CN113314212A - Training data processing method and electronic device - Google Patents

Abstract

The invention provides a training data processing method and an electronic device. The method comprises the following steps: obtaining a medical history data including at least one first disease suffered by a user; setting a plurality of disease types according to a target disease; setting a time interval; obtaining at least one second disease within the time interval from the medical history data; performing a preprocessing operation on the second disease according to the disease type to obtain processed data; and inputting the processed data into a neural network to train the neural network.

Description

Training data processing method and electronic device

Technical Field

The invention relates to a training data processing method and an electronic device.

Background

Other diseases such as dementia, which occur years or earlier than the diagnosis may be a precursor of dementia. Therefore, how to predict whether or not there is a risk of developing dementia using historical data of diseases is one of the problems to be solved by those skilled in the art.

Disclosure of Invention

The invention provides a training data processing method and an electronic device, which can enable the prediction effect of an established neural network model to be better than that of a traditional machine learning method.

The invention provides a training data processing method, which is used for an electronic device and comprises the following steps: obtaining a medical history data including at least one first disease suffered by a user; setting a plurality of disease types according to a target disease; setting a time interval; obtaining at least one second disease within the time interval from the medical history data; performing a preprocessing operation on the second disease according to the disease type to obtain processed data; and inputting the processed data into a neural network to train the neural network.

The invention provides an electronic device, comprising: an input circuit and a processor. The input circuit obtains a medical history data including at least a first disease suffered by a user. The processor sets a plurality of disease categories according to a target disease. The processor sets a time interval. The processor obtains at least a second disease located within the time interval from the medical history data. The processor performs a preprocessing operation on the second disease according to the disease type to obtain processed data. The processor inputs the processed data to a neural network to train the neural network.

Based on the above, the training data processing method and the electronic device of the present invention are used to pre-process the data used for training the model, so that the prediction effect of the neural network model established by using the processed data is better than that of the conventional machine learning method, and the application context of the established model conforms to the real application context.

Drawings

The accompanying drawings are included to provide a further understanding of the invention, and are incorporated in and constitute a part of this specification. The drawings illustrate embodiments of the invention and together with the description serve to explain the principles of the invention.

FIG. 1 is a schematic diagram of a training data processing method according to an embodiment of the invention;

FIGS. 2A and 2B are schematic diagrams illustrating time intervals according to an embodiment of the invention;

fig. 3 is a schematic diagram illustrating generation of word frequency information according to an embodiment of the present invention.

Detailed Description

Reference will now be made in detail to exemplary embodiments of the invention, examples of which are illustrated in the accompanying drawings. Wherever possible, the same reference numbers will be used throughout the drawings and the description to refer to the same or like parts.

The model training method of the present invention is applicable to an electronic device (not shown). The electronic device comprises an input circuit (not shown) and a processor (not shown). The input circuit is coupled to the processor. The input circuit is, for example, an input interface or a circuit for obtaining relevant data from outside of the electronic device or other sources, and is not limited herein.

The Processor may be a Central Processing Unit (CPU), or other programmable general purpose or special purpose Microprocessor (Microprocessor), Digital Signal Processor (DSP), programmable controller, Application Specific Integrated Circuit (ASIC), or other similar components or combinations thereof.

In addition, the electronic device may further include a memory circuit (not shown). The memory circuit may be any type of fixed or removable Random Access Memory (RAM), read-only memory (ROM), flash memory (flash memory), or the like, or any combination thereof.

In the exemplary embodiment, the memory circuit of the electronic device stores a plurality of code segments, and the code segments are installed and executed by the processor. For example, the memory circuit includes a plurality of modules, and each of the modules is used to perform each operation of the electronic device, wherein each of the modules is composed of one or more code segments. However, the invention is not limited thereto, and the operations of the electronic device may be implemented by using other hardware forms.

Fig. 1 is a schematic diagram of a training data processing method according to an embodiment of the present invention. In particular, the model (or neural network) trained by the present invention can be used to predict whether a subject will suffer from a target disease or the probability of suffering from the target disease.

In detail, referring to fig. 1, first, the input circuit obtains medical history data including a disease (also referred to as a first disease) suffered by a user (step S101). Thereafter, the processor sets a plurality of disease categories according to the target disease (step S103). The objective disease will be described below as dementia, but the present invention is not limited to the objective disease.

In more detail, the processor converts a plurality of predetermined diseases into a plurality of category data (or a plurality of categories). The processor selects the disease types corresponding to the diseases with higher correlation degree and moderate quantity in the medical field according to the target diseases to be predicted. For example, the processor may perform a screening (e.g., deleting or adding some diseases) from the aforementioned multiple categories of data according to the target disease to be predicted to obtain the last used disease category in step S103. It should be noted that when the number of disease categories is too small, the disease information may not be sufficient to predict the target disease; when the number of disease types is too large, the noise increases, and the prediction accuracy decreases.

For example, assuming the target disease is dementia, the processor may select a CCS single level diagnostics (CCS single level diagnostics) disease classification, of which there are 285 diseases in total. The processor may set the disease types corresponding to the 258 diseases in step S103.

After step S101, the processor sets a time interval (step S105), and obtains a second disease in the time interval from the medical history data (step S107). The processor performs a preprocessing operation on the second disease according to the disease type to obtain processed data (step S109). Finally, the processor inputs the processed data into the neural network to train the neural network (step S111).

For example, fig. 2A and 2B are schematic diagrams illustrating time intervals according to an embodiment of the invention. It should be noted that the user to which the medical history data obtained in step S101 belongs may be suffering from the target disease or not. While for these two different users, the second disease in the time interval may be obtained in different ways.

For example, referring to fig. 2A, fig. 2A is a flowchart illustrating an example of how to define a time interval for obtaining a second disease from medical history data of a user with a target disease. As shown in fig. 2A, time t0 is, for example, the time when the user suffered from (or was first diagnosed with) the target disease, time t1 (also referred to as the first time) is Z years from time t0 (i.e., time t1 is Z years before time t 0), time t2 (also referred to as the second time) is X years from time t1 (i.e., time t2 is X years before time t1), Z and X are positive numbers, and the time unit may be years or months according to the actual situation requirement. The time interval for acquiring the second disease in FIG. 2A is between time t1 and time t 2.

In addition, referring to fig. 2B, fig. 2B is a flowchart illustrating an example of how to define a time interval for obtaining a second disease from medical history data of a user who has not suffered from a target disease using the time interval. As shown in fig. 2B, time k is, for example, the time at which the medical history data of the user is obtained, time t3 (also referred to as a third time) is Z years away from time k (i.e., time t3 is Z years away from time k), time t4 (also referred to as a fourth time) is X years away from time t3 (i.e., time t4 is X years away from time t 3), and Z and X are positive numbers. The time interval for acquiring the second disease in FIG. 2B is between time t3 and time t 4. However, it should be noted that in other embodiments, the time point t3 may be any other time point earlier than the time point k.

It should be noted that the meaning of the definition of the time interval is that if the observation starting point (e.g., time point t1) is too early, the difference of the physical condition of the user with the target disease may not occur yet, and the medical history cannot be used for modeling prediction; if the observation start point (e.g., time t1) is too late, even if it is successfully predicted that the target disease is close to the disease, the effect of preventing the target disease in advance cannot be achieved. In this embodiment, since the target disease is dementia, the processor may set the aforementioned Z value to 5 and the aforementioned X value to 1. That is, taking the example of fig. 2A as an example, the time interval is between the first five years and the first six years of the time point t0 at which dementia occurs.

It is described here how a second disease located within a time interval is obtained from the medical history data. Two approaches can be used: (1) a disease sequence; and (2) word frequency information, and the like, which will be described below.

[ disease sequences ]

The disease sequence can be generated in two ways. In one embodiment, the processor finds the disease within the time interval from the diseases based on the earliest occurrence time of each disease (i.e., the first disease) in the medical history data. And finding a disease sequence consisting of at least one disease (also referred to as a second disease) from the diseases in the time interval. In particular, the second diseases in the disease sequence are ordered according to the earliest time of occurrence, the number of the second diseases is less than or equal to a predetermined number, and each of the second diseases occurs only once.

For example, assume that the predetermined number is 5, and assume that the sequence of visits (or diseases) in a time interval in a certain medical history "disease 2 → disease 1 → disease 2 → disease 4 → disease 3". The disease sequence "disease 2 → disease 1 → disease 4 → disease 3" can be obtained by using the earliest occurrence time ranking. In this sequence, the number of diseases (i.e., 4) is smaller than the predetermined number (i.e., 5). Also in this disease sequence, each disease occurs only once.

In the second mode, the processor finds out the diseases in the time interval from the diseases according to all the occurrence time of each disease in the medical history data, and sorts the diseases according to the occurrence time. The disease in the disease sequence generated in this manner may be repeated.

In addition, in one embodiment, the processor deletes a portion of the disease (also referred to as the third disease) in the medical history data to obtain a disease sequence consisting of a plurality of diseases (e.g., the second disease). Wherein the time of occurrence of the third disease is earlier than the time of occurrence of the disease in the disease sequence. The diseases in the disease sequence are ordered according to the earliest time of occurrence, and the number of diseases in the disease sequence is less than or equal to a preset number.

For example, assuming that the predetermined number is 5, and assuming that the second manner is described above, the sequence of the visits (or diseases) in the time interval in a certain medical history is "disease 2 → disease 1 → disease 2 → disease 4 → disease 3", since the number of the visits (or diseases) (i.e., 7) in the medical history data is greater than the predetermined number, the processor may delete the disease "disease 2 → disease 2" appearing earlier in the medical history data to obtain a disease sequence "disease 1 → disease 2 → disease 4 → disease 3".

After the second disease in the time interval is obtained in the foregoing manner, the second disease in the disease sequence may be encoded into one-dimensional or two-dimensional encoded data (or referred to as a vector) according to the disease type in step S109, and the encoded data may be used as processed data, and this processed data may be input to the neural network to train the neural network in step S111.

The following description will take the encoded data obtained by encoding the second disease into one dimension as an example. Assume that there are 5 disease categories in total, and that "disease 1", "disease 2", "disease 3", "disease 4" and "disease 5" are defined as "[ 1,0,0,0,0 ]", "[ 0,1,0,0,0 ]", "[ 0,0,1,0,0 ]", "[ 0,0,0,1,0 ]", and "[ 0,0,0,0,1 ]", respectively. Assuming that the disease sequence obtained via the foregoing manner is "disease 2 → disease 1 → disease 4 → disease 3", the processor may convert the disease sequence into: "[ 0,1,0,0,0] → [1,0,0,0,0] → [0,0,0,1,0] → [0,0,1,0,0 ]", and will in turn generate one-dimensional data "[ 0,1,0,0,0,1,0,0,0,0,0,0,0,1,0,0,0,1,0,0 ]". This one-dimensional data may then be input to a neural network that takes the one-dimensional data as input.

The following description will take an example in which the second disease is encoded into two-dimensional encoded data. Assume that there are 5 disease categories in total, and that "disease 1", "disease 2", "disease 3", "disease 4" and "disease 5" are defined as "[ 1,0,0,0,0 ]", "[ 0,1,0,0,0 ]", "[ 0,0,1,0,0 ]", "[ 0,0,0,1,0 ]", and "[ 0,0,0,0,1 ]", respectively. Assuming that the disease sequence obtained via the foregoing manner is "disease 2 → disease 1 → disease 4 → disease 3", the processor may convert the disease sequence into: "[ 0,1,0,0,0] → [1,0,0,0,0] → [0,0,0,1,0] → [0,0,1,0,0 ]", and will in turn produce two-dimensional data as a matrix:

this two-dimensional data may then be input to a neural network (e.g., LSTM) that takes the two-dimensional data as input.

In particular, since the one-dimensional or two-dimensional data codes the diseases in a chronological manner, the precedence relationship among the diseases is still maintained in the coded data.

In the process of training the neural network, the converted vector length M may be set, for example, using the method of sequence embedding, and may be trained with the neural network (e.g., LSTM).

[ word frequency information ]

Fig. 3 is a schematic diagram illustrating generation of word frequency information according to an embodiment of the present invention.

Referring to FIG. 3, in an embodiment, the processor can directly obtain the diseases D1-D2 located in the time interval from the medical history data DD as the second disease constituting the disease sequence. The processor weights the diseases D1-D2 (without limiting the weight), and then the diseases are regarded as words (words) and converted into word frequency information E1-E2 by using the TF-IDF algorithm.

It should be noted that the present invention is not limited to how the disease is weighted. In one embodiment, the weighting may be based on whether a diagnosis was seen. For example, the weight of a diagnosed disease may be set to 1, otherwise 0.

In another embodiment, the weighting may be based on the number of visits. Suppose a person's medical history is: "disease 2 → disease 1 → disease 2 → disease 4 → disease 3", then the weight value for disease 1 is 1, the weight value for disease 2 is 3, the weight value for disease 3 is 2, and the weight value for disease 4 is 1.

In another embodiment, the weighting may be based on other medical history information. Other medical history data such as: individual disease dosages, surgical information, chronic disease markers, other treatments, and the like, without limitation.

In another embodiment, disease dosages may also be used for weighting. Assuming that three people A, B and C have seen diabetes and the dosages are 2 units, 1 unit and 3 units respectively, the weight of diabetes of the three people is 2, 1 and 3 respectively.

In another embodiment, the importance of the disease may be ranked by other machine learning methods and then weighted.

After the weighted second diseases are respectively converted into the word frequency information, the processor takes the word frequency information as processed data and inputs the processed data into the neural network to train the neural network. In particular, the format of the word frequency information generally conforms to the general machine learning input data format, so that the word frequency information can be directly input into the neural network for training.

After the neural network training is completed in the above manner, when the neural network receives the medical history data of a subject, it can be determined whether a target disease (e.g., dementia) is likely to be caused or not or the probability of the target disease being caused by the neural network.

In summary, the training data processing method and the electronic device of the present invention are used to pre-process the data used for training the model, so that the prediction effect of the neural network model established by using the processed data is better than that of the conventional machine learning method, and the application context of the established model is in accordance with the real application context.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.

Claims (14)

1. A training data processing method for an electronic device, the method comprising:

obtaining medical history data including at least one first disease suffered by a user;

setting a plurality of disease categories according to the target disease;

setting a time interval;

obtaining at least one second disease within the time interval from the medical history data;

performing a pre-processing operation on the second disease according to the disease type to obtain processed data; and

inputting the processed data to a neural network to train the neural network.

2. The method of claim 1, wherein the user suffers from the target disease, the time interval is between a first time point and a second time point, the first time point is Z years before the time point suffering from the target disease, the second time point is X years before the first time point, and Z and X are positive numbers.

3. The method of claim 1, wherein the user is not suffering from the target disease, wherein the time interval is between a third time point and a fourth time point, the third time point is Z years before or any time point of the time point of obtaining the medical history data, and the fourth time point is X years before the third time point, and Z and X are positive numbers.

4. The training data processing method according to claim 1, wherein the step of obtaining the second disease located within the time interval from the medical history data includes:

obtaining a disease sequence consisting of the second diseases from the first diseases according to the earliest time of occurrence of each of the first diseases, wherein the second diseases in the disease sequence are ordered according to the earliest time of occurrence, the number of the second diseases is less than or equal to a preset number, and each of the second diseases appears only once.

5. The training data processing method according to claim 1, wherein the step of obtaining the second disease located within the time interval from the medical history data includes:

deleting at least one third disease in the medical history data to obtain a disease sequence consisting of the second disease, wherein the occurrence time of the third disease is earlier than the occurrence time of the second disease, the second diseases in the disease sequence are sorted according to the earliest occurrence time, and the number of the second diseases is less than or equal to a preset number.

6. The training data processing method according to claim 5, wherein the step of performing the preprocessing operation on the second disease according to the disease category to obtain the processed data includes:

and coding the second disease in the disease sequence into one-dimensional or two-dimensional coded data according to the disease type, and taking the coded data as the processed data.

7. The training data processing method according to claim 1, wherein the step of performing the preprocessing operation on the second disease according to the disease category to obtain the processed data includes:

weighting each of said second diseases;

and respectively converting the weighted second disease into at least one word frequency information, and taking the word frequency information as the processed data.

8. An electronic device, comprising:

an input circuit; and

a processor coupled to the input circuit, wherein

The input circuit obtains medical history data including at least one first disease suffered by the user,

the processor sets a plurality of disease categories according to a target disease,

the processor sets a time interval for which the processor,

the processor obtains at least a second disease within the time interval from the medical history data,

the processor performs a pre-processing operation on the second disease according to the disease category to obtain processed data,

the processor inputs the processed data to a neural network to train the neural network.

9. The electronic device of claim 8, wherein the time interval between a first time point and a second time point is between the first time point and the second time point, the first time point is Z years before the time point of the target disease, and the second time point is X years before the first time point, and Z and X are positive numbers.

10. The electronic device of claim 8, wherein the user is not suffering from the target disease, wherein the time interval is between a third time point and a fourth time point, the third time point is Z years ago or any time point of the time point of obtaining the medical history data, and the fourth time point is X years ago of the third time point, Z and X being positive numbers.

11. The electronic device of claim 8, wherein in operation to obtain the second disease within the time interval from the medical history data,

the processor obtains a disease sequence composed of the second diseases from the first diseases according to the earliest time of occurrence of each of the first diseases, wherein the second diseases in the disease sequence are sorted according to the earliest time of occurrence, the number of the second diseases is less than or equal to a preset number, and each of the second diseases only appears once.

12. The electronic device of claim 8, wherein in operation to obtain the second disease within the time interval from the medical history data,

the processor deletes at least one third disease in the medical history data to obtain a disease sequence composed of the second disease, wherein the occurrence time of the third disease is earlier than the occurrence time of the second disease, the second diseases in the disease sequence are sorted according to the earliest occurrence time, and the number of the second diseases is less than or equal to a preset number.

13. The electronic device of claim 12, wherein in the operation of performing the pre-processing operation on the second disease according to the disease type to obtain the processed data,

and the processor encodes the second disease in the disease sequence into one-dimensional or two-dimensional encoded data according to the disease type, and takes the encoded data as the processed data.

14. The electronic device of claim 8, wherein in the operation of performing the pre-processing operation on the second disease according to the disease type to obtain the processed data,

the processor weights each of the second diseases,

and the processor converts the weighted second disease into at least one word frequency information respectively and takes the word frequency information as the processed data.

Images