Medical model training method based on big data

Abstract

The present invention relates to the technical field of big data model training, and in particular to a medical model training method based on big data, comprising: screening emergency call information within a single acquisition cycle, merging emergency intelligence according to emergency events as tags, and forming corresponding basic data; when the number of resolution pools within any acquisition cycle is lower than a preset number, it is determined that the training is completed; the present invention uses emergency call information and emergency intelligence as basic training materials to train a corresponding language model, and uses the semantics generated by the language model as an auxiliary output of a medical model, and at the same time, repeatedly training the model using new data, effectively improving the universality of the medical model while effectively improving the medical model's ability to recognize fuzzy information.

Description

Technical Field

[0001] The present invention relates to the technical field of big data model training, and in particular to a medical model training method based on big data. Background Art

[0002] Medical first aid means emergency treatment. When any accident or disease occurs, the rescuer is sent to the hospital by medical staff, and the hospital provides emergency treatment for the patient according to the patient's disease or injury. First aid is an important public service and a basic livelihood guarantee. Statistics and analysis of the situations that occur during the first aid process can be used to deal with subsequent first aid situations faster and more effectively.

[0003] Chinese Patent Authorization Announcement No.: CN117077541 B discloses a method and system for efficient fine-tuning of parameters of a medical model. The method includes: collecting a data set of medical data and dividing it into a training set and a test set; introducing a convolutional inductive bias module into a Transformer model in the form of a residual link to obtain an improved Transformer model as a large model; using the training set to pre-train the improved Transformer model; using the test set to fine-tune the parameters of the convolutional inductive bias module to achieve fine-tuning of the pre-trained large model under low-data conditions. The solution proposed in this invention can achieve fine-tuning of pre-trained large models under low-data conditions, solving the problems of small amount of fine-tuning data and large domain information gap in medical scenarios.

[0004] However, the above method has the problem of low recognition accuracy for ambiguous speech. Summary of the invention

[0005] To this end, the present invention provides a medical model training method based on big data to overcome the problem of low recognition accuracy for ambiguous speech in the prior art.

[0006] To achieve the above object, the present invention provides a medical model training method based on big data, comprising:

[0007] Step S1, screening the emergency call information within a single collection cycle, merging the emergency intelligence according to the emergency event as a mark, and forming corresponding basic data;

[0008] The merging is to compare the emergency intelligence in each corresponding emergency call according to the address, number of patients and patient information corresponding to the emergency event, and calculate the global accuracy of the emergency intelligence and the emergency call information;

[0009] Step S2, performing fuzzy screening on the emergency call information according to the basic data and the global accuracy, and generating a plurality of fuzzy screening pools;

[0010] The fuzzy screening is to determine a number of accuracy standards according to the global accuracy, and map the basic data into each accuracy standard to form a number of fuzzy screening pools based on accuracy standards;

[0011] Step S3, extracting a number of emergency call information in the fuzzy screening pool, and generating a resolution pool of each fuzzy pool according to the volume and recognition of each emergency call information;

[0012] The recognition degree is to use the language model to recognize the emergency call information, generate the corresponding recognition result and input the language recognition model again to extract the corresponding recognition emergency information, and compare the extracted recognition emergency information with the corresponding emergency information to determine the corresponding recognition accuracy, and output the recognition accuracy to generate the corresponding resolution;

[0013] Step S4, dividing the data to be trained according to the resolution pool, and inputting the divided training pool into the corresponding fuzzy screening pool, and performing a preset number of convolutions using the language recognition model to form a fuzzy medical model;

[0014] Step S5, connecting the fuzzy medical model to the corresponding emergency center according to the collection cycle, inputting the new emergency call information and the corresponding emergency intelligence as new training data into the fuzzy medical model, and determining the corresponding recognition degree;

[0015] Step S6, when the number of the resolution pools in any acquisition cycle is less than a preset number, it is determined that the training is completed;

[0016] The preset number is the maximum number of convolutions, which is positively correlated with the sample size;

[0017] The preset number is proportional to the sample capacity and is set to a positive integer.

[0018] Furthermore, in the step S1, when screening the emergency call information, the emergency call information is converted into text information using the language recognition model;

[0019] The first aid information is the final information of first aid completion, and at least includes the address, number of patients and patient information of a single first aid.

[0020] Furthermore, in step S1, the global accuracy is to compare the first aid information of the single first aid with the corresponding address, number of patients and / or patient information in each first aid call information, and determine the ratio of the accurate amount of information of any one of the address, number of patients or patient information to the total amount of information.

[0021] Furthermore, in the step S2, the number of the accuracy standards is not more than the number of the emergency intelligence.

[0022] Further, when forming the fuzzy screening pool, the number of the global accuracies is used to generate a corresponding number of fuzzy screening pools;

[0023] A single fuzzy screening pool corresponds to a range of global accuracies.

[0024] Furthermore, for the single basic data, it is allocated in a corresponding fuzzy screening pool and a corresponding resolution pool.

[0025] Furthermore, in the step S3, when screening the recognition of a single emergency call, it includes:

[0026] Step S301, amplifying the volume of the emergency call information and inputting it into the language recognition model to generate corresponding recognition text information;

[0027] Step S302, inputting the recognized text information into the language recognition model to form secondary recognition information;

[0028] Step S303, inputting the secondary recognition information into the language recognition model to form semantic recognition information;

[0029] Step S304, comparing the semantic recognition information with the corresponding emergency intelligence, determining the corresponding global accuracy, and outputting it as the recognition accuracy rate.

[0030] Furthermore, in step S4, the fuzzy medical model outputs the semantic recognition information.

[0031] Furthermore, in the step S5, the new emergency call information and the corresponding emergency intelligence correspond to the collection period.

[0032] Furthermore, in step S6, when the resolution pool is set, the corresponding resolution range is a fixed interval. When there is no corresponding basic data for a single fixed interval, it is determined that the resolution pool is canceled.

[0033] Compared with the prior art, the beneficial effect of the present invention lies in that the corresponding language model is trained by using emergency call information and emergency intelligence as basic training materials, and the semantics generated by the language model is used as an auxiliary output of the medical model. At the same time, the model is repeatedly trained with new data, which effectively improves the universality of the medical model and the medical model's ability to recognize fuzzy information.

[0034] Furthermore, by distinguishing the emergency information, the complexity of the basic information is effectively increased while the emergency information is refined, thereby further improving the medical model's ability to recognize fuzzy information.

[0035] Furthermore, by setting up a fuzzy screening pool and a resolution pool, the degree of data merging is effectively improved, thereby further improving the medical model's ability to recognize fuzzy information.

[0036] Furthermore, by screening the recognition degree of a single emergency call, the recognition information is repeatedly input into the corresponding language recognition model, thereby self-training the language model. This effectively improves the utilization rate of basic data while further improving the medical model's ability to recognize ambiguous information. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A flowchart of the medical model training method for big data of the present invention;

[0038] Figure 2 The flowchart of screening the recognition degree of a single emergency call according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0039] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0040] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0041] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is merely for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0042] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0043] See also Figure 1 As shown, it is a flow chart of the medical model training method based on big data of the present invention, and the medical model training method based on big data includes:

[0044] Step S1, screening the emergency call information within a single collection cycle, merging the emergency intelligence according to the emergency event as a mark, and forming corresponding basic data;

[0045] The combination is to compare the emergency information in each emergency call according to the address, number of patients and patient information corresponding to the emergency event, and calculate the global accuracy of the emergency information and the emergency call information;

[0046] Step S2, fuzzy screening the emergency call information according to the basic data and the global accuracy, and generating a plurality of fuzzy screening pools;

[0047] Fuzzy screening is to determine several accuracy standards according to the global accuracy, and map the basic data into each accuracy standard to form several fuzzy screening pools based on the accuracy standard;

[0048] Step S3, extracting a number of emergency call information in the fuzzy screening pool, and generating a resolution pool of each fuzzy pool according to the volume and recognition of each emergency call information;

[0049] The recognition degree is to use the language model to recognize the emergency call information, generate the corresponding recognition result and input the language recognition model again to extract the corresponding recognition emergency information, and compare the extracted recognition emergency information with the corresponding emergency information to determine the corresponding recognition accuracy, and output the recognition accuracy to generate the corresponding resolution;

[0050] Step S4, dividing the data to be trained according to the resolution pool, and inputting the divided training pool into the corresponding fuzzy screening pool to perform a preset number of convolutions using the language recognition model to form a fuzzy medical model;

[0051] Step S5, connecting the fuzzy medical model to the corresponding emergency center according to the collection cycle, inputting the new emergency call information and the corresponding emergency intelligence as new training data into the fuzzy medical model, and determining the corresponding recognition degree;

[0052] Step S6, when the number of resolution pools in any acquisition cycle is less than a preset number, it is determined that the training is completed;

[0053] Among them, the preset number is the maximum number of convolutions, which is positively correlated with the sample size;

[0054] The preset number is proportional to the sample size and is set as a positive integer.

[0055] By using emergency call information and emergency intelligence as basic training materials, the corresponding language model is trained, and the semantics generated by the language model is used as an auxiliary output of the medical model. At the same time, the model is repeatedly trained with new data, which effectively improves the universality of the medical model and its ability to recognize ambiguous information.

[0056] Specifically, in step S1, when screening the emergency call information, the emergency call information is converted into text information using a language recognition model;

[0057] First aid information is the final information of first aid, and at least includes the address, number of patients and patient information of a single first aid.

[0058] In implementation, for a single emergency call, the emergency information time of the emergency call is fixed as the corresponding time when the marking is completed;

[0059] Example 1: In a certain emergency call, the location information includes:

[0060] 1. Crossroad A;

[0061] 2. Shop D near intersection A;

[0062] 3. Intersection B;

[0063] 4. School gate C between intersection B and intersection A;

[0064] At this time, if the first aid is finally completed at the intersection A, the location of the first aid information corresponding to the single first aid call is recorded as the intersection A;

[0065] At the same time, among the locations of the above alarms, information containing intersection A accounts for 75% of the emergency alarms.

[0066] Specifically, in step S1, the global accuracy is to compare the first aid intelligence of a single first aid with the corresponding address, number of patients and / or patient information in each first aid call information, and determine the ratio of the accurate amount of information of any one of the address, number of patients or patient information to the total amount of information.

[0067] By distinguishing emergency information, the complexity of basic information is effectively increased while the emergency information is refined, thereby further improving the medical model's ability to recognize fuzzy information.

[0068] Example 2: In addition to Example 1, the emergency information of the other 3 emergency calls is also intersection A, and it accounts for 60%-80% of the emergency alarms. The above 4 emergency calls are merged into the same type of calls, and intersection A is the key position of the corresponding target point appearing in the calls;

[0069] In several emergency information reports with intersection A as a key location, the content may include intersection B near intersection A, school gate C, and store D. When the location is mentioned next time, intersection A can be used as a feature point for prompting to achieve the purpose of rapid positioning.

[0070] Specifically, in step S2, the number of accuracy standards is not more than the number of emergency intelligence.

[0071] Specifically, when forming the fuzzy screening pool, the number of global accuracies is used to generate a corresponding number of fuzzy screening pools;

[0072] A single fuzzy screening pool corresponds to a range of global accuracies.

[0073] Example 3: Taking Example 2 as an example, it corresponds to a basic merged data with an accuracy rate of 60%-80%. In this data, the intersection A also includes the location intersection B, the school gate C, and the canteen D;

[0074] In the other two emergency rescues, the accuracy of the emergency intelligence formed was 100% for intersection A and 100% for intersection B;

[0075] At this time, the two basic merged data are not included in the intersection A, and the accuracy rate is 60%-80%.

[0076] It can be understood that, in the division of Example 3, taking Crossroad A as an example, for Crossroad A, the range of 60%-80% on the map is smaller than the range of 100% accuracy. At this time, the emergency intelligence of the several times is divided accordingly according to the division of accuracy.

[0077] In practice, the accuracy can be divided arbitrarily. In particular, each time a division is performed, the corresponding fuzzy screening pool will increase accordingly.

[0078] Specifically, for a single basic data, it is allocated in the corresponding fuzzy screening pool and the corresponding resolution pool.

[0079] By setting up a fuzzy screening pool and a resolution pool, the degree of data merging is effectively improved, thereby further improving the medical model's ability to recognize fuzzy information.

[0080] See also Figure 2 As shown, it is a flow chart of screening the recognition of a single emergency call according to an embodiment of the present invention, including:

[0081] Step S301, amplify the volume of the emergency call information, and input it into the language recognition model to generate corresponding recognition text information;

[0082] Step S302, inputting the recognized text information into a language recognition model to form secondary recognition information;

[0083] Step S303, inputting the secondary recognition information into a language recognition model to form semantic recognition information;

[0084] Step S304, comparing the semantic recognition information with the corresponding emergency intelligence, determining the corresponding global accuracy, and outputting it as the recognition accuracy rate.

[0085] In implementation, after the corresponding text information is generated, the corresponding location information can be obtained;

[0086] Taking Example 1 as an example, the four calls are:

[0087] 1. "Someone was injured at intersection A";

[0088] 2. "I saw someone injured at intersection A at store D near intersection A";

[0089] 3. "Someone was injured at intersection B while driving from south to north";

[0090] 4. "Someone was injured in front of the school gate C between intersection B and intersection A";

[0091] For a computer, it is difficult to effectively identify the exact location by reading calls 2-4 separately. At this time, the above content is input into the corresponding language model again. Combining 1-4, it can be concluded that someone was injured near intersection A and several people witnessed it from different directions, and the corresponding location can be determined, which can effectively identify the exact location of the alarm.

[0092] By screening the recognition level of individual emergency calls and repeatedly inputting the recognition information into the corresponding language recognition model, the language model is self-trained. This effectively improves the utilization rate of basic data while further enhancing the medical model's ability to recognize ambiguous information.

[0093] Specifically, in step S4, the fuzzy medical model outputs semantic recognition information.

[0094] Specifically, in step S5, new emergency call information and corresponding emergency intelligence correspond to the collection cycle.

[0095] Specifically, in step S6, when the resolution pool is set, the corresponding resolution range is a fixed interval. When there is no corresponding basic data in a single fixed interval, it is determined that the resolution pool is canceled.

[0096] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0097] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A medical model training method based on big data, characterized in that: include: Step S1, screening the emergency call information within a single collection cycle, merging the emergency intelligence according to the emergency event as a mark, and forming corresponding basic data; The merging is to compare the emergency intelligence in each corresponding emergency call according to the address, number of patients and patient information corresponding to the emergency event, and calculate the global accuracy of the emergency intelligence and the emergency call information; Step S2, performing fuzzy screening on the emergency call information according to the basic data and the global accuracy, and generating a plurality of fuzzy screening pools; The fuzzy screening is to determine a number of accuracy standards according to the global accuracy, and map the basic data into each accuracy standard to form a number of fuzzy screening pools based on accuracy standards; Step S3, extracting a number of emergency call information in the fuzzy screening pool, and generating a resolution pool of each fuzzy pool according to the volume and recognition of each emergency call information; The recognition degree is to use the language model to recognize the emergency call information, generate the corresponding recognition result and input the language recognition model again to extract the corresponding recognition emergency information, and compare the extracted recognition emergency information with the corresponding emergency information to determine the corresponding recognition accuracy, and output the recognition accuracy to generate the corresponding resolution; Step S4, dividing the data to be trained according to the resolution pool, and inputting the divided training pool into the corresponding fuzzy screening pool, and performing a preset number of convolutions using the language recognition model to form a fuzzy medical model; Step S5, connecting the fuzzy medical model to the corresponding emergency center according to the collection cycle, inputting the new emergency call information and the corresponding emergency intelligence as new training data into the fuzzy medical model, and determining the corresponding recognition degree; Step S6, when the number of the resolution pools in any acquisition cycle is less than a preset number, it is determined that the training is completed; The preset number is the maximum number of convolutions, which is positively correlated with the sample size; The preset number is proportional to the sample size and is set to a positive integer; When screening the emergency call information, converting the emergency call information into text information using the language recognition model; The first aid information is the final information of first aid completion, and at least includes the address, number of patients and patient information of a single first aid; The global accuracy is to compare the emergency information of the single emergency with the corresponding address, number of patients and / or patient information in each emergency call information, and determine the ratio of the accurate amount of information of any one of the address, number of patients or patient information to the total amount of information; In the step S3, when screening the recognition of a single emergency call, it includes: Step S301, amplifying the volume of the emergency call information and inputting it into the language recognition model to generate corresponding recognition text information; Step S302, inputting the recognized text information into the language recognition model to form secondary recognition information; Step S303, inputting the secondary recognition information into the language recognition model to form semantic recognition information; Step S304, comparing the semantic recognition information with the corresponding emergency intelligence, determining the corresponding global accuracy, and outputting it as the recognition accuracy rate.

2. The medical model training method based on big data according to claim 1, characterized in that: In the step S2, the number of the accuracy standards is not more than the number of the emergency intelligence.

3. The medical model training method based on big data according to claim 2 is characterized in that: When forming the fuzzy screening pool, generating a corresponding number of fuzzy screening pools according to the number of global accuracies; A single fuzzy screening pool corresponds to a range of global accuracies.

4. The medical model training method based on big data according to claim 3 is characterized in that: For a single basic data, it is allocated in the corresponding fuzzy screening pool and the corresponding resolution pool.

5. The medical model training method based on big data according to claim 4 is characterized in that: In the step S4, the fuzzy medical model outputs the semantic recognition information.

6. The medical model training method based on big data according to claim 5, characterized in that: In the step S5, the new emergency call information and the corresponding emergency intelligence correspond to the collection period.

7. The medical model training method based on big data according to claim 6, characterized in that: In step S6, when the resolution pool is set, the corresponding resolution range is a fixed interval. When there is no corresponding basic data in a single fixed interval, it is determined that the resolution pool is canceled.

Images