Depression symptom determination device, determination model generation device, and learning data generation method

Abstract

The present invention comprises a depression symptom determination unit 13 for determining a depression symptom of a subject, who is subject to determination, by inputting a feature vector generated on the basis of a feature amount of a conversation carried out by the subject to a machine-learned determination model, the determination being performed by the determination model, which is generated by machine learning using, as training data, conversation data for subjects satisfying prescribed extraction conditions and exclusion conditions relating to depression symptoms. By using, as an exclusion condition, a condition that subjects diagnosed with manic depression and subjects for whom a prescribed manic depression evaluation scale score is equal to or greater than a manic depression threshold value are excluded, it becomes possible to perform machine learning of the determination model without being affected by conversation data for when a manic depression patient is temporarily in a depressed state or a manic state, and it becomes possible to determine a depression symptom of the subject in a state of having the depression symptom as a personal characteristic of the subject instead of as a temporary state.

Description

[Technical Field]

[0001] The present invention relates to a depressive symptom assessment device, a judgment model generation device, and a learning data generation method, and in particular to a device that assesses a person's depressive symptoms using a machine-learned judgment model, a device that generates the judgment model, and a method for generating learning data to be used in machine learning. [Background technology]

[0002] Conventionally, there is known a technique for estimating the presence or severity of a depressive state using an estimation model trained using training data (see, for example, Patent Document 1: WO2020 / 122227). Patent Document 1 discloses training an estimation model by machine learning using training data in which multiple types of feature quantities extracted from the biometric data of each subject are used as input vectors and the assessment of the presence or absence of a depressive state for each subject by a doctor or other expert is used as a label.

[0003] Patent Document 1 also indicates that doctors use the Hamilton Depression Scale (HAMD), a common diagnostic index for depression, to diagnose depression, and that a cutoff point for evaluation values ​​on the HAMD 17 is set at 7 points, with a diagnosis of depression being made when the total score exceeds 7 points. HAMD 17 involves a doctor or other expert asking 17 questions and assessing the severity of depression based on the responses from the subject, with each item assigned a score of 3 to 5 points (hereinafter referred to as the HAMD score) being assessed as follows: 0 to 7 points indicates normal, 8 to 13 points indicates mild, 14 to 18 points indicates moderate, 19 to 22 points indicates severe, and 23 points or more indicates extremely severe. Summary of the Invention [Problem to be solved by the invention]

[0004] The technology described in Patent Document 1 makes it possible to distinguish between healthy individuals with an estimated HAMD score of 7 or less and depressed patients with an estimated HAMD score of 8 or more, and to estimate the severity of depressed patients, by configuring an estimation model to estimate the HAMD score. Patent Document 1 explains, with reference to Figures 4 to 6, that there is a high correlation between the results of estimating the presence or severity of a depressive state using the estimation model and the results of a diagnosis by a doctor using HAMD17.

[0005] However, even if a subject actually has depressive symptoms, there are cases where the HAMD score is 7 points or less, and there is a problem that the estimation model described in Patent Document 1 may judge such a subject as a healthy person. This is because machine learning of the estimation model is performed by labeling data of subjects whose doctor's diagnosis using HAMD17 is 7 points or less as a healthy person.

[0006] The present invention has been made to solve such problems, and aims to make it possible to determine whether subjects with high scores on a depression assessment scale, as well as subjects with low scores on a depression assessment scale, have depressive symptoms using a machine-learned judgment model. [Means for solving the problem]

[0007] To solve the above-mentioned problems, the present invention determines a subject's depressive symptoms by inputting a feature vector calculated based on features of the conversation of the subject to be assessed into a machine-learned determination model. The determination model is machine-learned using, as training data, feature vectors of multiple subjects that satisfy predetermined extraction and exclusion conditions regarding depressive symptoms. Here, the extraction conditions are a condition for extracting subjects diagnosed with depression who have a predetermined depression rating scale score equal to or above the depression threshold, and subjects who have not been diagnosed with either bipolar disorder or depression. The exclusion conditions are a condition for excluding subjects diagnosed with bipolar disorder and subjects whose predetermined bipolar disorder rating scale score equals or above the bipolar disorder threshold. [Effects of the Invention]

[0008] According to the present invention configured as described above, it is possible to determine the depressive symptoms of a subject to be assessed using a machine-learned determination model that is not influenced by the feature vectors of subjects whose depression rating scale scores are equal to or greater than the depression threshold when a bipolar disorder patient is temporarily in a depressed state, or subjects whose depression rating scale scores are less than the depression threshold when a bipolar disorder patient is temporarily in a manic state. Therefore, it is possible to determine the depressive symptoms of a subject with conversational characteristics based on the characteristics of conversation when the subject is not only experiencing temporary depressive symptoms, but also when the subject has depressive symptoms that are not temporary but are a characteristic of the person. This makes it possible to determine the presence of non-temporary depressive symptoms based on the characteristics of the subject's conversation, not only for subjects whose depression rating scale scores are equal to or greater than the depression threshold, but also for subjects whose depression rating scale scores are less than the depression threshold. [Brief explanation of the drawings]

[0009] [Figure 1] 1 is a block diagram showing an example of the functional configuration of a depression symptom assessment device according to an embodiment of the present invention; [Figure 2] FIG. 2 is a block diagram showing a specific example of the functional configuration of a feature vector calculation unit according to the present embodiment. [Figure 3] 10 is a diagram for explaining a group of text index values ​​calculated by an index value vector calculation unit of the present embodiment. FIG. [Figure 4] 1 is a block diagram illustrating an example of a functional configuration of a decision model generating device according to an embodiment of the present invention. [Figure 5] 1 is a block diagram illustrating an example of a functional configuration of a learning object data generating device according to an embodiment of the present invention. [Figure 6] FIG. 10 is a diagram showing the results of depressive symptom assessment performed using the depressive symptom assessment device of this embodiment. [Figure 7] FIG. 10 is a diagram showing the results of depressive symptom assessment performed using the depressive symptom assessment device of this embodiment. [Figure 8]10A and 10B are diagrams illustrating feature amounts focused on by a determination model of the present embodiment and feature amounts focused on by a determination model generated as a comparative example. [Figure 9] 1 is a block diagram illustrating an example of the functional configuration of a training data generating device and a determination model generating device according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0010] An embodiment of the present invention will be described below with reference to the drawings. Fig. 1 is a block diagram showing an example of the functional configuration of a depressive symptom assessment device 1 according to this embodiment. As shown in Fig. 1, the depressive symptom assessment device 1 of this embodiment includes, as its functional configuration, a assessment target data input unit 11, a feature vector calculation unit 12, and a depressive symptom assessment unit 13. In addition, a assessment model storage unit 14 serving as a storage medium is connected to the depressive symptom assessment device 1 of this embodiment.

[0011] The functional blocks 11 to 13 can be configured using any of hardware, a DSP (Digital Signal Processor), and software. For example, the functional blocks 11 to 13 are realized by the operation of a program stored in a storage medium such as a RAM, a ROM, a hard disk, or a semiconductor memory under the control of a microcomputer configured with a CPU, a RAM, a ROM, etc. Instead of or in addition to the CPU, a GPU (Graphics Processing Unit), an FPGA (Field Programmable Gate Array), an ASIC (Application Specific Integrated Circuit), a DSP, etc. may be used.

[0012] The judgment target data input unit 11 inputs m pieces of conversation data each representing the content of a conversation between m subjects (m is an arbitrary integer equal to or greater than 1) who are to be judged for depressive symptoms as judgment target data. In this embodiment, as an example of conversation data, text data representing the content of the conversation is input as judgment target data.

[0013] For example, the judgment target data input unit 11 converts the audio data of a series of conversations between a subject whose depressive symptoms are unknown and a doctor into text data, extracts the text data of the subject's speech from that data, and inputs it as judgment target data.

[0014] The conversation between the subject and the doctor is conducted in the form of a medical interview, lasting, for example, 5 to 10 minutes. That is, the doctor repeatedly asks the subject questions, and the subject answers them. The conversation is input and recorded using a microphone, and the audio data of the conversation is converted into text data by manual transcription or using automatic speech recognition technology.

[0015] Here, when multiple exchanges take place between a subject and a doctor, the series of conversations will contain multiple speeches by the subject and the doctor. In this embodiment, as an example, the character data of these multiple speeches is collected and treated as a single sentence. That is, for one conversation (series of dialogue) of one subject, one sentence is defined as including two or more sentences, each generally separated by a period. This means that when the judgment target data input unit 11 inputs judgment target data of m subjects, m sentences are input.

[0016] The feature vector calculation unit 12 calculates and vectorizes the feature amounts of the conversation data input by the judgment target data input unit 11 to obtain a feature vector. When a sentence (character data) expressing the content of the conversation is used as an example of the conversation data, the feature vector calculation unit 12 calculates and vectorizes the feature amounts of the sentence. The calculation content for vectorization is arbitrary, but it is possible to calculate a feature vector by the method shown in Fig. 2, for example.

[0017] Fig. 2 is a block diagram showing a specific example of the functional configuration of feature vector calculation unit 12. As shown in Fig. 2, feature vector calculation unit 12 includes, as functional components, a word extraction unit 121, a vector calculation unit 122, and an index value vector calculation unit 123. As more specific functional components, vector calculation unit 122 includes a sentence vector calculation unit 122a and a word vector calculation unit 122b.

[0018] The word extraction unit 121 analyzes m sentences input as data to be determined by the data to be determined input unit 11, and extracts n words (n is any integer equal to or greater than 2) from the m sentences. As a method for analyzing sentences, for example, a known morphological analysis can be used. Here, the word extraction unit 121 may extract all morphemes of parts of speech divided by the morphological analysis as words, or may extract only morphemes of specific parts of speech as words.

[0019] Note that the same word may be contained multiple times in the m sentences. In this case, the word extraction unit 121 does not extract multiple instances of the same word, but extracts only one. In other words, the n words extracted by the word extraction unit 121 mean n types of words. Here, the word extraction unit 121 may measure the frequency with which the same word is extracted from the m sentences, and extract the n words (n types) with the highest frequency of appearance, or the n words (n types) with an appearance frequency equal to or higher than a threshold.

[0020] The vector calculation unit 122 calculates m sentence vectors and n word vectors from m sentences and n words. Here, the sentence vector calculation unit 122a calculates m sentence vectors each consisting of q axis components by vectorizing each of the m sentences that have been analyzed by the word extraction unit 121 into q dimensions (q is any integer equal to or greater than 2) according to a predetermined rule. Furthermore, the word vector calculation unit 122b calculates n word vectors each consisting of q axis components by vectorizing each of the n words extracted by the word extraction unit 121 into q dimensions according to a predetermined rule.

[0021] In this embodiment, as an example, sentence vectors and word vectors are calculated as follows: Now, suppose there is a set S=<d∈D,w∈W> Here, each sentence d i (i=1,2,···,m) and each word w j (j=1,2,...,n) for each sentence vector d i → and word vector w j → (hereinafter, the symbol "→" indicates a vector). Then, any word w j and any sentence d i For this, the probability P(w j |d i ) is calculated.

[0022]

number

[0023] Furthermore, this probability P(w j |d i ) can be calculated, for example, by following the probability p disclosed in the paper "Distributed Representations of Sentences and Documents" by Quoc Le and Tomas Mikolov, Google Inc., Proceedings of the 31st International Conference on Machine Learning Held in Bejing, China on June 22-24, 2014, which describes the evaluation of sentences and documents using paragraph vectors. This paper describes, for example, predicting "on" as the fourth word when there are three words, "the," "cat," and "sat," and provides a formula for calculating the prediction probability p. The probability p(wt|wt-k, ,wt+k) described in the paper is the probability of correct prediction when predicting a single word, wt, from multiple words, wt-k, ,wt+k.

[0024] In contrast, the probability P(w j |d i ) is one sentence d out of m sentences. i From n words, one word w j represents the expected probability of correct answer. i One word from w j Specifically, predicting a sentence d i appears, and the word w j This means predicting the possibility that

[0025] In equation (1), an exponential function value is used, where e is the base and the inner product value of the word vector w→ and the sentence vector d→ is the exponent. i and the word w j The exponential function value calculated from the combination of i and n words w k The ratio of the sum of n exponential function values ​​calculated from each combination of (k=1,2,...,n) to the sum of n exponential function values ​​calculated from each combination of (k=1,2,...,n) is i The first word j is calculated as the expected probability of correct answer.

[0026] where the word vector w j → and sentence vector d i The dot product value with → is the word vector w j → the sentence vector d i → is the scalar value when projected in the direction of the word vector w j → has a sentence vector d i It can also be said to be the component value in the direction of →. This is the word w j is sentence d i Therefore, the exponential function value calculated using this dot product can be used to calculate the degree to which n words w k The sum of the exponential function values ​​calculated for (k=1,2,...,n) for one word w j To find the ratio of the exponential function values ​​calculated for one sentence d i One word w out of n wordsj This is equivalent to finding the predicted probability of correctness.

[0027] In addition, equation (1) is d i and w j Since it is symmetric with respect to j From the m sentences, one sentence d i The expected probability P(d i |w j ) can be calculated for one word w j One sentence from d i To predict a word w j appears, it is sentence d i In this case, the sentence vector d i → and word vector w j The inner product value with → is the sentence vector d i → word vector w j → is the scalar value when projected in the direction of the sentence vector d i → has a word vector w j It can also be said to be the component value in the direction of →. i The word w j This can be thought of as representing the degree to which the

[0028] Note that, although an example of calculation using an exponential function value with the dot product value of the word vector w→ and the sentence vector d→ as the exponent has been shown here, the use of an exponential function value is not essential. Any calculation formula using the dot product value of the word vector w→ and the sentence vector d→ will suffice, and for example, the probability may be calculated from the ratio of the dot product value itself (however, this may include performing a predetermined calculation (for example, dot product value + 1) to ensure that the dot product value is always a positive value).

[0029] Next, the vector calculation unit 122 calculates the probability P(w j |d i ) over all sets S, the sentence vector d that maximizes the sum of L i → and word vector wj That is, the sentence vector calculation unit 122a and the word vector calculation unit 122b calculate the probability P(w j |d i ) for all combinations of m sentences and n words, and the sum of these is used as the target variable L. The sentence vector d that maximizes the target variable L is calculated as follows: i → and word vector w j →Calculate.

[0030]

number

[0031] The probability P(w j |d i ) is to maximize the sum L of the sentence d i A word w from (i=1,2,...,m) j (j=1, 2, . . . , n) maximizes the expected probability of correct answer. In other words, the vector calculation unit 122 calculates the sentence vector d that maximizes the probability of correct answer. i → and word vector w j → can be said to calculate.

[0032] As described above, in this embodiment, the vector calculation unit 122 calculates the vector of m sentences d i By vectorizing each of these into q dimensions, we obtain m sentence vectors d consisting of q axis components. i → and vectorize each of the n words into q dimensions to obtain n word vectors w consisting of q axis components. j → is calculated by varying the q axis directions and calculating the sentence vector d that maximizes the target variable L mentioned above. i → and word vector w j This is equivalent to calculating →.

[0033] The index value vector calculation unit 123 calculates m sentence vectors di → and n word vectors w j By taking the dot product of each of them, m sentences d i and n words w j In this embodiment, the index value vector calculation unit 123 calculates m×n relationship index values ​​that reflect the relationships between the m text vectors d i →each q axis component (d 11 ~d mq ) and n word vectors w j →each q axis component (w 11 ~w nq ) as elements of the word matrix W, and calculate the index value matrix DW, each element of which is an m×n relationship index value. t is the transpose of the word matrix.

[0034]

number

[0035] Each element dw of the index value matrix DW calculated in this way ij (i=1,2,···,m, j=1,2,···,n) can be said to represent the degree to which each word contributes to each sentence. For example, the element dw in the first row and second column 12 is a value that represents the degree to which word w2 contributes to sentence d1. As a result, each row of the index value matrix DW can be used to evaluate the similarity of sentences, and each column can be used to evaluate the similarity of words.

[0036] The index value vector calculation unit 123 calculates the relationship index values ​​of one sentence d by using the index value matrix DW (m×n relationship index values) calculated as in equation (3). i n relationship index values ​​dw ij A set of sentence index values ​​consisting of (j=1, 2, . . . , n) is identified as an index value vector. i The index value vector of sentence d iThe feature vector of the conversation data of subject i is output as the feature vector of

[0037] 3 is a diagram for explaining a sentence index value group (index value vector). As shown in FIG. 3, for example, in the case of the first sentence d1, the sentence index value group is the n relationship index values ​​dw included in the first row of the index value matrix DW. 11 ~dw 1n Similarly, for the second sentence d2, the n relationship index values ​​dw included in the second row of the index value matrix DW are 21 ~dw 2n The following corresponds to the mth sentence d. m A set of sentence index values ​​(n relationship index values ​​dw m1 ~dw mn ) and so on.

[0038] While the example in which the feature vector is constructed from the sentence index value group of each column in the index value matrix DW as shown in Fig. 3 has been described, the present invention is not limited to this. For example, the sentence vector calculated by the sentence vector calculation unit 122a may be used as the feature vector.

[0039] Returning to Fig. 1, the depressive symptom determination unit 13 determines the depressive symptoms of the subject by inputting the feature vector calculated by the feature vector calculation unit 12 into a machine-learned determination model stored in the determination model storage unit 14. This determination model is a model that classifies the subject to be determined into two values: either a depressed patient or a healthy subject, and takes the feature vector as input and outputs an evaluation value indicating the presence or absence of depressive symptoms.

[0040] This decision model can be generated by ensemble learning such as XGBoost, which is a gradient boosting technique. Note that the form of the decision model is not limited to this. For example, other tree models such as decision trees, regression trees, and random forests may also be used. Alternatively, a neural network model or a clustering model may also be used.

[0041] The determination model of this embodiment is machine-trained using, as training data, feature vectors of multiple subjects who satisfy predetermined extraction and exclusion conditions regarding depressive symptoms. The extraction condition is a condition for extracting subjects who have been diagnosed with depression by a doctor and whose scores on a predetermined depression assessment scale are equal to or greater than the depression threshold, as well as subjects who have not been diagnosed with either bipolar disorder or depression. The exclusion condition is a condition for excluding subjects who have been diagnosed with bipolar disorder by a doctor and whose scores on a predetermined bipolar disorder assessment scale are equal to or greater than the bipolar disorder threshold.

[0042] In this embodiment, the Hamilton Rating Scale for Depression (HAMD17) is used as an example of a depression assessment scale. As mentioned above, in the HAMD17, a subject with a HAMD score of 7 or less is generally diagnosed as a healthy subject, and a subject with a HAMD score of 8 or more is diagnosed as a depressed patient (including mild, moderate, severe, and very severe depression). Following this, in this embodiment, the depression threshold in the extraction conditions is set to 8 points, and subjects with a HAMD score of 8 or more and subjects who have not been diagnosed with either bipolar disorder or depression are extracted.

[0043] In this embodiment, the Young Mania Rating Scale (YMRS) is used as an example of a manic-depressive illness rating scale. The YMRS is a rating scale based on a clinical interview and consists of 11 items, including elation and increased activity. In this embodiment, the threshold for manic-depressive illness, which is an exclusion criterion, is set to 8 points, and training data is generated by excluding subjects whose total score for each item (hereinafter referred to as the YMRS score) is 8 points or more, and subjects who have been diagnosed with manic-depressive illness by a doctor.

[0044] In this embodiment, the judgment model is machine-trained using feature vectors calculated from the conversation data of each subject, with subjects who have been diagnosed with depression as positive examples and subjects who have not been diagnosed with depression as negative examples among the subjects who meet the above-mentioned extraction and exclusion conditions.

[0045] Fig. 4 is a block diagram showing an example of the functional configuration of a determination model generating device 2 according to this embodiment. As shown in Fig. 4, the determination model generating device 2 of this embodiment includes, as its functional configuration, a learning target data input unit 21, a feature vector calculation unit 22, and a determination model generating unit 23. In addition, a determination model storage unit 24 and a learning target data storage unit 25 are connected to the determination model generating device 2 of this embodiment as storage media.

[0046] The functional blocks 21 to 23 can be configured using any of hardware, DSP, and software. For example, the functional blocks 21 to 23 are realized by the operation of a program stored in a storage medium such as RAM, ROM, a hard disk, or a semiconductor memory under the control of a microcomputer configured with a CPU, RAM, ROM, etc. Instead of or in addition to the CPU, a GPU, FPGA, ASIC, DSP, etc. may be used.

[0047] The learning object data input unit 21 inputs, as learning object data, a plurality of conversation data representing the contents of conversations between a plurality of subjects (hereinafter referred to as condition-applicable subjects) who satisfy predetermined extraction and exclusion conditions regarding depressive symptoms. In this embodiment, as an example of conversation data, text data representing the contents of the conversation is input as learning object data.

[0048] The processing content for the learning subject data input unit 21 to input conversation data of multiple subjects as sentences is the same as that of the judgment subject data input unit 11 shown in Fig. 1. The difference from the judgment subject data input unit 11 is that the learning subject data input unit 21 inputs conversation data related to subjects that meet the conditions as learning subject data.

[0049] For example, conversation data of a subject meeting a condition (which may be voice data of the conversation or text data obtained by converting the conversation data into text) is stored in the learning object data storage unit 25. The learning object data input unit 21 inputs the learning object data by reading out the conversation data of the subject meeting a condition from the learning object data storage unit 25. Here, if voice data is stored in the learning object data storage unit 25, the learning object data input unit 21 replaces the voice data of the conversation read out from the learning object data storage unit 25 with text data, and sets this as the learning object data.

[0050] In this example, the training data stored in the training data storage unit 25 is generated by a training data generation device 3 having the functions of a training data generation unit 31, as shown in FIG. 5. In the example shown in FIG. 5, the conversation data storage unit 32 stores conversation data (which may be audio data of the conversation or text data obtained by converting the conversation data into text) of subjects who do not satisfy the specified extraction and exclusion conditions (hereinafter referred to as non-condition-satisfying subjects) in addition to conversation data of subjects who meet the conditions. The conversation data storage unit 32 also stores information necessary for determining whether the specified extraction and exclusion conditions are met, in association with the conversation data. The information necessary for determining whether the conditions are met includes information indicating whether the subject has been diagnosed with depression or bipolar disorder by a doctor, and the subject's HAMD score and YMRS score. The HAMD score and YMRS score were obtained by conducting an evaluation when the conversation data was recorded.

[0051] The learning subject data generation unit 31 generates learning subject data by extracting conversation data of subjects meeting the condition from the conversation data storage unit 32 based on information stored in association with the conversation data in the conversation data storage unit 32, and stores the generated learning subject data in the learning subject data storage unit 25. Here, the learning subject data generation unit 31 assigns a positive example label to conversation data of subjects who have been diagnosed with depression among the conversation data of the extracted condition-matching subjects, and assigns a negative example label to conversation data of subjects who have not been diagnosed with depression.

[0052] In addition, when the conversation data stored in the conversation data storage unit 32 is voice data, the learning target data generation unit 31 may store the voice data read from the conversation data storage unit 32 in the learning target data storage unit 25 as the learning target data, or may replace the voice data read from the conversation data storage unit 32 with character data and store the character data in the learning target data storage unit 25 as the learning target data.

[0053] The method for generating the learning object data is not limited to this. For example, conversations may be recorded only for subjects who satisfy predetermined extraction and exclusion conditions, and the resulting conversation data may be stored in the learning object data storage unit 25 as learning object data.

[0054] Alternatively, the function of the learning object data generation unit 31 may be provided in the learning object data input unit 21. In this case, the learning object data input unit 21 has both the function of generating and inputting learning object data. That is, the learning object data input unit 21 generates learning object data by extracting (inputting) conversation data of a subject that meets a condition from the conversation data of multiple subjects stored in the conversation data storage unit 32.

[0055] Returning to Fig. 4, the feature vector calculation unit 22 calculates and vectorizes the feature amounts of the multiple pieces of conversation data input by the learning target data input unit 21, thereby obtaining feature vectors. When a sentence (character data) expressing the content of a conversation is used as an example of conversation data, the feature vector calculation unit 22 calculates and vectorizes the feature amounts of the sentence. The process for vectorization is the same as that of the feature vector calculation unit 12 shown in Fig. 1. The feature vectors calculated by the feature vector calculation unit 22 are used as learning data when machine learning a determination model.

[0056] The claimed training data generation method is realized by the processing of the training data generation unit 31, the training data input unit 21, and the feature vector calculation unit 22. That is, the training data generation unit is configured by the training data generation unit 31, the training data input unit 21, and the feature vector calculation unit 22.

[0057] The determination model generation unit 23 generates a determination model for determining depressive symptoms of the subject based on the feature vector by performing machine learning using, as training data, the feature vector calculated by the feature vector calculation unit 22. As described above, in this embodiment, machine learning is performed using, as training data, the feature vector calculated from the learning target data generated based on the conversation data of the subject meeting the conditions.

[0058] Here, the judgment model generation unit 23 performs machine learning using feature vectors generated from conversation data of subjects meeting the conditions that are labeled as positive examples (conversation data of subjects diagnosed with depression) as positive examples, and feature vectors generated from conversation data that are labeled as negative examples (conversation data of subjects not diagnosed with depression) as negative examples.

[0059] Then, the judgment model generation unit 23 stores the judgment model generated by machine learning in the judgment model storage unit 24. The judgment model stored in the judgment model storage unit 24 is stored in the judgment model storage unit 14 shown in Fig. 1. Note that the judgment model storage unit 24 shown in Fig. 4 may be the same as the judgment model storage unit 14 shown in Fig. 1.

[0060] Although the above description has been given of an example in which the depression symptom determination device 1 and the determination model generation device 2 are configured separately, they may be configured to share some of their components. For example, the feature vector calculation units 12 and 22 may be shared.

[0061] As described above, in this embodiment, training data is constructed by excluding subjects with a YMRS score of 8 or more and subjects diagnosed with bipolar disorder by a doctor, and machine learning of a determination model is performed using the training data constructed in this manner. The determination model machine-learned using such training data can be said to be a determination model machine-learned without being influenced by conversation data of subjects whose HAMD scores are 8 or more when bipolar disorder patients are temporarily in a depressed state, or whose HAMD scores are less than 8 when bipolar disorder patients are temporarily in a manic state.

[0062] In this embodiment, the depressive symptoms of the subject to be assessed are assessed using the assessment model configured as described above. Therefore, it is possible to assess the depressive symptoms of subjects with conversational characteristics based on the characteristics of conversations when the subject is not only experiencing temporary depressive symptoms but also when the subject has depressive symptoms that are not temporary but are a characteristic of the person. As a result, even though the training data is generated using extraction conditions that limit subjects with a HAMD score of 8 or more to subjects with a HAMD score of less than 8, it is possible to determine whether the subject has a depressive symptom that is not temporary based on the characteristics of the subject's conversation.

[0063] It is generally said that there are two types of anxiety related to depressive symptoms. One is trait anxiety (trait) and the other is state anxiety (state). Trait anxiety is derived from a person's personality and refers to a tendency to become anxious, and does not change much depending on the situation at hand. On the other hand, state anxiety refers to a temporary anxiety reaction felt in response to a specific point in time, scene, event, or object. The determination model of this embodiment is particularly effective in determining the presence or absence of depressive symptoms caused by trait anxiety (trait).

[0064] In the above embodiment, the two exclusion conditions are described as subjects diagnosed with manic depression and subjects whose predetermined manic depression rating scale score is equal to or greater than the manic depression threshold. However, a further exclusion condition may be added: subjects whose depression rating scale score is equal to or greater than a second depression threshold. For example, a further condition may be added to exclude subjects whose HAMD score is 19 points or greater (patients with severe or severe depression).

[0065] The inventors confirmed that the feature vectors calculated from the conversation data of subjects with a HAMD score of 19 or more were significantly different from the feature vectors calculated from the conversation data of subjects with a HAMD score of 18 or less. Therefore, the conversation data of subjects with a HAMD score of 19 or more was excluded to generate training data, and machine learning of a judgment model was performed based on this data. It was confirmed that the accuracy of judging depressive symptoms in subjects with a HAMD score of 18 or less improved.

[0066] Figure 6 shows the results of a depressive symptom assessment performed using the depressive symptom assessment device 1 of this embodiment, with conversation data from depressed patients with a HAMD score of 8 or more and conversation data from healthy individuals as assessment targets. The results shown here are from a assessment performed using a machine-learned assessment model based on training data generated with the condition that subjects with a HAMD score of 19 or more be excluded (the same applies to Figures 7 and 8 shown below). As shown in Figure 6, the number of false negatives (FN) and false positives (FP) is very low compared to the number of true negatives (TN) and true positives (TP), with an accuracy rate of 90%, a recall rate of 89.25%, and a precision rate of 92.22%.

[0067] 7 shows the results of a depressive symptom assessment performed using the depressive symptom assessment device 1 of this embodiment, using conversation data from depressed patients with a HAMD score of 7 or less and conversation data from healthy individuals as assessment targets. As shown in FIG. 7, the number of false negatives (FN) and false positives (FP) is extremely low compared to the number of true negatives (TN) and true positives (TP), with an accuracy rate of 88.52%, a recall rate of 96.43%, and a precision rate of 87.38%. As such, depressive symptoms can be accurately assessed even for depressed patients with a HAMD score of 7 or less.

[0068] Here, to confirm that depressive symptoms can be accurately determined even for depressed patients with a HAMD score of 7 or less, a judgment model was generated by machine learning using, as a comparative example, feature vectors of subjects extracted by changing the extraction condition "HAMD score of 8 or more" to "HAMD score of 7 or less" as training data. In this way, by machine learning a judgment model using the feature vectors of subjects with a HAMD score of 7 or less as positive examples, a judgment model that can accurately determine depressive symptoms in subjects with a HAMD score of 7 or less is generated.

[0069] FIG. 8 is a diagram showing the feature amounts focused on by the judgment model of this embodiment (FIG. 8(a) on the left) and the feature amounts focused on by a judgment model generated as a comparative example (FIG. 8(b) on the right). The feature amounts shown here are elements of the feature vector calculated by the feature vector calculation unit 12. FIG. 8 shows the results of calculating the well-known Shap value as an index value indicating which element of the feature vector affected how the judgment of depressive symptoms was made.

[0070] 8(a) and 8(b), the features that the judgment model of this embodiment focuses on when determining depressive symptoms are common to many of the features that the judgment model of the comparative example focuses on when determining depressive symptoms (the common features are underlined). From the calculation results of this Shap value, it can be inferred that the judgment model generated by the judgment model generation device 2 of this embodiment can also accurately determine depressive symptoms in depressed patients with a HAMD score of 7 or less.

[0071] In the above embodiment, the character data of multiple utterances included in one conversation of a single subject is collectively defined as one sentence, but the character data of multiple utterances may be treated as multiple sentences. In this case, the determination model is generated as a model for determining depressive symptoms by inputting multiple feature vectors for one subject.

[0072] In the above embodiment, a sentence representing the content of a conversation is used as an example of conversation data, and the sentence index value group shown in FIG. 3 is used as a feature vector. However, the feature vector is not limited to this. In other words, any vector may be used as long as its elements are multiple features representing the content of the conversation or speech characteristics of the subject. For example, a feature vector may be generated by extracting multiple types of acoustic features from conversational speech (prosodic features such as pause duration, pitch, and energy measurement values; phonetic features such as fundamental frequency, formant frequency, and average Hilbert envelope; various cepstral coefficients, etc.).

[0073] In the above embodiment, as described above, an example was described in which a HAMD score of 8 points (the minimum value determined to be mild) was used as the depression threshold for the extraction condition, but this is not limiting. For example, a HAMD score of 14 points (the minimum value determined to be moderate) may be used. In the above embodiment, an example was described in which a YMRS score of 8 points was used as the manic-depressive illness threshold for the exclusion condition, but this is not limiting.

[0074] In the above embodiment, the Hamilton Depression Rating Scale (HAMD17) is used as an example of a depression rating scale, and the Young Mania Rating Scale (YMRS) is used as an example of a manic-depressive rating scale. However, this is not limiting. For example, the Hamilton Anxiety Scale (HAMA), the CPRG Depression Rating Scale (CPRG-D), the Inventory of Depressive Symptomatology (IDS), etc. may be used instead of the HAMD17. Furthermore, the Bipolar Depression Rating Scale (BDRS), the CPRG Mania Rating Scale (CPRG-M), the Manic Diagnostic and Severity Scale (MADS), etc. may be used instead of the YMRS.

[0075] In the above embodiment, the depressive symptom determination device 1 is provided with the feature vector calculation unit 12. However, the present invention is not limited to this. For example, the feature vector calculation unit 12 may be provided in a device separate from the depressive symptom determination device 1, and the feature vector generated by the separate device may be input to the depressive symptom determination device 1.

[0076] Similarly, the feature vector calculation unit 22 may be provided in a device separate from the determination model generation device 2, and the feature vector generated in the separate device may be input to the determination model generation device 2. For example, as shown in Fig. 9, a configuration may be provided that includes a training data generation device 4 shown in Fig. 9(a) and a determination model generation device 2' shown in Fig. 9(b).

[0077] As shown in FIG. 9(a), the training data generation device 4 has, as its functional configuration, a training object data generation unit 31 and a feature vector calculation unit 22. These functions are the same as those shown in FIGS. 4 and 5. The feature vector calculation unit 22 stores the calculated feature vector as training data in the training data storage unit 41. In this case, the training object data generation unit 31 and the feature vector calculation unit 22 constitute the training data generation unit.

[0078] As shown in FIG. 9(b), the judgment model generating device 2′ has, as its functional configuration, a training data input unit 42 and a judgment model generating unit 23. The function of the judgment model generating unit 23 is the same as that shown in FIG. 4. The training data input unit 42 inputs training data (feature vectors) stored in a training data storage unit 41. The judgment model generating unit 23 generates a judgment model by performing machine learning using the training data input by the training data input unit 42.

[0079] Furthermore, the above-described embodiments are merely examples of specific embodiments for carrying out the present invention, and the technical scope of the present invention should not be construed as being limited thereby. In other words, the present invention can be carried out in various forms without departing from the gist or main characteristics thereof. [Explanation of symbols]

[0080] 1. Depression symptom assessment device 2,2' Decision model generation device 3. Learning data generation device 4. Training data generation device 11. Judgment target data input section 12 Feature vector calculation unit 13 Depression Symptom Assessment Section 14 Decision model memory unit 21 Learning data input section 22 Feature vector calculation unit 23 Decision model generation unit 24 Decision model memory unit 25 Learning data storage unit 31 Learning data generation unit 32 Conversation data storage unit 41 Learning data storage unit 42 Learning data input section 121 Word Extraction Unit 122 Vector calculation unit 122a Text vector calculation unit 122b Word vector calculation unit 123 Index value vector calculation unit

Claims

1. a depression symptom determination unit that determines depression symptoms of a subject by inputting a feature vector calculated based on features of a conversation conducted by the subject to be determined into a machine-learned determination model; the determination model is machine-trained using the feature vectors of a plurality of subjects who satisfy predetermined extraction and exclusion conditions regarding depressive symptoms as training data; The extraction condition is a condition to extract subjects who have been diagnosed with depression and whose predetermined depression assessment scale score is equal to or greater than the depression threshold, and subjects who have not been diagnosed with either bipolar disorder or depression, The above exclusion criteria are to exclude subjects who have been diagnosed with bipolar disorder and subjects whose scores on a predetermined bipolar disorder rating scale are equal to or greater than the threshold for bipolar disorder. A depression symptom assessment device characterized by:

2. 2. The depressive symptom assessment device according to claim 1, wherein the exclusion condition is a condition that subjects whose depression assessment scale score is equal to or greater than a second depression threshold that is greater than the depression threshold are further excluded.

3. a determination model generation unit that performs machine learning using feature vectors calculated based on feature amounts of conversations between a plurality of subjects that satisfy predetermined extraction conditions and exclusion conditions regarding depressive symptoms, and generates a determination model for determining depressive symptoms of the subjects based on the feature vectors; The extraction condition is a condition to extract subjects who have been diagnosed with depression and whose predetermined depression assessment scale score is equal to or greater than the depression threshold, and subjects who have not been diagnosed with either bipolar disorder or depression, The above exclusion criteria are to exclude subjects who have been diagnosed with bipolar disorder and subjects whose scores on a predetermined bipolar disorder rating scale are equal to or greater than the threshold for bipolar disorder. A decision model generating device comprising:

4. 4. The apparatus for generating a judgment model according to claim 3, wherein the exclusion condition is a condition that subjects whose depression assessment scale scores are equal to or greater than a second depression threshold that is greater than the depression threshold are further excluded.

5. The judgment model generating device described in claim 3 or 4, characterized in that the judgment model generating unit performs machine learning using the feature vectors of each subject, treating subjects who have been diagnosed with depression as positive examples and subjects who have not been diagnosed with depression as negative examples among the subjects who satisfy the extraction conditions and the exclusion conditions.

6. A method for generating learning data to be used in machine learning of a determination model for determining depressive symptoms in a subject, comprising: a learning data generation unit of the computer extracting a plurality of conversation data representing the contents of conversations held by a plurality of subjects that satisfy predetermined extraction conditions and exclusion conditions set for the depressive symptoms, and generating the learning data; The extraction condition is a condition to extract subjects who have been diagnosed with depression and whose predetermined depression assessment scale score is equal to or greater than the depression threshold, and subjects who have not been diagnosed with either bipolar disorder or depression, The above exclusion criteria are to exclude subjects who have been diagnosed with bipolar disorder and subjects whose scores on a predetermined bipolar disorder rating scale are equal to or greater than the threshold for bipolar disorder. A training data generation method comprising:

7. 7. The training data generating method according to claim 6, wherein the exclusion condition is a condition that subjects whose depression assessment scale score is equal to or greater than a second depression threshold that is greater than the depression threshold are further excluded.

8. 8. The method for generating training data according to claim 6, wherein the training data is obtained by labeling subjects who have been diagnosed with depression as positive examples and subjects who have not been diagnosed with depression as negative examples, among subjects who satisfy the extraction conditions and the exclusion conditions.

Images