium, phosphorus and parathyroid hormone (PTH); The abnormal blood sample includes part or all of the abnormal calcium blood sample, abnormal phosphorus blood sample and abnormal PTH blood sample in the M blood samples; The control blood sample includes part or all of the blood samples in the M blood samples except the blood sample with abnormal calcium, the blood sample with abnormal phosphorus, and the blood sample with abnormal PTH.

6. The feature processing method for assisting disease diagnosis according to any one of claims 1 to 5, characterized in that: The step of performing correlation analysis on the N1 candidate feature values ​​of each abnormal blood sample and the N1 candidate feature values ​​of each control blood sample in the subsample set by using a sequential forward selection algorithm, and screening out a candidate feature set associated with the target feature from the N1 candidate features according to the result of the correlation analysis, includes: Determining an F-Test score for each of the N1 candidate features based on the N1 candidate feature values ​​of each abnormal blood sample and the N1 candidate feature values ​​of each control blood sample in the subsample set; Sort the N1 candidate features in descending order of F-Test scores to obtain a candidate feature sequence; According to the arrangement order of the N1 candidate features in the candidate feature sequence, N1 iteration operations are performed on the N1 candidate features to obtain the associated candidate feature set, wherein the i+1th iteration operation in the N1 iteration operations includes: Determine the candidate feature ranked at the i+1th position in the candidate feature sequence as the candidate feature to be processed; Determining, based on the candidate features added to the associated candidate feature set in the i-th iteration operation, the classification contribution of the candidate features to be processed to the target feature; If the determined classification contribution is greater than the classification contribution determined in the i-th iterative operation, then the candidate feature to be processed is determined to be added to the associated candidate feature set; wherein: The i is a value greater than 0 and less than N1; and when the i is 1, in the i-th iteration operation, the candidate feature ranked at the first position in the candidate feature sequence is directly added to the associated candidate feature set.

7. The feature processing method for assisting disease diagnosis according to claim 6, characterized in that: The classification contribution is the area under the receiver operating characteristic (ROC) curve (AUC).

8. The feature processing method for assisting disease diagnosis according to any one of claims 1 to 5, characterized in that: The step of cross-validating the candidate features in the associated candidate feature set by at least one machine learning algorithm, and selecting at least one associated feature of the target feature from the associated candidate feature set based on the result of the cross-validation, comprises: Performing sub-cross validation on the candidate features in the associated candidate feature set using each of the at least one machine learning algorithm, and screening out a sub-associated feature set of the target feature from the associated candidate feature set based on the results of the sub-cross validation; The candidate features in each of the screened sub-association feature sets are verified, and the at least one association feature is screened out from each of the sub-association feature sets.

9. The feature processing method for assisting disease diagnosis according to claim 8, characterized in that: The at least one machine learning algorithm includes one or more of a K-nearest neighbor algorithm (KNN), a linear regression algorithm, a logistic regression algorithm, and a support vector machine algorithm.

10. A feature processing device for assisting disease diagnosis, characterized in that: The device includes a processor and a memory, wherein the memory is used to store a program executable by the processor, and the processor is used to read the program in the memory and execute the steps of the auxiliary diagnosis method for the disease according to any one of claims 1 to 9.

11. A computer storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the feature processing method for assisting disease diagnosis as described in any one of claims 1 to 9 are implemented.

Images