WO2021085947A1

WO2021085947A1 - Parkinson's disease diagnostic application

Info

Publication number: WO2021085947A1
Application number: PCT/KR2020/014568
Authority: WO
Inventors: Seung Jun Jeon
Original assignee: Infoshare Co., Ltd
Priority date: 2019-10-28
Filing date: 2020-10-23
Publication date: 2021-05-06
Also published as: KR20210050107A; KR102399118B1

Abstract

A diagnostic application for Parkinson's disease is disclosed. The diagnostic application for Parkinson's disease diagnoses Parkinson's disease by extracting acoustic features of voice input to a smart terminal through frequency conversion of the voice, the acoustic features including an average, a deviation, a frequency variation (F0 variation), a harmonics to noise ratio (HNR), intensity of the voice, jitter of the voice, and shimmer of the voice, subdividing a voice recognition measurement scheme according to speaker independence, a pronunciation pattern, and the number of vocabularies based on Parkinson's disease diagnosis data through an application installed in the smart terminal, using the extracted acoustic features, and canceling noise introduced during voice recording through a cepstrum.

Description

PARKINSON'S DISEASE DIAGNOSTIC APPLICATION

The present disclosure relates to a diagnostic application for Parkinson's disease, and more particularly, to a diagnostic application for Parkinson's disease, which performs an acoustic analysis on a sentence uttered by a patient with Parkinson's disease, extracts acoustic features from the sentence, and diagnoses whether the patient suffers from Parkinson's disease based on the extracted acoustic features.

Parkinson's disease is a degenerative brain disease. The main symptoms include resting tremor, bradykinesia, cogwheel rigidity, and postural reflexes. If two or more of these symptoms and pathological findings are observed in a patient, the patient is diagnosed as idiopathic Parkinson's disease. These physical clinical findings appear only when 80% or more of cells implicated in the production of the neurotransmitter, dopamine are lost, which makes initial diagnosis of Parkinson's disease very difficult. Moreover, it is not easy to distinguish the physical characteristics of Parkinson's disease from normal aging.

Because there has been no practical and objective test tool for diagnosing Parkinson's disease so far, medical diagnosis for Parkinson's disease is made based on general clinical findings. Accordingly, neurologists make different diagnoses for a patient with Parkinson's disease, and even specialists in other departments may not distinguish Parkinson's disease from normal aging symptoms. Patients and their families frequently visit different medical institutions (oriental medicine, orthopedics, rehabilitation medicine, and so on) for an accurate diagnosis, thereby suffering financially and mentally.

It was reported that 84% of patients with Parkinson's disease had experienced misdiagnosis in two medical institutions in Korea. After the onset of Parkinson's disease, no fewer than 5 years is taken before 17% of patients with Parkinson's disease are finally diagnosed. The resulting personal and social losses are great.

More than 70% of patients with Parkinson's disease are known to have speech disorders. This implies that the presence or absence of Parkinson's disease for a patient may be identified indirectly from the way the patient speaks (speech pattern) or sentences that the patient utters (uttered sentence). Specifically, this prediction is possible through the phonetic prosody analysis of the patient's sentence.

Speech is composed of segmental sounds and prosody. In the Korean language, a segmental sound is a unit of speech usually expressed by a consonant of Hangul, and prosody refers to a very important musical element that allows these segmental sounds to be woven together to produce a physical sound in the form of a sound wave. The prosody of an utterance is mainly composed of pitch, intensity, and length.

Although patients with Parkinson's disease are not distinctively different from normal persons in terms of segmental sound, most of them show characteristic and consistent patterns in terms of prosody. Therefore, analysis of the prosody features of a patient's spoken text may be used to diagnose whether the patient has Parkinson's disease.

Existing voice analysis methods, for example, products using the voice analysis program of Kay or Tiger, mainly perform voice quality analysis, and have limitations in analyzing the prosody aspect of an entire sentence.

It is difficult to accurately diagnose Parkinson's disease only by analyzing the sound quality of voice. This is because a more accurate diagnosis is possible only when the prosody elements of a patient's speech sentence are integrally observed and analyzed.

More importantly, the existing analysis tools are mainly based on research conducted by foreign entrepreneurs on foreigners and thus not suitable for direct application to Korean patients in many cases. Therefore, there is a pressing need to develop a diagnostic tool based on the results of research conducted on Parkinson's disease patients in Korea.

As a relevant prior art document for solving the above problems, Korean Patent Registration No. 10-1182069 entitled by "Apparatus and Method for Diagnosing Idiopathic Parkinson's Disease through Prosody Analysis of Uttered Sentence" discloses a Parkinson's disease diagnostic apparatus including a recorder that records a sentence uttered by a patient, a storage that stores a pre-built diagnostic model, an analyzer that analyzes the uttered sentence recorded in the recorder, a diagnoser that diagnoses the presence or absence of Parkinson's disease in the patient from analysis results obtained by the analyzer using the diagnosis model stored in the storage, and a controller that controls each of the recorder, the storage, the analyzer, and the diagnoser, wherein the diagnosis model is built by predetermining sentences to be uttered by the patient, recording diagnostic sentences uttered by a patient group of patients who have been confirmed as having Parkinson's disease and diagnostic sentences uttered by a control group of normal persons, analyzing the prosody of the recorded uttered sentences, and performing multivariate statistical analysis and an automatic machine learning algorithm using results of the prosody analysis.

In the prior art document, a series of processes related to diagnosis are non-invasive, thus causing no pain to patients, and the diagnosis process is convenient even to a patient suffering movement disorders due to voice recording through a recorder. Further, since a portable digital recorder may be easily handled by ordinary people, a patient's family may record utterance of the patient directly in a quiet room without assistance from an expert, thereby enhancing the convenience of use.

However, a shortcoming with the above-described prior art is a high error rate in diagnosis of Parkinson's disease due to the introduction of ambient noise during recording.

An aspect of the present disclosure devised to solve the problem is to provide a diagnostic application for Parkinson's disease, which diagnoses Parkinson's disease by subdividing a voice recognition measurement scheme according to speaker independence, a pronunciation pattern, and the number of vocabularies based on accumulated result data of Parkinson's disease diagnosis through an application installed in a smart terminal and canceling noise introduced during voice recording through a cepstrum.

In accordance with an aspect of the present disclosure, a diagnostic application for Parkinson's disease is provided. In the diagnostic application for Parkinson's disease, Parkinson's disease is diagnosed by extracting acoustic features of voice input to a smart terminal through frequency conversion of the voice, the acoustic features including an average, a deviation, a frequency variation (F0 variation), a harmonics to noise ratio (HNR), intensity of the voice, jitter of the voice, and shimmer of the voice, subdividing a voice recognition measurement scheme according to speaker independence, a pronunciation pattern, and the number of vocabularies based on Parkinson's disease diagnosis data through an application installed in the smart terminal, using the extracted acoustic features, and canceling noise introduced during voice recording through a cepstrum.

The diagnostic application for Parkinson's disease may include a model generator, a Kalman filter, a decision theory unit, a combination type generator, and a prediction generator. The model generator may be configured to store the voice and classify the stored voice into voice responsive to the Kalman filter and voice unresponsive to the Kalman filter through the application installed in the smart terminal by a user intending to diagnose Parkinson's disease. The Kalman filter may be configured to receive a measurement

of acoustic data of the voice responsive to the Kalman filter, which has been classified by the model generator and output an estimate

by performing prediction, Kalman gain calculation, and estimate calculation, for diagnosis of Parkinson's disease for the user, the decision theory unit may be configured to detect occurrence of Parkinson's disease by applying a probabilistic decision theory based on a residual generated by the Kalman filter. The combination type generator may be configured to combine acoustic data based on data generated from the Kalman filter by type and store the combined data by type, for fast analysis of acoustic data of the same type. The prediction generator may be configured to predict Parkinson's disease based on acoustic data analyzed by the Kalman filter, when voice from a patient with Parkinson's disease has a tremor due to weakening of muscles controlling intensity of the voice and the tremor increases to or above a threshold, when a bias phenomenon occurs in the bass of the voice, when a spike occurs suddenly in an output value of the voice, when the intensity of the voice suddenly decreases, when jitter of the voice rapidly increases, when noise of the voice rapidly increases and forms a pattern, when pitch of the voice gradually decreases, or when an output of the voice non-linearly changes.

The Kalman filter may identify an abnormal acoustic feature in which voice from a patient with Parkinson's disease has a tremor due to weakening of muscles controlling intensity of the voice, the tremor increases to or above a threshold, and thus a hard over error occurs.

The Kalman filter may identify an abnormal acoustic feature in which voice from a patient with Parkinson's disease is biased in bass of the voice and thus a biased error occurs.

The Kalman filter may identify an abnormal acoustic feature in which a spike occurs suddenly in voice from a patient with Parkinson's disease.

The Kalman filter may identify an abnormal acoustic feature in which intensity of voice from a patient with Parkinson's disease suddenly decreases.

The Kalman filter may identify an abnormal acoustic feature in which jitter of voice from a patient with Parkinson's disease rapidly increases.

The Kalman filter may identify an abnormal acoustic feature in which noise of voice from a patient with Parkinson's disease rapidly increases and the rapidly increased noise has a specific pattern.

The Kalman filter may identify an abnormal acoustic feature in which pitch of voice from a patient with Parkinson's disease gradually decreases, that is, the pitch of the voice continuously decreases.

The Kalman filter may identify an abnormal acoustic feature in which an output of voice from a patient with Parkinson's disease non-linearly changes, that is, the pitch of the voice gradually decreases and then gradually increases.

According to the present disclosure, as Parkinson's disease is diagnosed by subdividing a voice recognition measurement scheme according to speaker independence, a pronunciation pattern, and the number of vocabularies based on accumulated result data of Parkinson's disease diagnosis through an application installed in a smart terminal and canceling noise introduced during voice recording through a cepstrum, the diagnosis of Parkinson's disease is accurate.

Further, as Parkinson's disease is diagnosed by performing acoustic analysis and feature extraction on sentences uttered by patients with Parkinson's disease based on accumulated data of users' Parkinson's disease diagnosis results through an application installed in a smart terminal and extracting eight typical feature types from the extracted features using a Kalman filter, an accurate diagnosis of Parkinson's disease is made.

FIG. 1 is a block diagram illustrating a Parkinson's disease prediction process using a diagnostic application for Parkinson's disease according to the present disclosure.

FIG. 2 is a diagram illustrating an algorithm of a Kalman filter illustrated in FIG. 1.

FIG. 3 is a diagram illustrating a starting state of diagnosis of Parkinson's disease in the diagnostic application for Parkinson's disease according to the present disclosure.

FIG. 4 is a diagram illustrating states of voice input to the diagnostic application for Parkinson's disease according to the present disclosure.

FIG. 5 is a diagram illustrating a state in which a voice diagnosis is successfully made in the diagnostic application for Parkinson's disease according to the present disclosure.

FIG. 6 is a diagram illustrating a state in which a voice diagnosis is failed in the diagnostic application for Parkinson's disease according to the present disclosure.

FIG. 7 is a diagram illustrating results of a voice diagnosis in the diagnostic application for Parkinson's disease according to the present disclosure.

FIG. 8 is a graph illustrating an abnormal acoustic feature in which voice from a patient with Parkinson's disease has rapidly increasing tremors as a result of a voice diagnosis in the diagnostic application for Parkinson's disease according to the present disclosure.

FIG. 9 is a graph illustrating an abnormal acoustic feature in which voice from a patient with Parkinson's disease has rapidly increasing noise in a specific pattern as a result of a voice diagnosis in the diagnostic application for Parkinson's disease according to the present disclosure.

FIG. 10 is a graph illustrating an abnormal acoustic feature in which voice from a patient with Parkinson's disease has a gradually decreasing pitch, that is, a continuously decreasing pitch as a result of a voice diagnosis in the diagnostic application for Parkinson's disease according to the present disclosure.

FIG. 11 is a graph illustrating an abnormal acoustic feature in which voice from a patient with Parkinson's disease has a non-linearly changing output, that is, the pitch of the voice gradually decreases and then gradually increases as a result of a voice diagnosis in the diagnostic application for Parkinson's disease according to the present disclosure.

FIG. 12 illustrates a screen displaying a diagnosis page with a "checkup" button on a user interface (UI) of an application installed for diagnosis of Parkinson's disease in a smart terminal according to the present disclosure.

FIG. 13 illustrates a screen displayed when voice of 0.8db or higher triggers diagnosis, upon pressing of the "checkup start" button on the screen of the diagnosis page according to the present disclosure.

FIG. 14 illustrates a screen displayed when diagnosis is 100% completed about 10 seconds after the "checkup start" button is pressed on the screen of the diagnosis page according to the present disclosure.

FIG. 15 illustrates a screen displaying a message "Please measure again", when measurement is difficult due to too weak voice to be diagnosed or the number of available measurement data smaller than a reference after the "checkup start" button is pressed on the screen of the diagnosis page according to the present disclosure.

FIG. 16 illustrates a screen displaying the results of successful diagnosis made according to the present disclosure in comparison with average values of a patient with Parkinson's disease.

A diagnostic application for Parkinson's disease according to the present disclosure will be described in detail with reference to the attached drawings.

The diagnostic application for Parkinson's disease according to the present disclosure is characterized in that Parkinson's disease is diagnosed by extracting acoustic features of voice input to a smart terminal through frequency conversion of the voice, relying on the property that as the disease progresses, voice tremor increases, voice intensity decreases, and the period of vocalization decreases, due to muscle atrophy, the acoustic features including an average, a deviation, a frequency variation (F0 variation), a harmonics to noise ratio (HNR), intensity of the voice, jitter of the voice, and shimmer of the voice, subdividing a voice recognition measurement scheme according to speaker independence, a pronunciation pattern, and the number of vocabularies based on Parkinson's disease diagnosis data through an application installed in the smart terminal, using the extracted acoustic features, and canceling noise introduced during voice recording through a cepstrum.

With reference to the attached drawings, a more detailed description will be given of the diagnostic application for Parkinson's disease which has the above-described features.

FIG. 1 is a block diagram illustrating a process of predicting Parkinson's disease in a diagnostic application for Parkinson's disease according to the present disclosure, and FIG. 2 is a diagram illustrating an algorithm of a Kalman filter illustrated in FIG. 1.

Referring to FIGS. 1 and 2, the diagnostic application for Parkinson's disease according to the present disclosure largely includes a model generator 10, a Kalman filter 20, a decision theory unit 30, a combination type generator 40, and a prediction generator 50.

The model generator 10 stores voice through an application installed in a smart terminal by a user intending to diagnose Parkinson's disease, and classifies the stored voice into voice responsive to the Kalman filter 20 and voice unresponsive to the Kalman filter 20.

The Kalman filter 20 receives a measurement

of acoustic data of the voice responsive to the Kalman filter 20, which has been classified by the model generator 10 to diagnose Parkinson's disease of the user, and outputs an estimate

by estimate prediction, Kalman gain calculation, and estimate calculation.

For the input of the measurement

, the Kalman filter 20 outputs the estimate

in multiple steps.

A first step is an estimation step. Two variables

and

are calculated for use in second, third and fourth steps.

In the second step, a Kalman gain

is calculated, a value calculated in the first step is used as the variable

, and H and R are predetermined values.

In the third step, an estimate is calculated from the input measurement, and a value calculated in the first step is used as the variable

.

In the fourth step, an error covariance is calculated from the input measurement and the variable

is calculated.

The above-described variables and equations are listed as follows.

External input	Measurement
Final output	Estimate
Model	A, H, Q, R
For calculation	, P _k ^-, P _k,

The model variables A and Q are used in the prediction process, and the model variables H and R are used in the estimation process. Modeling may be performed based on these variables.

[Equation 1]

[Equation 2]

and

x _k = state variable, (nx1) column vector

z _k = measurement, (mx1) column vector

A = (nxn) state transition matrix

H = (mxn) matrix

w _k = noise, (nx1) column vector

v _k= noise measurement, (mx1) column vector

w _k represents noise affecting the state variable, and v _k represents noise measured by a sensor.

The Kalman filter 20 predicts and calculates the estimate in the first and third steps. The estimate may be predicted by the following.

[Equation 3]

[Equation 4]

The estimate calculation of the Kalman filter 20 in the third step may be expressed as the following equation.

[Equation 5]

The above equation implies that noise may not be predicted, and instead, the noise may be only statistically estimated. According to this method, the Kalman filter 20 may express the noise of the state model as the following covariance matrix.

Q = covariance matrix of w _k, (nxn) diagonal matrix

R = covariance matrix of v _k, (mxm) diagonal matrix

The covariance matrix is a matrix composed of the variances of variables. Assuming that the variance of each noise is

, the covariance matrix R of the measured noise v _k may be constructed in the same manner, as follows.

[Equation 6]

The Kalman gain calculation formula may be applied to the matrix R.

[Equation 7]

Assuming that all variables in Equation 7 are scalars, the inverse matrix amounts to division. Therefore, Equation (3-25) may be expressed as follows.

[Equation 8]

In Equation (8), as R increases, the Kalman gain decreases. However, if the Kalman gain decreases, the estimate may be expressed as follows.

[Equation 9]

As the Kalman gain decreases, a ratio at which the measurement is reflected in the estimate calculation decreases. On the contrary, a ratio at which the predicted value is reflected increases. In other words, since the external measurement affects less, the change in the estimate is reduced. Therefore, the value of the matrix R is increased in order to obtain an estimate which is less affected by the measurement and slowly changes. The predicted error covariance is calculated by Equation 10.

[Equation 10]

In Equation 10, as Q increases, the predicted error covariance also increases.

Estimate prediction, Kalman gain calculation, and estimate calculation are performed in the above process.

The decision theory unit 30 detects occurrence of Parkinson's disease by applying a probabilistic decision theory based on a residual value generated by the Kalman filter 20.

The sequence of detecting the occurrence of Parkinson's disease is basically four steps.

1: Voice signal input (Happening)

2: Association with an observer by the mechanism of a voice signal

3: Observation with noise

4: Observer makes a decision on the cause associated with failure.

In order to specifically represent the above mechanism, the concept of four spaces is introduced. Thus, when an event occurs, the occurrence of the event may be represented in a message space M, and transition occurs from the message space M to a signal space S by a signaling mechanism. Then, the procedure goes through an observation space Z to a decision space D by a decision (or estimation) rule. These four spaces are illustrated in FIG. 3.

The occurrence of Parkinson's disease is detected in the process using a decision theory as described above.

The combination type generator 40 combines acoustic data based on data generated by the Kalman filter 20 by type and stores the combined acoustic data by type in the smart terminal, for fast analysis of the same type of acoustic data.

The prediction generator 50 predicts Parkinson's disease based on acoustic data analyzed by the Kalman filter 20, when voice from a patient with Parkinson's disease has a tremor due to weakening of muscles controlling intensity of the voice and the tremor increases to or above a threshold, when a bias phenomenon occurs in the bass of the voice, when a spike occurs suddenly in an output value of the voice, when the intensity of the voice suddenly decreases, when jitter of the voice rapidly increases, when noise of the voice rapidly increases and forms a pattern, when pitch of the voice gradually decreases, or when an output of the voice non-linearly changes.

The diagnostic application for Parkinson's disease according to the present disclosure, which is configured as described above, predicts a change in voice by the Kalman filter, and determines Parkinson's disease when an abnormal change occurs in the predicted value. The types to be described below are abnormal acoustic features frequently observed in patients with Parkinson's disease.

FIG. 4 illustrates an abnormal acoustic feature in which voice from a patient with Parkinson's disease has a tremor due to weakening of muscles controlling intensity of the voice, the tremor increases to or above a threshold, and thus a hard over error occurs.

FIG. 5 illustrates a case in which when a patient with Parkinson's disease utters voice, the bass of the voice is biased, which is also an abnormal speech feature.

FIG. 6 illustrates a case in which a spike occurs suddenly in an output value of voice from a patient with Parkinson's disease, which is an abnormal acoustic feature.

FIG. 7 illustrates a case in which the intensity of voice from a patient with Parkinson's disease suddenly decreases, which is an abnormal acoustic feature.

FIG. 8 illustrates a case in which jitter of voice from a patient with Parkinson's disease rapidly increases, which is an abnormal acoustic feature.

FIG. 9 illustrates a case in which noise of voice from a patient with Parkinson's disease rapidly increases and the rapidly increased noise has a specific pattern, which is an abnormal acoustic feature.

FIG. 10 illustrates an abnormal acoustic feature in which the pitch of voice from a patient with Parkinson's disease gradually decreases, that is, the pitch of the voice continuously decreases.

FIG. 11 illustrates an abnormal acoustic feature in which an output of voice from a patient with Parkinson's disease non-linearly changes, that is, the pitch of the voice gradually decreases and then gradually increases.

The eight typical feature types of signals are used to identify an acoustic feature type for diagnosis of Parkinson's disease, and the identified acoustic feature type is analyzed through the application, thus making a diagnosis of Parkinson's disease.

A process of predicting Parkinson's disease using the diagnostic application for Parkinson's disease having the above-described configuration according to the present disclosure will be described below.

When a UI of the application installed in the smart terminal is pressed to diagnose a patient as Parkinson's disease, a diagnosis page is displayed as illustrated in FIG. 12.

When a checkup start button is pressed on the diagnosis page displayed on the smart terminal and a sound /ah/ is pronounced continuously from low to high, the diagnosis starts with a sound of 0.8db or higher as a trigger.

As described above, a voice input through the smart terminal is displayed in real time on the screen (or viewer) of the smart terminal as illustrated in FIG. 13. When the diagnosis is completed 100% in about 10 seconds, the screen is initialized and a diagnosis success message indicating that the diagnosis has been successfully made is displayed as illustrated in FIG. 14.

On the contrary, when the voice is too weak to be diagnosed or the number of available measurement data is smaller than a reference and thus measurement is difficult, a message "Please measure again" is displayed as illustrated in FIG. 15.

When the diagnosis is successful as described above, the result is displayed on the screen, in comparison with the average values of patients with Parkinson's disease.

As illustrated in FIG. 16, each vertex represents an acoustic feature variable indicating a significant difference between a patient with Parkinson's disease and a normal person. If at least one feature is true in the comparison with each feature, a message recommending medical diagnosis in a clinic is displayed.

As described above, Parkinson's disease is diagnosed by subdividing a voice recognition measurement scheme according to speaker independence, a pronunciation pattern, and the number of vocabularies based on accumulated result data of Parkinson's disease diagnosis through an application installed in a smart terminal and canceling noise introduced during voice recording through a cepstrum.

While the disclosure includes details of a number of specific implementations, these should not be construed as limiting the scope of any invention or claim, but rather as a description of features that may be peculiar to a particular embodiment of a particular invention. Certain features described herein in the context of separate embodiments may be implemented in combination in a single embodiment. On the contrary, various features described in the context of a single embodiment may also be implemented in a plurality of embodiments individually or in any suitable sub-combination. Furthermore, although features operate in a particular combination and may be initially described as claimed, one or more features from a claimed combination may in some cases be excluded from the combination, and the claimed combination may be changed to a sub-combination or a modification of the sub-combination.

Embodiments described and illustrated in the disclosure and drawings are merely specific examples presented to help understanding, not intended to limit the scope of the disclosure. It will be apparent to those skilled in the art that other modifications and variations can be made in the disclosure without departing from the spirit or scope of the disclosure.

Claims

A diagnostic application for Parkinson's disease, wherein Parkinson's disease is diagnosed by extracting acoustic features of voice input to a smart terminal through frequency conversion of the voice, the acoustic features including an average, a deviation, a frequency variation (F0 variation), a harmonics to noise ratio (HNR), intensity of the voice, jitter of the voice, and shimmer of the voice, subdividing a voice recognition measurement scheme according to speaker independence, a pronunciation pattern, and the number of vocabularies based on Parkinson's disease diagnosis data through an application installed in the smart terminal, using the extracted acoustic features, and canceling noise introduced during voice recording through a cepstrum.
The diagnostic application for Parkinson's disease according to claim 1, comprising:

a model generator 10;

a Kalman filter 20;

a decision theory unit 30;

a combination type generator 40; and

a prediction generator 50,

wherein the model generator 10 is configured to store the voice and classify the stored voice into voice responsive to the Kalman filter 20 and voice unresponsive to the Kalman filter 20 through the application installed in the smart terminal by a user intending to diagnose Parkinson's disease, the Kalman filter 20 is configured to receive a measurement
of acoustic data of the voice responsive to the Kalman filter 20, which has been classified by the model generator 10 and output an estimate
by performing prediction, Kalman gain calculation, and estimate calculation, for diagnosis of Parkinson's disease for the user, the decision theory unit 30 is configured to detect occurrence of Parkinson's disease by applying a probabilistic decision theory based on a residual generated by the Kalman filter 20, the combination type generator 40 is configured to combine acoustic data based on data generated from the Kalman filter 20 by type and store the combined data by type, for fast analysis of acoustic data of the same type, and the prediction generator 50 is configured to predict Parkinson's disease based on acoustic data analyzed by the Kalman filter 20, when voice from a patient with Parkinson's disease has a tremor due to weakening of muscles controlling intensity of the voice and the tremor increases to or above a threshold, when a bias phenomenon occurs in the bass of the voice, when a spike occurs suddenly in an output value of the voice, when the intensity of the voice suddenly decreases, when jitter of the voice rapidly increases, when noise of the voice rapidly increases and forms a pattern, when pitch of the voice gradually decreases, or when an output of the voice non-linearly changes.
The diagnostic application for Parkinson's disease according to claim 2, wherein the Kalman filter 20 identifies an abnormal acoustic feature in which voice from a patient with Parkinson's disease has a tremor due to weakening of muscles controlling intensity of the voice, the tremor increases to or above a threshold, and thus a hard over error occurs.
The diagnostic application for Parkinson's disease according to claim 2, wherein the Kalman filter 20 identifies an abnormal acoustic feature in which voice from a patient with Parkinson's disease is biased in bass of the voice and thus a bias error occurs.
The diagnostic application for Parkinson's disease according to claim 2, wherein the Kalman filter 20 identifies an abnormal acoustic feature in which a spike occurs suddenly in voice from a patient with Parkinson's disease.
The diagnostic application for Parkinson's disease according to claim 2, wherein the Kalman filter 20 identifies an abnormal acoustic feature in which intensity of voice from a patient with Parkinson's disease suddenly decreases.
The diagnostic application for Parkinson's disease according to claim 2, wherein the Kalman filter 20 identifies an abnormal acoustic feature in which jitter of voice from a patient with Parkinson's disease rapidly increases.
The diagnostic application for Parkinson's disease according to claim 2, wherein the Kalman filter 20 identifies an abnormal acoustic feature in which noise of voice from a patient with Parkinson's disease rapidly increases and the rapidly increased noise has a specific pattern.
The diagnostic application for Parkinson's disease according to claim 2, wherein the Kalman filter 20 identifies an abnormal acoustic feature in which pitch of voice from a patient with Parkinson's disease gradually decreases, that is, the pitch of the voice continuously decreases.
The diagnostic application for Parkinson's disease according to claim 2, wherein the Kalman filter 20 identifies an abnormal acoustic feature in which an output of voice from a patient with Parkinson's disease non-linearly changes, that is, the pitch of the voice gradually decreases and then gradually increases.