CN111968740A - Diagnostic label recommendation method and device, storage medium and electronic equipment - Google Patents
Diagnostic label recommendation method and device, storage medium and electronic equipment Download PDFInfo
- Publication number
- CN111968740A CN111968740A CN202010918380.2A CN202010918380A CN111968740A CN 111968740 A CN111968740 A CN 111968740A CN 202010918380 A CN202010918380 A CN 202010918380A CN 111968740 A CN111968740 A CN 111968740A
- Authority
- CN
- China
- Prior art keywords
- label
- sample
- target sample
- samples
- probability
- 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.)
- Granted
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
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- 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)
- Medical Informatics (AREA)
- Public Health (AREA)
- Data Mining & Analysis (AREA)
- Biomedical Technology (AREA)
- Databases & Information Systems (AREA)
- Pathology (AREA)
- Epidemiology (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Medical Treatment And Welfare Office Work (AREA)
Abstract
The application provides a diagnostic label recommendation method, a diagnostic label recommendation device, a storage medium and electronic equipment, wherein the recommendation method comprises the following steps: obtaining a target sample of a patient, the target sample including diseased features of the patient; selecting K adjacent samples from a known sample set of the same department as the target sample, wherein each known sample in the known sample set comprises at least one label, and the K adjacent samples share L labels; respectively calculating probability information of each label in the L kinds of labels in the K adjacent samples, and calculating a matching probability value of the label matched with the target sample according to the probability information of each label; and determining at least one label from the L labels according to the matching probability value of each label to be used as a recommended label. By combining the existing diagnosis decision, the diagnosis decision similar to the new diseased characteristic can be found out, and the found diagnosis decision is recommended to a doctor, so that the diagnosis and treatment speed of the doctor can be increased, the analysis links of the doctor are reduced, and the careless analysis of the doctor is avoided.
Description
Technical Field
The invention relates to the technical field of medical assistance, in particular to a diagnostic label recommendation method, a diagnostic label recommendation device, a storage medium and electronic equipment.
Background
With the rapid increase of medical resource demand, improving the execution efficiency and accuracy of hospitals is the key to improving medical strength. How to provide reference for future diagnosis by using the accumulated medical experience is an important step for improving the medical efficiency and the diagnosis accuracy.
The current hospital diagnosis modes mainly comprise: (1) doctors give diagnosis and treatment schemes by virtue of medical experience; (2) the medical auxiliary system based on word frequency recommendation is utilized to recommend some possible diagnosis and treatment schemes to doctors, the recommendation principle is simple, and the calculation amount is small. However, both of these approaches have drawbacks: firstly, the traditional doctor diagnosis needs to rely on personal experience of doctors, whether the diagnosis is accurate or not is related to the knowledge and experience of the doctors, meanwhile, the judgment of the doctors is time-consuming, and the manual judgment is relatively easy to make mistakes; the latter is a medical assistance system, and the word frequency recommendation has the disadvantages that rare word frequencies cannot be recommended and long tails cannot be considered, for example, for the symptoms of patients, the recommendation is performed only based on the occurrence frequency of the symptom keywords, but the factors such as sex, age and the like of the patients cannot be considered, and the recommendation accuracy is not enough.
Disclosure of Invention
An embodiment of the present application provides a diagnostic tag recommendation method, an apparatus, a storage medium, and an electronic device, so as to solve the above technical problem.
In order to achieve the above purpose, the present application provides the following technical solutions:
in a first aspect, an embodiment of the present application provides a diagnostic tag recommendation method, including: obtaining a target sample of a patient, the target sample including a diseased feature of the patient; selecting K neighboring samples from a known sample set of the same department as the target sample, wherein each known sample in the known sample set comprises at least one label, and the label is a diagnosis decision made by a doctor according to a diseased feature in the known sample, and the K neighboring samples have L labels in total; respectively calculating probability information of each label in the L labels in the K adjacent samples, and calculating matching probability values of the labels matched with the target sample according to the probability information of each label to obtain L matching probability values; and determining at least one label from the L kinds of labels according to the matching probability value of each kind of label to be used as a recommended label.
According to the method and the device, the diagnosis decision similar to the new diseased characteristic can be found out by combining the diagnosis decisions in the known sample, so that the found existing diagnosis decision can be recommended to a doctor, the diagnosis and treatment speed of the doctor is further accelerated, the analysis links of the doctor are reduced, the occurrence of careless analysis of the doctor is avoided, and the accuracy is higher compared with a word frequency recommendation scheme.
In an alternative embodiment, the calculating a matching probability value of the tag matching the target sample according to the probability information of each tag includes: for each tag, the match probability value J is calculated by the following formula:
J＝P(HL＝N 1)·P(EL＝N C1|HL＝N 1)-P(HL＝N 0)·P(EL＝N C0|HL＝N 0)；
wherein, P (H)L＝N 1) The first prior probability of the label represents the probability of the label existing in K adjacent samples of the target sample; p (H)L＝N 0) A second prior probability of the label, representing K neighbors in the target sampleThe probability that the label is not present in the sample; p (E)L＝N C1|HL＝N 1) The first posterior probability of the label represents the probability that the number of the labels in the K adjacent samples corresponding to the target sample is C1 under the condition that the label exists in the K adjacent samples of the target sample, and C1 is the number of the labels in the K adjacent samples of the target sample; p (E)L＝N C0|HL＝N 0) The second posterior probability of the label indicates that, under the condition that the label does not exist in the K neighbor samples of the target sample, the number of the labels that do not exist in the K neighbor samples corresponding to the neighbor samples of the target sample is the probability of C0, and C0 is the number of the labels that do not exist in the K neighbor samples of the target sample.
After L labels corresponding to K neighbor samples are obtained, probability information is calculated for each label in the L labels respectively, wherein the probability information comprises a first prior probability, a second prior probability, a first posterior probability and a second posterior probability of the label, and then a matching probability value can be obtained based on the first prior probability, the second prior probability, the first posterior probability and the second posterior probability.
In an alternative embodiment, the diseased features include n features, and n is an integer greater than or equal to 1; the selecting K neighboring samples from a known sample set of the same department as the target sample comprises:
calculating the similarity f of the target sample and each known sample in the set of known samples by the following formulamatch：
Wherein, MCiRepresenting the number of matches of the ith feature in the diseased features of the target sample with the ith feature in the diseased features of the known sample, wiRepresents the weight corresponding to the ith feature, CiInformation indicating the number of the i-th feature, i being taken from 1 to n；
Taking the first K known samples with high similarity as the K adjacent samples.
Each of the diseased features has different weights in the calculation process according to different degrees of importance for diagnosis decision, and the calculated similarity can be more accurate.
In an optional embodiment, the method further comprises: according to the recommendation label of each target sample in a preset time period and the fact label of each target sample, counting the recommendation hit rate in the preset time period, wherein the fact label of each target sample is a diagnosis decision actually made by a doctor according to the diseased features in the target sample; and when the recommended hit rate is not greater than a first preset value, adjusting the number K of the neighbor samples of the target sample.
The recommendation method has self-learning capability, the K value can be adjusted in time according to the recommendation hit rate, self deviation is corrected, and the recommendation result is guaranteed to have better accuracy.
In an optional implementation, the adjusting the number K of neighbor samples of the target sample includes: dividing a known sample set of the department into a training sample set and a testing sample set; obtaining the adjustment range of K; selecting an unselected K value from the adjustment range, and recording the unselected K value as K0; for each test sample in the test sample set, selecting K0 neighbor samples corresponding to the test sample from the training sample set, and determining a recommended label corresponding to the test sample according to the K0 neighbor samples; calculating the recommended hit rate of the test sample set according to the recommended label of each test sample and the fact label of each test sample, wherein the fact label of each test sample is a diagnosis decision actually made by a doctor according to the diseased features in the test samples; judging whether the recommended hit rate corresponding to the K0 is greater than the first preset value; if the recommended hit rate corresponding to the K0 is greater than the first preset value, determining the K0 as the number K of the neighbor samples of the target sample.
In an optional implementation manner, after determining whether the recommendation hit rate corresponding to the K0 is greater than the first preset value, the method further includes: if the recommended hit rate corresponding to the K0 is not greater than the first preset value, judging whether the K values in the adjustment range are all selected; if all the K values are selected, selecting a target K value with the highest recommended hit rate from the K values in the adjustment range, and determining the target K value as the number K of the neighbor samples of the target sample; otherwise, the step of selecting an unselected K value from the adjustment range is skipped.
In the above embodiment, the K value is continuously selected from the preset adjustment range, the recommendation hit rate corresponding to the K value is obtained, and then a new K value is determined according to the recommendation hit rate, and the accuracy of the obtained recommendation result is higher for the new K value than the original K value.
In an optional embodiment, the determining at least one label from the L kinds of labels according to the matching probability value of each label as a recommended label includes: respectively judging whether the matching probability value of each label is greater than a second preset value; and selecting labels with the matching probability value larger than a second preset value as undetermined labels, and selecting a preset number of labels from the undetermined labels according to the matching probability value to serve as the recommended labels.
In a second aspect, an embodiment of the present application provides a diagnostic tag recommendation apparatus, including: a target sample acquisition module for acquiring a target sample of a patient, the target sample including a diseased feature of the patient; a neighbor sample acquiring module, configured to select K neighbor samples from a known sample set of the same department as the target sample, where each known sample in the known sample set includes at least one label, and the label is a diagnostic decision made by a doctor according to a diseased feature in the known sample, and the K neighbor samples have L kinds of labels in total; a probability calculation module, configured to calculate probability information of each of the L kinds of tags in the K neighboring samples, respectively, and calculate a matching probability value of the tag matching the target sample according to the probability information of each of the L kinds of tags, so as to obtain L matching probability values in total; and the recommending module is used for determining at least one label from the L kinds of labels as a recommended label according to the matching probability value of each kind of label.
In a third aspect, an embodiment of the present application provides a storage medium, where a computer program is stored on the storage medium, and when the computer program is executed by a processor, the method according to any one of the first aspect and the optional implementation manner of the first aspect is performed.
In a fourth aspect, an embodiment of the present application provides an electronic device, including: a processor, a memory and a bus, the memory storing processor-executable machine-readable instructions, the processor and the memory communicating via the bus when the electronic device is running, the machine-readable instructions when executed by the processor performing the method according to any one of the first aspect and the optional implementation manner of the first aspect.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are required to be used in the embodiments of the present application will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered as limiting the scope, and that those skilled in the art can also obtain other related drawings based on the drawings without inventive efforts.
FIG. 1 is a flow chart illustrating a diagnostic tag recommendation method provided by an embodiment of the present application;
FIG. 2 illustrates a flow chart of self-learning of a diagnostic tag recommendation method provided by an embodiment of the present application;
fig. 3 is a schematic diagram illustrating a diagnostic tag recommendation apparatus according to an embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be described below with reference to the drawings in the embodiments of the present application. It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, it need not be further defined and explained in subsequent figures. The terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element. The terms "first," "second," "third," and the like are used solely to distinguish one from another and are not to be construed as indicating or implying relative importance.
In order to solve the technical problems in the prior art, an embodiment of the present application provides a diagnostic tag recommendation method, please refer to fig. 1, which includes the following steps:
step 110: a target sample of a patient is obtained, the target sample including a diseased feature of the patient.
Step 120: k adjacent samples are selected from a known sample set of the same department as the target sample, and the K adjacent samples share L labels.
The set of known samples includes a plurality of known samples, each known sample including: a diseased characteristic and at least one label corresponding to the diseased characteristic, the label being a diagnostic decision made by a physician based on the diseased characteristic in a known sample. The diseased features and labels in the known sample are historical information and the target sample for the patient may be current real-time information. After obtaining a target sample of a patient, K neighboring samples which are closest to the target sample are selected from a known sample set, each neighboring sample has at least one label, and the labels among the K neighboring samples can be repeated, wherein the total number of the labels is L. It can be understood that if the department visited by the patient is the department of gastroenterology, K neighboring samples are selected from the known sample set corresponding to the department of gastroenterology, if the department visited by the patient is the department of respiration, K neighboring samples are selected from the known sample set corresponding to the department of respiration, and for the target samples of different departments, the known sample set corresponding to the department is selected.
In this embodiment, the diagnostic label recommendation method is used to infer a label to which a new diseased feature belongs according to the past historical diseased feature and the corresponding label.
In this embodiment, the disease condition characteristic may be a disease condition of the patient, the disease condition is a set of a series of characteristic attributes of the patient, the disease condition may include a plurality of symptoms, sex, age, past medical history, genetic medical history and other factors of the patient, which are associated with the diagnosis to different degrees, and the label is a diagnosis decision made by the doctor according to the disease condition, for example, a name of the disease condition.
In this embodiment, the disease characteristic may also be at least one disease name of the patient, and the label is a diagnosis decision made by the doctor according to the at least one disease name, for example, a check item corresponding to the disease name, for example, the label may be a blood test, a urine test, a CT photograph, etc. Therefore, according to the recommendation method, a plurality of possible labels can be deduced according to the disease name of the patient and the relationship between the past disease name stored in the database and the issued examination/test item.
In the above embodiment, the disease is characterized by at least one disease name of the patient, and the label may be a drug corresponding to the at least one disease name, so that, by the recommendation method, a plurality of possible labels can be inferred from the disease name of the patient and the relationship between the past disease name and the prescribed drug stored in the database.
For convenience of understanding, the technical solution will be described below by taking the case of deducing the name of a disease corresponding to the disease condition of a patient.
In the step 110, the disease condition information of the patient, which is the target sample of the patient, is obtained, and the disease condition information is divided into multiple features, such as department, symptom, physical examination result, sex, age, past medical history, and genetic medical history.
In the above step 120, similarity calculation is performed on each known sample in the known sample set and the target sample to obtain K neighboring samples, which share L kinds of labels. Before that, the diagnosis and treatment records in the database can be traversed, and a known sample set is obtained according to all the diagnosis and treatment records of the same department as the target sample, wherein each known sample in the known sample set corresponds to one diagnosis and treatment record.
Step 130: and respectively calculating probability information of each label in the L labels in the K adjacent samples.
Step 140: and calculating the matching probability value of the label and the target sample according to the probability information of each label, and obtaining L matching probability values.
Step 150: and determining at least one label from the L labels according to the matching probability value of each label to be used as a recommended label.
In the step 130, after L kinds of labels corresponding to the K neighboring samples are obtained, probability information is calculated for each of the L kinds of labels. Wherein, the probability information comprises a first prior probability, a second prior probability, a first posterior probability and a second posterior probability of the label.
Specifically, step 130 includes:
1. calculating a first prior probability P (H) of such a labelL＝N 1)
P(HL＝N 1) The first prior probability of such a label represents the probability that such a label is present in the K neighbor samples of the target sample. Counting the frequency of each label in K adjacent samples, and assuming that a certain label appears in A1 adjacent samples in the K adjacent samples, the first prior probability P (H) of the certain labelL＝N 1)＝A1/K。
2. Calculating a second prior probability P (H) of such labelL＝N 0)
P(HL＝N 0) The second prior probability of the label represents the probability that the label does not exist in the K adjacent samples of the target sample. The frequency of each label not appearing in K adjacent samples is counted, and A0 near samples of a certain label in the K adjacent samples are assumedIf no occurrence in the neighbor sample, then a second prior probability P (H) of that type of labelL＝N 0)＝A0/K。
Since the denominators of the first prior probability and the second prior probability are the number K of neighboring samples, the sum of the first prior probability and the second prior probability is 1. For example, if the number K of neighboring samples is 10, and a certain label appears in 4 of the 10 neighboring samples and does not appear in the remaining 6 neighboring samples, the first prior probability is 4/10, and the second prior probability is 6/10.
3. Calculating a first posterior probability P (E) of such a labelL＝N C1|HL＝N 1)
P(EL＝N C1|HL＝N 1) The first posterior probability of the type of label represents the probability that the number of the type of label in the K neighboring samples corresponding to the neighboring sample of the target sample is C1 under the condition that the type of label exists in the K neighboring samples of the target sample, and C1 is the number of the type of label in the K neighboring samples of the target sample.
Firstly, counting the label L in K adjacent samples of the target sampleNThe number of occurrences, denoted C1, is for example the label "cold" occurred in 3 of the K neighbor samples, and the value of C1 is 3.
Then, the known sample set corresponding to the department is traversed, and K adjacent samples are selected again from the known sample set for each adjacent sample of the target sample. For example, K neighbor samples of the target sample are respectively neighbor sample K1, neighbor sample K2, neighbor sample K3, and …, K neighbor samples of neighbor sample K1 are selected from the known sample set, K neighbor samples of neighbor sample K2 are selected from the known sample set, K neighbor samples of neighbor sample K3 are selected from the known sample set, and so on until K neighbor samples corresponding to each neighbor sample of the target sample are obtained.
For convenience of description, a neighboring sample of the target sample is temporarily referred to as a first neighboring sample, and a neighboring sample of the first neighboring sample is referred to as a second neighboring sample.
Then, the user can use the device to perform the operation,according to K second neighbor samples corresponding to each first neighbor sample of the target sample, finding out the label L in the K second neighbor samples of each first neighbor sampleNThe number of first neighbor samples that occur the same number of times as C1 is denoted as EL＝N C1。
Specifically, the label L is obtained by the following formulaNFirst posterior probability of (a):
P(EL＝N C1|HL＝N 1)＝EL＝N C1/K＝(E1+E2+E3+…+EK)/K；
wherein E1 represents the label L in K first neighbor samples of the target sample, K second neighbor samples of the first neighbor sample K1NWhether the number of occurrences is the same as that of C1, if so, the value of E1 is denoted by 1, and if not, the value of E1 is denoted by 0. For example, if the label "cold" in K second neighboring samples of the first neighboring sample K1 also appears 3 times, then E1 is recorded as 1, and then the calculation of the value E2 corresponding to the next first neighboring sample K2 is continued. In the above formula, E2 to EK are calculated based on the corresponding first neighboring samples, and the meaning and calculation method are the same as those of E1, which is not repeated herein.
4. Calculating a second posterior probability P (E) of such labelL＝N C0|HL＝N 0)
P(EL＝N C0|HL＝N 0) The second posterior probability of the label indicates that, under the condition that the label does not exist in the K neighbor samples of the target sample, the number of the labels that do not exist in the K neighbor samples corresponding to the neighbor samples of the target sample is the probability of C0, and C0 is the number of the labels that do not exist in the K neighbor samples of the target sample.
Firstly, counting the label L in K adjacent samples of the target sampleNThe number of non-occurrences, denoted C0, e.g., the label "cold" does not occur in 7 of the K neighbor samples, then the value of C0 is 7.
Then, the known sample set corresponding to the department is traversed, and K adjacent samples are selected again from the known sample set for each adjacent sample of the target sample. For example, K neighbor samples of the target sample are respectively neighbor sample K1, neighbor sample K2, neighbor sample K3, and …, K neighbor samples of neighbor sample K1 are selected from the known sample set, K neighbor samples of neighbor sample K2 are selected from the known sample set, K neighbor samples of neighbor sample K3 are selected from the known sample set, and so on until K neighbor samples corresponding to each neighbor sample of the target sample are obtained.
It will be appreciated that the above-described step of selecting K neighbor samples of each neighbor sample of the target sample may be performed only once when calculating the first posterior probability and the second posterior probability.
For convenience of description, a neighboring sample of the target sample is temporarily referred to as a first neighboring sample, and a neighboring sample of the first neighboring sample is referred to as a second neighboring sample.
Then, according to K second neighboring samples corresponding to each first neighboring sample of the target sample, finding out the label L in the K second neighboring samples of each first neighboring sampleNThe number of first neighbor samples that do not occur as many times as C0 is denoted as EL ＝N C0。
Specifically, the label L is obtained by the following formulaNSecond posterior probability of (2):
P(EL＝N C0|HL＝N 0)＝EL＝N C0/K＝(F1+F2+F3+…+FK)/K；
wherein F1 represents the label L in K first neighbor samples of the target sample, K second neighbor samples of the first neighbor sample K1NWhether the number of non-occurrences is the same as that of C0 or not is judged, and if so, the value of F1 is 1, and if not, the value of F1 is 0. For example, if the number of times that the label "cold" does not appear in the K second neighbor samples of the first neighbor sample K1 is also 7, then F1 is recorded as 1, and then the calculation of the value F2 corresponding to the next first neighbor sample K2 is continued. In the above formula, F2 to FK are calculated based on the corresponding first neighboring samples, and the meaning and calculation method are the same as F1, which is not described herein again.
After the first prior probability, the second prior probability, the first posterior probability and the second posterior probability of each label are obtained through calculation, probability information of each label is obtained, and then step 140 is executed.
In the above step 140, the matching probability value J of each tag is calculated by the following formula:
J＝P(HL＝N 1)·P(EL＝N C1|HL＝N 1)-P(HL＝N 0)·P(EL＝N C0|HL＝N 0)；
wherein, P (H)L＝N 1) Is the first prior probability, P (E), of such a labelL＝N C1|HL＝N 1) Is the first posterior probability of such a label, P (H)L＝N 0) Is the second prior probability, P (E), of such a labelL＝N C0|HL＝N 0) Is the second posterior probability of such a tag.
In the above step 150, it is determined whether to select each of the tags as a recommended tag according to the calculated matching probability value of the tag.
In an embodiment, whether the matching probability value of each label is greater than a second preset value or not is respectively judged, and the labels with the matching probability values greater than the second preset value are all taken as recommended labels. For example, the second preset value may be 0.4, and if the matching probability value of a certain tag is greater than 0.4, the certain tag may be regarded as a recommended tag.
In another embodiment, the tags with the matching probability value larger than the second preset value are used as undetermined tags, the undetermined tags are sorted according to the size of the matching probability value, and a preset number of tags are selected from the undetermined tags according to the sequence from large to small to serve as recommended tags. Similarly, the second preset value may be 0.4, and the tag with the matching probability value greater than 0.4 is taken as the pending tag. In a specific embodiment, three tags with the highest matching probability values are selected from the tags to be determined as recommended tags.
After the recommendation label is determined, the recommendation label can be recommended to a doctor and displayed on a page of a doctor terminal for the doctor to select, in addition, an actual diagnosis decision made by the doctor according to the diseased features in the target sample is obtained, a fact label of the target sample is obtained and is input into a database to be used as a basis for recommending the diagnosis labels of other target samples in the future.
Optionally, in step 120, K neighboring samples are screened out by calculating the similarity between the target sample and each known sample.
The disease characteristics of the target sample comprise n characteristics, wherein n is an integer greater than or equal to 1, if the disease characteristics are the illness state of the patient, the n characteristics respectively comprise the symptoms, the sex, the age, the past medical history, the genetic medical history and the like of the patient, and if the disease characteristics are the illness names of the patient, each characteristic corresponds to one disease name. The formula for calculating the similarity between the target sample and the known sample is as follows:
wherein f ismatchFor similarity between the target sample and the known sample, MCiRepresenting the number of matches of the ith feature in the diseased features of the target sample with the ith feature in the diseased features of the known sample, wiRepresents the weight corresponding to the i-th feature, CiInformation indicating the number of the i-th feature, i being taken from 1 to n. Each of the disease features is weighted differently in the calculation process according to the degree of importance of the disease feature in the diagnosis decision, for example, the past medical history, the genetic medical history, etc. are weighted less than the symptom, sex, age, etc. features.
Taking the disease characteristics as an example, assuming that the disease characteristics include five characteristics, namely symptoms, sex, age, past medical history and family/genetic medical history, a complete similarity calculation formula is obtained:
wherein, MCsymptomIs a target sample and is knownNumber of symptom matches between samples, WsymptomWeights corresponding to symptom characteristics; MC (monomer casting)genderIs the number of gender matches, the value is 1 if the gender characteristics of the patient between the target sample and the known sample match, otherwise the value is 0, WgenderThe weight corresponding to the sex characteristics; MC (monomer casting)ageIs the age group match number, the value is 1 if the age group of the patient matches between the target sample and the known sample, otherwise the value is 0, WageThe weight corresponding to the age characteristic; MC (monomer casting)pmhNumber of past medical history matches, WpmhThe weight corresponding to the characteristics of the prior medical history; MC (monomer casting)fmhNumber of family/genetic history matches, WfmhWeights corresponding to family/genetic history characteristics; csymptomTotal number of symptoms in the target sample; cgenderThe number of gender characteristics of the target sample is 1 by default; cageThe number of age features of the target sample, the value of which is 1 by default; cpmhThe past medical history of the target sample; cfmhFamily/genetic history of the target sample.
After the similarity between the target sample and each known sample is obtained through calculation, the previous K known samples with high similarity are taken as K adjacent samples of the target sample.
In the calculation process of the similarity, unlike the matching of gender and age, the symptom data is recorded in a character string manner, and thus the matching of symptoms is generally performed by vocabulary matching. Generally, the description of the symptom includes, in addition to the keyword of the symptom, a description of a degree of the keyword, such as "fever 3 days", "mild headache", and the like, since in this embodiment, the diagnosis and treatment records of the target sample and the known sample belonging to the same department do not differ too much in duration and severity of the disease condition, the descriptions of the degree of the disease condition may be ignored, and for example, the descriptions of "fever 1 day", "fever 2 days", and "fever" are considered to be matched.
In this embodiment, the diagnostic label recommendation method is to judge the label to which the target sample belongs based on K neighboring samples of the target sample and all known samples together, where the neighboring samples are mainly used to calculate the prior probability, all known samples are mainly used to calculate the posterior probability, and the value of the number K of the neighboring samples will affect the accuracy of the finally obtained recommended label. The value of K must be appropriate, and an excessively large value of K will incorporate a label with low correlation with the target sample, while an excessively small value of K will amplify the influence of the noise sample.
The value of K can be set according to the test effect of the test sample set or the actual use effect of the hospital. In one embodiment, K is in the range of 5 to 10.
Optionally, the method further includes: according to the recommendation label of each target sample in a preset time period and the fact label of each target sample, calculating the recommendation hit rate in the preset time period; and when the recommended hit rate is not greater than the first preset value, adjusting the number K of the neighbor samples of the target sample. Wherein the fact label of the target sample is a diagnostic decision actually made by the physician based on the diseased features in the target sample. Illustratively, the recommendation hit rate is calculated according to the recommendation label given to each target sample and the corresponding fact label every morning at 3:00 a day, if the recommendation hit rate is lower than 0.65 (a first preset value), the number K of the neighbor samples is adjusted, and of course, the recommendation hit rate can be calculated and updated in real time every day.
In this embodiment, the K value may be adjusted in a self-learning manner, and the self-learning step refers to the flow illustrated in fig. 2, which includes:
step 210: the known sample set of the department is divided into a training sample set and a testing sample set.
Step 220: an adjustment range for K is determined.
In one embodiment, K is adjusted to a range of [5,10], so K can be selected from 5, 6, 7, 8, 9, 10.
Step 230: an unselected value of K is selected from the adjustment range and is marked as K0.
Step 240: for each test sample in the test sample set, selecting K0 neighbor samples corresponding to the test sample from the training sample set, and determining a recommended label corresponding to the test sample according to the K0 neighbor samples.
The step of determining the recommended label corresponding to each test sample is consistent with the step of determining the recommended label corresponding to the target sample, which is not repeated herein, and reference may be made to the description of the foregoing embodiments.
Step 250: and calculating the recommended hit rate of the test sample set according to the recommended label of each test sample and the fact label of each test sample.
Wherein the fact label of each test sample is a diagnosis decision actually made by a doctor in the past according to the diseased features in the test sample.
Step 260: judging whether the recommended hit rate corresponding to K0 is greater than a first preset value; if the recommendation hit rate corresponding to the K0 is greater than the first preset value, jumping to step 270; otherwise, go to step 280.
Step 270: this K0 is determined as the number K of neighbor samples of the target sample.
And obtaining K0 neighbor samples according to the K0 selected in the step 230, calculating the recommended hit rate of the test sample set based on the K0 neighbor samples, and if the recommended hit rate corresponding to the K0 is greater than a first preset value, using the K0 as the number K of neighbor samples when labels are recommended to the target sample later, and finishing the self-learning step.
Step 280: judging whether the K values in the adjusting range are all selected; if all the selection is already done, go to step 290; otherwise, go to step 230.
Step 290: and selecting a target K value with the highest recommended hit rate from the plurality of K values in the adjustment range, and determining the target K value as the number K of the neighbor samples of the target sample.
And if the recommended hit rates corresponding to all the K values in the adjustment range are not larger than the first preset value, selecting one K value with the highest recommended hit rate, taking the K value as the number K of the neighbor samples when the labels are recommended to the target samples later, and finishing the self-learning step. Meanwhile, an alarm mail can be sent to the technical department to indicate manual intervention and parameter adjustment.
Further, this embodiment provides a technical solution of a full-flow diagnosis and treatment, that is, the method includes each stage of diagnosis and treatment, for example, possible diagnosis results of disease names are given according to disease conditions, possible diagnosis results of examination items or test items are given according to diagnosis of disease names, and possible diagnosis results of drugs are given according to diagnosis of disease names, so as to provide recommendations for doctors in each stage of diagnosis and treatment.
Illustratively, a path of a full-flow medical treatment is as follows:
(1) when a patient goes to a doctor in a hospital, the illness state information of the patient is obtained, three recommendation labels with the highest matching degree with the illness state information are determined through the recommendation method, three options are displayed on a doctor terminal, each option corresponds to an illness name represented by one recommendation label, and a doctor only needs to select the displayed three options according to an actually made illness name diagnosis result, so that the information entry process of the doctor is simplified. If the three given options do not have the disease name diagnosis result actually made by the doctor, the doctor can input the disease name diagnosis result in other input modes, for example, a large category of the disease name is selected in a menu bar of a page and then selected from each small category of the large category, or a first spelling of the disease name diagnosis result is input in an input bar on the page, and then a matched result can be automatically filtered out.
(2) After a disease name diagnosis result given by a doctor is obtained, the disease name diagnosis result is a preliminary diagnosis result, 1-2 recommended labels with the highest matching degree with the preliminary diagnosis result are determined through the recommendation method, 1-2 options are displayed on a doctor terminal, each option corresponds to an examination/inspection item represented by one recommended label, and the doctor can check from the displayed 1-2 options according to the actually made examination/inspection item diagnosis result.
(3) After the patient finishes the examination/inspection, the doctor can give the disease name diagnosis result again according to the examination/inspection result of the patient, the disease name diagnosis result is the final diagnosis result, a plurality of recommendation labels with the highest matching degree with the final diagnosis result are determined through the recommendation method, a plurality of options are displayed on the doctor terminal, each option corresponds to the medicine represented by one recommendation label, and the doctor only needs to select the option from the displayed options according to the actually made medicine diagnosis result.
The embodiment of the application is combined with the existing diagnosis decision to find out the diagnosis decision similar to the new diseased characteristic, and the existing diagnosis decision is recommended to a doctor, so that the diagnosis and treatment speed of the doctor is increased, the analysis links of the doctor are reduced, and the occurrence of careless analysis of the doctor is avoided; in addition, the recommendation method can be used for guiding training by doctors with higher levels and popularized to doctors with less experience for use, so that a good guiding effect can be provided for primary doctors, the diagnosis and treatment difficulty is reduced, and the diagnosis and treatment level of the primary doctors is improved. Moreover, the recommendation method has self-learning capability, can correct self deviation in time, and ensures that the recommendation result has better accuracy.
It is understood that, in the present embodiment, the similarity between the target sample and the known sample may also be calculated by using a similarity algorithm based on the extensions of the euler distance, the included vector angle, the degree of coincidence, and the like.
It is understood that the disease characteristics in this embodiment may include other characteristics besides the above listed characteristics such as symptoms, physical examination results, sex, age, past medical history, and genetic medical history, for example, the work and rest habits, eating habits, and bad taste of the patient may be used as one of the characteristics of the disease, and of course, some of the above listed characteristics may be discarded.
Based on the same inventive concept, an embodiment of the present application provides a diagnostic tag recommendation apparatus, please refer to fig. 3, the apparatus includes: a target sample acquisition module 310, a neighbor sample acquisition module 320, a probability calculation module 330, and a recommendation module 340.
Wherein the target sample acquiring module 310 is configured to acquire a target sample of a patient, the target sample including a diseased feature of the patient.
A neighboring sample obtaining module 320, configured to select K neighboring samples from a known sample set of the same department as the target sample, where each known sample in the known sample set includes at least one label, and the label is a diagnostic decision made by a doctor according to a diseased feature in the known sample, and the K neighboring samples have L kinds of labels in total.
A probability calculating module 330, configured to calculate probability information of each of the L kinds of labels in the K neighboring samples, respectively, and calculate a matching probability value of the label matching the target sample according to the probability information of each label, so as to obtain L matching probability values in total.
And the recommending module 340 is configured to determine at least one label from the L kinds of labels according to the matching probability value of each kind of label, and use the determined at least one label as a recommended label.
Optionally, the probability calculating module 330 is specifically configured to calculate the matching probability value J by using the following formula for each tag: j ═ P (H)L＝N 1)·P(EL＝N C1|HL＝N 1)-P(HL＝N 0)·P(EL＝N C0|HL＝N 0) (ii) a Wherein, P (H)L＝N 1) The first prior probability of the label represents the probability of the label existing in K adjacent samples of the target sample; p (H)L＝N 0) A second prior probability of the label, which represents the probability that the label does not exist in the K adjacent samples of the target sample; p (E)L＝N C1|HL＝N 1) The first posterior probability of the label represents the probability that the number of the labels in the K adjacent samples corresponding to the target sample is C1 under the condition that the label exists in the K adjacent samples of the target sample, and C1 is the number of the labels in the K adjacent samples of the target sample; p (E)L＝N C0|HL＝N 0) The second posterior probability of the label indicates that, under the condition that the label does not exist in the K neighbor samples of the target sample, the number of the labels that do not exist in the K neighbor samples corresponding to the neighbor samples of the target sample is the probability of C0, and C0 is the number of the labels that do not exist in the K neighbor samples of the target sample.
Optionally, the diseased features include n features, and n is an integer greater than or equal to 1; the neighbor sample acquisition module 320 is specifically configured to calculate the target sample and each known sample in the set of known samples by the following formulaSimilarity f of samplesmatch：
Wherein, MCiRepresenting the number of matches of the ith feature in the diseased features of the target sample with the ith feature in the diseased features of the known sample, wiRepresents the weight corresponding to the ith feature, CiInformation indicating the number of the ith feature, i being taken from 1 to n;
the neighboring sample obtaining module 320 is further configured to take K previous known samples with high similarity as the K neighboring samples.
Optionally, the apparatus further comprises:
the hit rate calculation module is used for counting the recommended hit rate in a preset time period according to the recommended label of each target sample in the preset time period and the fact label of each target sample, wherein the fact label of each target sample is a diagnosis decision actually made by a doctor according to the diseased features in the target samples;
and the self-learning module is used for adjusting the number K of the neighbor samples of the target sample when the recommendation hit rate is not greater than the first preset value.
Optionally, the self-learning module includes:
the sample set dividing submodule is used for dividing the known sample set of the department into a training sample set and a testing sample set;
the adjustment range acquisition submodule is used for acquiring the adjustment range of the K;
the value selection submodule is used for selecting an unselected K value from the adjustment range and recording the unselected K value as K0;
a recommended label determining submodule, configured to, for each test sample in the test sample set, select, from the training sample set, K0 neighboring samples corresponding to the test sample, and determine, according to the K0 neighboring samples, a recommended label corresponding to the test sample;
the hit rate calculation submodule is used for calculating the recommended hit rate of the test sample set according to the recommended label of each test sample and the fact label of each test sample, wherein the fact label of each test sample is a diagnosis decision actually made by a doctor according to the diseased features in the test samples;
the first judgment submodule is used for judging whether the recommendation hit rate corresponding to the K0 is greater than the first preset value or not;
and the first K value determining submodule is used for determining the K0 as the number K of the neighbor samples of the target sample when the recommended hit rate corresponding to the K0 is greater than the first preset value.
Optionally, the self-learning module further includes:
the second judgment submodule is used for judging whether the K values in the adjustment range are all selected or not when the recommended hit rate corresponding to the K0 is not greater than the first preset value; if all the K values are selected, jumping to a second K value determination submodule; otherwise, jumping to a value selection submodule so that the value selection submodule executes the step of selecting an unselected K value from the adjustment range;
and the second K value determining submodule is used for selecting a target K value with the highest recommended hit rate from the plurality of K values in the adjusting range and determining the target K value as the number K of the neighbor samples of the target sample.
Optionally, the recommending module 340 is specifically configured to respectively determine whether the matching probability value of each tag is greater than a second preset value, use the tag with the matching probability value greater than the second preset value as an undetermined tag, and select a preset number of tags from the undetermined tags according to the size of the matching probability value, so as to use the tags as the recommended tags.
The implementation principle and the technical effect of the diagnostic label recommendation device provided by the embodiment of the application have been introduced in the foregoing method embodiment, and for the sake of brief description, reference may be made to corresponding contents in the method embodiment where no part of the embodiment of the device is mentioned.
Optionally, an embodiment of the present application further provides a storage medium, where a computer program is stored on the storage medium, and when the computer program is executed by a processor, the method for recommending a diagnostic tag provided by the present application is executed.
Optionally, an embodiment of the present application further provides an electronic device, including: the diagnostic tag recommendation system comprises a processor, a memory and a bus, wherein the memory stores machine readable instructions executable by the processor, the processor and the memory are communicated through the bus when the electronic device runs, and the machine readable instructions are executed by the processor to execute the diagnostic tag recommendation method provided by the application. It is understood that the electronic device may be a physical device, such as a PC, a laptop, a tablet, a server, an embedded device, etc., or may be a virtual device, such as a virtual machine, a virtualized container, etc.
In the embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. The above-described embodiments of the apparatus are merely illustrative, and for example, the division of the unit is only a logical division, and other divisions may be realized in practice. Furthermore, the functional modules in the embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.
The above description is only an example of the present application and is not intended to limit the scope of the present application, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims (10)
1. A diagnostic tag recommendation method, comprising:
obtaining a target sample of a patient, the target sample including a diseased feature of the patient;
selecting K neighboring samples from a known sample set of the same department as the target sample, wherein each known sample in the known sample set comprises at least one label, and the label is a diagnosis decision made by a doctor according to a diseased feature in the known sample, and the K neighboring samples have L labels in total;
respectively calculating probability information of each label in the L labels in the K adjacent samples, and calculating matching probability values of the labels matched with the target sample according to the probability information of each label to obtain L matching probability values;
and determining at least one label from the L kinds of labels according to the matching probability value of each kind of label to be used as a recommended label.
2. The method of claim 1, wherein the calculating a matching probability value of the tag matching the target sample according to the probability information of each tag comprises:
for each tag, the match probability value J is calculated by the following formula:
J＝P(HL＝N 1)·P(EL＝N C1|HL＝N 1)-P(HL＝N 0)·P(EL＝N C0|HL＝N 0)；
wherein, P (H)L＝N 1) The first prior probability of the label represents the probability of the label existing in K adjacent samples of the target sample; p (H)L＝N 0) A second prior probability of the label, which represents the probability that the label does not exist in the K adjacent samples of the target sample; p (E)L＝N C1|HL＝N 1) The first posterior probability of the label represents the probability that the number of the labels in the K adjacent samples corresponding to the target sample is C1 under the condition that the label exists in the K adjacent samples of the target sample, and C1 is the number of the labels in the K adjacent samples of the target sample; p (E)L ＝N C0|HL＝N 0) The second posterior probability of the label indicates that the number of the labels in the K neighbor samples corresponding to the neighbor samples of the target sample does not exist under the condition that the label does not exist in the K neighbor samples of the target sampleThe probability of C0, C0, is the number of K nearest neighbors of the target sample for which this kind of label is not present.
3. The method according to claim 1, wherein the diseased features include n features, and n is an integer greater than or equal to 1; the selecting K neighboring samples from a known sample set of the same department as the target sample comprises:
calculating the similarity f of the target sample and each known sample in the set of known samples by the following formulamatch：
Wherein, MCiRepresenting the number of matches of the ith feature in the diseased features of the target sample with the ith feature in the diseased features of the known sample, wiRepresents the weight corresponding to the ith feature, CiInformation indicating the number of the ith feature, i being taken from 1 to n;
taking the first K known samples with high similarity as the K adjacent samples.
4. The method of claim 1, further comprising:
according to the recommendation label of each target sample in a preset time period and the fact label of each target sample, counting the recommendation hit rate in the preset time period, wherein the fact label of each target sample is a diagnosis decision actually made by a doctor according to the diseased features in the target sample;
and when the recommended hit rate is not greater than a first preset value, adjusting the number K of the neighbor samples of the target sample.
5. The method of claim 4, wherein the adjusting the number K of neighbor samples of the target sample comprises:
dividing a known sample set of the department into a training sample set and a testing sample set;
obtaining the adjustment range of K;
selecting an unselected K value from the adjustment range, and recording the unselected K value as K0;
for each test sample in the test sample set, selecting K0 neighbor samples corresponding to the test sample from the training sample set, and determining a recommended label corresponding to the test sample according to the K0 neighbor samples;
calculating the recommended hit rate of the test sample set according to the recommended label of each test sample and the fact label of each test sample, wherein the fact label of each test sample is a diagnosis decision actually made by a doctor according to the diseased features in the test samples;
judging whether the recommended hit rate corresponding to the K0 is greater than the first preset value;
if the recommended hit rate corresponding to the K0 is greater than the first preset value, determining the K0 as the number K of the neighbor samples of the target sample.
6. The method of claim 5, wherein after determining whether the recommendation hit rate corresponding to the K0 is greater than the first preset value, the method further comprises:
if the recommended hit rate corresponding to the K0 is not greater than the first preset value, judging whether the K values in the adjustment range are all selected;
if all the K values are selected, selecting a target K value with the highest recommended hit rate from the K values in the adjustment range, and determining the target K value as the number K of the neighbor samples of the target sample;
otherwise, the step of selecting an unselected K value from the adjustment range is skipped.
7. The method of claim 1, wherein the determining at least one of the L tags as a recommended tag according to the matching probability value of each tag comprises:
respectively judging whether the matching probability value of each label is greater than a second preset value;
and selecting labels with the matching probability value larger than a second preset value as undetermined labels, and selecting a preset number of labels from the undetermined labels according to the matching probability value to serve as the recommended labels.
8. A diagnostic tag recommendation device, comprising:
a target sample acquisition module for acquiring a target sample of a patient, the target sample including a diseased feature of the patient;
a neighbor sample acquiring module, configured to select K neighbor samples from a known sample set of the same department as the target sample, where each known sample in the known sample set includes at least one label, and the label is a diagnostic decision made by a doctor according to a diseased feature in the known sample, and the K neighbor samples have L kinds of labels in total;
a probability calculation module, configured to calculate probability information of each of the L kinds of tags in the K neighboring samples, respectively, and calculate a matching probability value of the tag matching the target sample according to the probability information of each of the L kinds of tags, so as to obtain L matching probability values in total;
and the recommending module is used for determining at least one label from the L kinds of labels as a recommended label according to the matching probability value of each kind of label.
9. A storage medium, having stored thereon a computer program which, when executed by a processor, performs the method according to any one of claims 1-7.
10. An electronic device, comprising: a processor, a memory and a bus, the memory storing processor-executable machine-readable instructions that when executed by the electronic device communicate over the bus between the processor and the memory, the machine-readable instructions when executed by the processor perform the method of any of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010918380.2A CN111968740B (en) | 2020-09-03 | 2020-09-03 | Diagnostic label recommendation method and device, storage medium and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010918380.2A CN111968740B (en) | 2020-09-03 | 2020-09-03 | Diagnostic label recommendation method and device, storage medium and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111968740A true CN111968740A (en) | 2020-11-20 |
| CN111968740B CN111968740B (en) | 2021-04-27 |
Family
ID=73391940
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010918380.2A Active CN111968740B (en) | 2020-09-03 | 2020-09-03 | Diagnostic label recommendation method and device, storage medium and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111968740B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113761340A (en) * | 2021-01-15 | 2021-12-07 | 北京京东拓先科技有限公司 | Information recommendation method and device, electronic equipment and computer readable medium |
| CN114093506A (en) * | 2021-11-19 | 2022-02-25 | 北京欧应信息技术有限公司 | System and storage medium for assisting disease reasoning |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106874643A (en) * | 2016-12-27 | 2017-06-20 | 中国科学院自动化研究所 | Build the method and system that knowledge base realizes assisting in diagnosis and treatment automatically based on term vector |
| CN107845424A (en) * | 2017-11-15 | 2018-03-27 | 海南大学 | The method and system of diagnostic message Treatment Analysis |
| CN108764280A (en) * | 2018-04-17 | 2018-11-06 | 中国科学院计算技术研究所 | A kind of medical data processing method and system based on symptom vector |
| CN109686445A (en) * | 2018-12-29 | 2019-04-26 | 成都睿码科技有限责任公司 | A kind of intelligent hospital guide's algorithm merged based on automated tag and multi-model |
| US20200107787A1 (en) * | 2019-05-31 | 2020-04-09 | Light AI Inc. | Image Processing of Streptococcal Infection in Pharyngitis Subjects |
| WO2020073114A1 (en) * | 2018-10-09 | 2020-04-16 | Light AI Inc. | Image processing of streptococcal infection in pharyngitis subjects |
| CN111326243A (en) * | 2020-02-05 | 2020-06-23 | 安徽科大讯飞医疗信息技术有限公司 | Triage recommendation method and device, electronic equipment and storage medium |
-
2020
- 2020-09-03 CN CN202010918380.2A patent/CN111968740B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106874643A (en) * | 2016-12-27 | 2017-06-20 | 中国科学院自动化研究所 | Build the method and system that knowledge base realizes assisting in diagnosis and treatment automatically based on term vector |
| CN107845424A (en) * | 2017-11-15 | 2018-03-27 | 海南大学 | The method and system of diagnostic message Treatment Analysis |
| CN108764280A (en) * | 2018-04-17 | 2018-11-06 | 中国科学院计算技术研究所 | A kind of medical data processing method and system based on symptom vector |
| WO2020073114A1 (en) * | 2018-10-09 | 2020-04-16 | Light AI Inc. | Image processing of streptococcal infection in pharyngitis subjects |
| CN109686445A (en) * | 2018-12-29 | 2019-04-26 | 成都睿码科技有限责任公司 | A kind of intelligent hospital guide's algorithm merged based on automated tag and multi-model |
| US20200107787A1 (en) * | 2019-05-31 | 2020-04-09 | Light AI Inc. | Image Processing of Streptococcal Infection in Pharyngitis Subjects |
| CN111326243A (en) * | 2020-02-05 | 2020-06-23 | 安徽科大讯飞医疗信息技术有限公司 | Triage recommendation method and device, electronic equipment and storage medium |
Non-Patent Citations (3)
| Title |
|---|
| AWAD ALSAID ALYOUSEF: "Latent Class Multi-Label Classification to Identify Subclasses of Disease for Improved Prediction", 《2019 IEEE 32ND INTERNATIONAL SYMPOSIUM ON COMPUTER-BASED MEDICAL SYSTEMS (CBMS)》 * |
| 何松: "面向重大疾病辅助诊疗的多组学数据整合分析方法和工具", 《中国博士学位论文全文数据库（基础科学辑）》 * |
| 贾鹏飞、胡小芳: "《玩转科技制作》", 31 December 2018 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113761340A (en) * | 2021-01-15 | 2021-12-07 | 北京京东拓先科技有限公司 | Information recommendation method and device, electronic equipment and computer readable medium |
| CN114093506A (en) * | 2021-11-19 | 2022-02-25 | 北京欧应信息技术有限公司 | System and storage medium for assisting disease reasoning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111968740B (en) | 2021-04-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11626195B2 (en) | Labeling medical scans via prompt decision trees | |
| WO2018236497A1 (en) | Medical scan assisted review system | |
| US20220172809A9 (en) | Report generating system and methods for use therewith | |
| US20200395119A1 (en) | Medical scan labeling quality assurance system | |
| CN111968740B (en) | Diagnostic label recommendation method and device, storage medium and electronic equipment | |
| US20210397996A1 (en) | Methods and systems for classification using expert data | |
| US20220037019A1 (en) | Medical scan artifact detection system and methods for use therewith | |
| US20220005565A1 (en) | System with retroactive discrepancy flagging and methods for use therewith | |
| WO2018082921A1 (en) | Precision clinical decision support with data driven approach on multiple medical knowledge modules | |
| US20230060794A1 (en) | Diagnostic Tool | |
| Zhao et al. | Construction of guideline-based decision tree for medication recommendation | |
| CN114201613B (en) | Test question generation method, test question generation device, electronic device, and storage medium | |
| CN116259396A (en) | Treatment expense prediction method, system, equipment and storage medium based on machine learning |
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 |