CN104462858A - Health warning method based on multi-order hidden Markov model - Google Patents

Abstract

The invention provides a health warning method based on a multi-order hidden Markov model. The method comprises the steps of establishing a standard health knowledge base, establishing a personal physical sign data information base, and establishing the multi-order hidden Markov model for personal prediction and warning. According to the method, an interactive active health service mode is provided, the health condition of a person changes along with time, and warning can be conducted when the condition becomes abnormal through comparison between the condition and corresponding indexes in the health knowledge base. The state conversion spaces can be divided into multiple reachable equivalence classes. On one hand, the physical sign data development trend and health trend be found by means of big data, and on the other hand, when the health condition of a new member needs to be judged, a corresponding matched state equivalence class can be found out, and then judgment is conducted by means of a state which is easy to judge (disease symptoms and development trend are similar to speed) so that warning and suggestions are provided.

Description

Based on the healthy early warning method of multistage hidden Markov model

Technical field

The invention belongs to computerized information field, particularly relate to a kind of healthy early warning method based on multistage hidden Markov model.

Background technology

Present stage people recognize that " centered by health " and " preventiveing treatment of disease " theory is the development trend of modern medical service service gradually, health service pattern starts, from the unidirectional passive health service pattern of tradition to the active health service Mode change of interaction, to be embodied in: one is " prevention of diseae " (i.e. prevention and health care) thus; Two is " both disease was preapred for an unfavorable turn of events " or " sick morning is controlled "; Three is successional medical services (consulting services, disease control, healthy early warning are seriation), realizes early intervention, viewpoint reach, the active service mode centered by disease/people.

Human body physical sign change has certain rule, and before chronic disease generation, human body has had some continuation extremely.In theory, if large data have grasped such abnormal conditions, just chronic disease prediction can have been carried out.Utilize the large data of platform to do the correlation analysis of disease forecasting, even can provide good suggestion to new drug.

Summary of the invention

The problem to be solved in the present invention is a kind of healthy early warning method based on multistage hidden Markov model of design, realizes early intervention, viewpoint reach, the active service mode centered by disease/people.

It should be noted that, the present invention is based on the healthy early warning method of multistage hidden Markov model, it is the one application of information science, the method of the healthy early warning being applicable to self is obtained by information analysis, not belong to the Diagnosis and Treat method of disease, therefore do not violate the relevant regulations of Patent Law Article 25.

In order to achieve the above object, the technical scheme that the present invention takes is: a kind of healthy early warning method based on multistage hidden Markov model, is characterized in that, comprise the steps:

(1) Criterion health knowledge storehouse: the normal value of record sign data, the normal value of health check-up and routine inspection data, for health data comparison true in individual in the future;

(2) set up individual sign data information bank, store the health and fitness information of individual, comprise sign data, health check-up and routine inspection data.For with the storehouse comparison of standard health knowledge, tracking data change, to analyze health status tendency, provide healthy early warning and personalized improving countermeasure;

(3) multistage hidden Markov model is set up,

The health status S of people at a time t _trepresent, then people is in a series of state representation of the health status of different time, and people regards the transfer of a series of state as in the health status of different times;

To the health status index S of people _trepresent with m achievement data, then people represents with a series of m dimensional vector in the health status of different time, S _t=(S _{t, 1}, S _{t, 2}... S _t,m);

A people is from t state S _tto t+1 moment state S _t+1the probability of transfer is P _ij, then from t state to the state transition probability matrix in t+1 moment be A=(P _ij) _{n × N}, the formula (1) namely:

S _t+1＝S _tA (1)

Wherein

0≤P _ij≤1 $\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{P}_{\mathrm{ij}}=1;\∀j=\mathrm{1,2,3}...N;$

(4) personal prediction and early warning, first reads corresponding sign data, then obtains its original state, then give a forecast with multistage hidden Markov model from individual sign data information bank.

Preferably, in described step (1), sign data comprises body weight, heart rate, blood pressure, pulse frequency, respiratory rate, body temperature, thermal losses, amount of exercise, amount of sleep blood sugar and blood oxygen, hormone and BMI index, body fat content; Described health check-up and routine inspection data comprise routine urinalysis every, biochemical full item, lectin from hemolymph 9, tumor markers 3.

Preferably, described step (3) transition probability matrix A initial value adopts expert given, then according to the case information Data Update that sufferer information data and doctor increase.

Further, described Data Update is pressed following formula (2) and is upgraded transition probability Pij,

New Pij＝old Pij+dij/N， (2)

Wherein dij is the corresponding weight value increment because the data of sufferer change, and-1 < dij < 1, N is total status number.

Further, described Data Update adopts the machine learning method iteration of probabilistic neural network to upgrade with perfect.

Beneficial effect of the present invention is: by method of the present invention, provides interactive active health service pattern, the health status of people along with passage of time, when with the corresponding index comparison of health knowledge storehouse, can emphasis prompting early warning when occurring abnormal.These State Transferring spaces may be divided into and much can reach equivalence class, on the one hand, utilize large data can find sign data development trend and healthy trend; On the other hand, when the member's health status needing judgement one new, corresponding matching status equivalence class can be looked for, judge (illness and development trend state and speed similar) to provide early warning and suggestion with this by the state wherein easily judged.

Accompanying drawing explanation

Fig. 1 is the schematic diagram of the multistage hidden Markov model of people's health status.

Embodiment

Below in conjunction with specific embodiment, the present invention will be further described.

The following work of this programme design:

1. member's health knowledge storehouse: record sign data is (as body weight, heart rate, blood pressure, pulse frequency, respiratory rate, body temperature, thermal losses, amount of exercise, amount of sleep blood sugar and blood oxygen, hormone and BMI index, body fat content) normal value, health check-up and routine inspection (every, the biochemical full item of routine urinalysis, lectin from hemolymph (9), tumor markers 3 etc.) normal value, for health data comparison true in member in the future.

2. member's sign data information bank: store the health and fitness information logging in the member of website, comprise sign data, health check-up and routine inspection data.For with the comparison of standard knowledge storehouse, tracking data change, to analyze health status tendency, provide healthy early warning and personalized improving countermeasure.

3. based on M-HMM (Multistage HiddenMarkovModel, multistage hidden Markov model) healthy trend prediction and early warning:

Markov model (MarkovModel) utilizes known all information till now to predict algorithm model in the future.The health status of people at a time t is represented with St, then people is in a series of state representation of the health status of different time, all possible like this health status composition member's health status space.Because this person regards the transfer of a series of state as in the health status of different times.As pneumonia I, II, III phase and phthisical I, II, III phase was its each state, wherein dead or can not be cured as absorbing state (stop state, no longer change).

And a series of achievement data is often needed to the health status index S t of people, as with m achievement data St _,1, St _,2 ... St _,m represents, namely form m dimensional vector, then people represents with a series of m dimensional vector in the health status of different time, and the health status as t is St=(St _,1, St _,2 ... St _,m).So a state needs several correlated phenomena to observe decision, so adopt hidden Markov model.Again because people's state is not at a time an index, so belong to multistage hidden Markov model.As shown in Figure 1:

As pneumonia I, II, III phase and phthisical I, II, III phase, its main detection shows as heating, night sweat, general malaise and cough, expectoration, spitting of blood, pectoralgia, expiratory dyspnea etc.Need the project detected to be routine blood test, routine urinalysis, stool routine, x-ray inspection, humoral immunity detects, liver function test, kidney function test, microbe growth, CT examination, endoscopy.If the probability that people shifts from t state St to t+1 moment state St+1 is Pij, then from t state to the state transition probability matrix in t+1 moment be A=(Pij) _{n × N}, i.e. formula (1)

St+1＝St A (1)

As above-mentioned pneumonia I, II, III phase and phthisical I, II, III phase added death or can not be cured and health totally 8 kinds of states.Namely A is 8*8 matrix, supposes as follows:

0.83 0.04 0.05 0.01 0.01 0.03 0.02 0.01

0.01 0.93 0.01 0.01 0.01 0.01 0.01 0.01

0.01 0.01 0.82 0.00 0.04 0.03 0.06 0.03

0.03 0.00 0.01 0.78 0.05 0.05 0.01 0.01

0.01 0.01 0.01 0.01 0.93 0.01 0.01 0.01

0.02 0.03 0.01 0.04 0.00 0.88 0.01 0.01

0.01 0.02 0.03 0.04 0.05 0.02 0.80 0.03

0.02 0.04 0.01 0.03 0.01 0.06 0.00 0.83

Its transition probability matrix A initial value adopts expert given, then according to the case information Data Update that this member and other members increase with ward mate's information data and doctor.

Data Update wherein can adopt and following upgrade calculating and be:

(1) if in a simple manner decoupled, then press following formula (2) assignment statement according to the data of certain member and upgrade transition probability Pij,

New Pij＝old Pij+dij/N， (2)

Wherein dij is the corresponding weight value increment because the data of certain member change, and-1 < dij < 1, N is total status number.

(2) if having large Data support and hardware condition can support complicated intelligent method, then the machine learning method iteration of probabilistic neural network can be adopted to upgrade with perfect.

For prediction and the early warning of a member, first from above-mentioned information database 2, read corresponding sign data, obtain its original state, then give a forecast with M-HMM, according to member's requirement, the prediction of various time domain can be done.

The health status of people along with passage of time, when with the corresponding index comparison of health knowledge storehouse, can emphasis prompting early warning when occurring abnormal.These State Transferring spaces may be divided into and much can reach equivalence class, on the one hand, utilize large data can find sign data development trend and healthy trend; On the other hand, when the member's health status needing judgement one new, corresponding matching status equivalence class can be looked for, judge (illness and development trend state and speed similar) to provide early warning and suggestion with this by the state wherein easily judged.

The foregoing is only specific embodiments of the invention, the protection domain be not intended to limit the present invention, within the spirit and principles in the present invention all, any amendment made, equivalent replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (5)

1., based on a healthy early warning method for multistage hidden Markov model, it is characterized in that, comprise the steps:

(1) Criterion health knowledge storehouse: the normal value of record sign data, the normal value of health check-up and routine inspection data, for health data comparison true in individual in the future;

(2) set up individual sign data information bank, store the health and fitness information of individual, comprise sign data, health check-up and routine inspection data.For with the storehouse comparison of standard health knowledge, tracking data change, to analyze health status tendency, provide healthy early warning and personalized improving countermeasure;

(3) multistage hidden Markov model is set up,

The health status S of people at a time t _trepresent, then people is in a series of state representation of the health status of different time, and people regards the transfer of a series of state as in the health status of different times;

To the health status index S of people _trepresent with m achievement data, then people represents with a series of m dimensional vector in the health status of different time, S _t=(S _{t, 1}, S _{t, 2}... S _t,m);

A people is from t state S _tto t+1 moment state S _t+1the probability of transfer is P _ij, then from t state to the state transition probability matrix in t+1 moment be A=(P _ij) _{n × N}, the formula (1) namely:

S _t+1＝S _tA (1)

Wherein 0≤P _ij≤ 1 $\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{P}_{\mathrm{ij}}=1;\&ForAll;j=\mathrm{1,2,3}\&CenterDot;\&CenterDot;\&CenterDot;N$

(4) personal prediction and early warning, first reads corresponding sign data, then obtains its original state, then give a forecast with multistage hidden Markov model from individual sign data information bank.

2. a kind of healthy early warning method based on multistage hidden Markov model according to claim 1, it is characterized in that, in described step (1), sign data comprises body weight, heart rate, blood pressure, pulse frequency, respiratory rate, body temperature, thermal losses, amount of exercise, amount of sleep blood sugar and blood oxygen, hormone and BMI index, body fat content; Described health check-up and routine inspection data comprise routine urinalysis every, biochemical full item, lectin from hemolymph 9, tumor markers 3.

3. a kind of healthy early warning method based on multistage hidden Markov model according to claim 1, it is characterized in that, described step (3) transition probability matrix A initial value adopts expert given, then according to the case information Data Update that sufferer information data and doctor increase.

4. a kind of healthy early warning method based on multistage hidden Markov model according to claim 3, is characterized in that, described Data Update is pressed following formula (2) and upgraded transition probability Pij,

New Pij＝old Pij+dij/N， (2)

Wherein dij is the corresponding weight value increment because the data of sufferer change, and-1 < dij < 1, N is total status number.

5. a kind of healthy early warning method based on multistage hidden Markov model according to claim 3, is characterized in that, described Data Update adopts the machine learning method iteration of probabilistic neural network to upgrade with perfect.