Artificial Intelligence Assisted Clinical Psychiatric Decision Support System

Abstract

The AI-assisted mental health medical decision support system according to the present invention may include a storage storage that stores psychiatric clinical multimodal data collected about a patient, and a local server that trains an artificial intelligence model based on the psychiatric clinical multimodal data stored in the storage storage and infers and outputs a psychiatric treatment policy for a target patient using the trained artificial intelligence model.

Description

Technology Field

[0001] The present invention relates to an AI-assisted mental health medical decision support system, and more specifically, to an AI-assisted mental health medical decision support system that can assist a physician in making decisions by training an artificial intelligence model using psychiatric clinical multimodal data and inferring a psychiatric treatment plan for a target patient. Background Technology

[0002] For a long time, in the field of psychiatric care, diagnosis and treatment strategies have been determined based on the patient's subjective complaints and the clinical experience of medical professionals. Although various information such as the patient's symptoms, medical history, and medication history is collected during the treatment process, this information is often recorded and utilized according to each medical professional's treatment style rather than being systematically structured and accumulated.

[0003] Meanwhile, with the advancement of information and communication technology and the introduction of electronic medical record systems, an environment in which various clinical information about patients is stored and retrieved in digital form has become commonplace. While such systems are useful for managing medical records, test results, and prescription history, their ability to actively support the determination of specific treatment strategies during psychiatric treatment by analyzing accumulated data is often limited.

[0004] Recently, with the introduction of artificial intelligence technology across the medical field, there has been an increasing number of attempts to utilize AI for mental health status assessment, risk prediction, simple counseling support, and brainwave analysis. However, psychiatric care in actual clinical settings is characterized by the fact that it relies entirely on subjective judgments based on diverse clinical information and the individual capabilities of medical professionals; consequently, current technology alone makes it difficult to provide consistent, real-time support for medical professionals' decision-making in the consultation room.

[0005] Furthermore, there is also a problem in that treatment guidelines presented in various literature and research papers regarding the treatment of diseases or symptoms exhibited by patients do not show consistent treatment responses when applied in actual clinical practice.

[0006] Therefore, a method to resolve these problems is required. Prior art literature

[0007] Korean Published Patent No. 10-2014-0110534 The problem to be solved

[0008] The present invention is an invention devised to solve the problems of the aforementioned prior art, and aims to train an artificial intelligence model using psychiatric clinical multimodal data to support psychiatric treatment guidelines for target patients based on the model, and to support further advanced treatment guidelines through retraining by receiving feedback on the treatment response that occurs when such treatment methods are applied to actual patients.

[0009] The problems of the present invention are not limited to those mentioned above, and other unmentioned problems will be clearly understood by those skilled in the art from the description below. means of solving the problem

[0010] The AI-assisted mental health medical decision support system of the present invention for achieving the above-mentioned purpose may include a storage storage that stores psychiatric clinical multimodal data collected about a patient, and a local server that trains an artificial intelligence model based on the psychiatric clinical multimodal data stored in the storage storage and infers and outputs a psychiatric treatment policy for a target patient using the trained artificial intelligence model.

[0011] In addition, the above psychiatric clinical multimodal data may include at least one of the following: patient symptom checkbox data, medical conversation data, medical record text data, clinical scale data, medical history data, medication history data, MRI image data, and electroencephalogram (EEG) data.

[0012] In addition, the above psychiatric treatment guidelines may include at least one of the following: a drug prescription guideline, a neuromodulation application guideline including tDCS (transcranial direct current stimulation), TMS (transcranial magnetic stimulation), TUS (transcranial ultrasound stimulation), and tACS (transcranial alternating current stimulation), a psychotherapy including psychological counseling, and a psychosocial treatment guideline including cognitive behavioral therapy (CBT).

[0013] At this time, the local server may include a learning execution unit that trains the artificial intelligence model using the psychiatric clinical multimodal data, and an inference execution unit that infers the psychiatric treatment policy by inputting the psychiatric clinical multimodal data corresponding to the target patient into the artificial intelligence model.

[0014] In addition, the local server may further include a report generation unit that generates and outputs a report based on the inference results of the inference execution unit.

[0015] Meanwhile, the local server may further include a data processing unit that receives and preprocesses the psychiatric clinical multimodal data from the storage.

[0016] In addition, the local server can receive medical feedback data input from a medical terminal owned by a medical professional and reflect it in the retraining of the artificial intelligence model.

[0017] At this time, the medical feedback data may include at least one of the physician's drug prescription decision data, drug side effect data, and treatment effect assessment data.

[0018] In addition, the local server can process the artificial intelligence model to infer the psychiatric treatment policy by referring to a previously stored clinical paper database.

[0019] In addition, the local server can provide a psychiatric clinical application capable of inputting and querying the psychiatric clinical multimodal data and querying the psychiatric treatment policy output by the artificial intelligence model. Effects of the invention

[0020] The AI-assisted mental health medical decision support system of the present invention for solving the above-mentioned problem has the advantage of being able to provide data-based consistent treatment policy inference in the psychiatric treatment process, which previously relied on the personal experience and subjective judgment of medical professionals, by systematically storing psychiatric clinical multimodal data in a storage and utilizing it for the training of an artificial intelligence model.

[0021] In addition, the present invention has the advantage of enabling medical professionals to quickly receive objective reference information in the process of determining treatment policies for individual patients by having a local server infer and output a psychiatric treatment policy for a target patient using a trained artificial intelligence model.

[0022] In addition, since the present invention has a structure for training an artificial intelligence model by integrating various forms of clinical data in a multimodal manner, it has the advantage of enabling inference of psychiatric treatment guidelines from a more comprehensive perspective compared to analysis based on a single type of data.

[0023] The effects of the present invention are not limited to those mentioned above, and other unmentioned effects will be clearly understood by those skilled in the art from the description in the claims. Brief explanation of the drawing

[0024] FIG. 1 is a diagram showing a flowchart of an AI-assisted mental health medical decision support system according to one embodiment of the present invention. FIG. 2 is a diagram showing detailed items of psychiatric clinical multimodal data in an AI-assisted mental health medical decision support system according to one embodiment of the present invention. FIG. 3 is a diagram showing detailed items of a psychiatric treatment policy in an AI-assisted mental health medical decision support system according to one embodiment of the present invention. FIG. 4 is a diagram showing the sub-configuration of a local server in an AI-assisted mental health medical decision support system according to one embodiment of the present invention. FIG. 5 is a diagram showing detailed items of medical feedback data in an AI-assisted mental health medical decision support system according to one embodiment of the present invention. Specific details for implementing the invention

[0025] In this specification, where a component (or region, layer, part, etc.) is described as being "on," "connected," or "combined" with another component, it means that it may be directly placed / connected / combined with the other component, or that a third component may be placed between them.

[0026] Identical reference numerals denote identical components. Additionally, in the drawings, the thicknesses, proportions, and dimensions of the components are exaggerated for the effective illustration of the technical content.

[0027] "And / or" includes all one or more combinations that the associated configurations can define.

[0028] Terms such as "first," "second," etc., may be used to describe various components, but said components should not be limited by said terms. These terms are used solely for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be named the second component, and similarly, the second component may be named the first component. A singular expression includes a plural expression unless the context clearly indicates otherwise.

[0029] Additionally, terms such as "below," "lower side," "above," and "upper side" are used to describe the relationships between the components depicted in the drawings. These terms are relative concepts and are described based on the directions indicated in the drawings.

[0030] Unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as generally understood by those skilled in the art to which the present invention pertains. Additionally, terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and are explicitly defined herein unless interpreted in an ideal or overly formal sense.

[0031] Terms such as "include" or "have" are intended to specify the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0032] Furthermore, when it is stated in this specification that a first component operates or is executed on (ON) a second component, it should be understood that the first component operates or is executed in an environment where the second component operates or is executed, or operates or is executed through direct or indirect interaction with the second component.

[0033] Where any component, device, or system is described as including a component consisting of a program or software, it should be understood that, even without explicit mention, that component, device, or system includes hardware (e.g., memory, CPU, etc.) or other programs or software (e.g., an operating system or drivers required to run the hardware) necessary for the execution or operation of that program or software.

[0034] Furthermore, unless otherwise specified regarding the implementation of a component, it should be understood that the component may be implemented in software, hardware, or both software and hardware.

[0035] Furthermore, the terms used herein are for describing the embodiments and are not intended to limit the invention. In this specification, the singular form includes the plural form unless specifically stated otherwise in the text. As used herein, 'comprises' and / or 'comprising' do not exclude the presence or addition of one or more other components to the mentioned components.

[0036] Additionally, in this specification, terms such as 'part', 'device', etc., may be intended to refer to hardware and the functional and structural combination of software driven by said hardware or for driving said hardware. For example, the hardware here may be a data processing device including a CPU or other processor. Furthermore, the software driven by the hardware may refer to a running process, object, executable, thread of execution, program, etc.

[0037] Furthermore, it can be easily inferred by an average expert in the art of the present invention that the above terms may refer to a specific code and a logical unit of hardware resources for executing the said specific code, and do not necessarily refer to physically connected code or a single type of hardware.

[0038] FIG. 1 is a diagram showing a flowchart of an AI-assisted mental health medical decision support system according to one embodiment of the present invention.

[0039] Figure 2 is a diagram showing detailed items of psychiatric clinical multimodal data in an AI-assisted mental health medical decision support system according to one embodiment of the present invention, and Figure 3 is a diagram showing detailed items of psychiatric treatment policies in an AI-assisted mental health medical decision support system according to one embodiment of the present invention.

[0040] In addition, FIG. 4 is a diagram showing the sub-configuration of a local server (200) in an AI-assisted mental health medical decision support system according to one embodiment of the present invention, and FIG. 5 is a diagram showing the detailed items of medical feedback data in an AI-assisted mental health medical decision support system according to one embodiment of the present invention.

[0041] Referring collectively to FIGS. 1 to 5, the AI-assisted mental health medical decision support system according to the present embodiment may include a storage (100) and a local server (200).

[0042] The storage (100) can store psychiatric clinical multimodal data collected from patients. Such storage (100) can be implemented as a cloud-based storage service and can receive and store psychiatric clinical multimodal data received from a medical professional terminal (10). However, the storage (100) can be implemented in various forms other than the cloud form, such as an on-premise database server or a network attached storage (NAS), and is not limited to only these forms.

[0043] Psychiatric clinical multimodal data is a concept that encompasses multiple different types of data collected during a patient's psychiatric treatment process. As illustrated in FIG. 2, in this embodiment, psychiatric clinical multimodal data may include at least one of the following: patient symptom checkbox data, medical conversation data, medical record text data, clinical scale data, medical history data, medication history data, MRI image data, and electroencephalogram (EEG) data.

[0044] Symptom checkbox data may be structured data in which symptom items complained of by a patient are selected in a predefined checkbox format. Symptom checkbox data may be generated by a medical professional or a patient selecting the corresponding item through a medical professional terminal (10).

[0045] Medical conversation data may be data recorded in text form containing the content of conversations that take place between a patient and a medical professional during the medical process. Such medical conversation data can be collected or processed using large-scale language models (LLM).

[0046] Medical record text data may be data stored in text form of medical records written by a medical professional during the course of treatment. Such medical record text data may include descriptive content recorded by the medical professional during treatment, such as the patient's main complaint, current medical history, findings from a mental status examination, diagnosis, and treatment plan. Medical record text data may be extracted from an electronic medical record system or collected by the medical professional directly inputting it through a medical professional terminal (10).

[0047] Clinical scale data may be quantitative score data calculated using standardized clinical assessment tools for the evaluation of psychiatric disorders such as depression, anxiety, and schizophrenia. For example, scores calculated by assessment tools such as the Beck Depression Inventory (BDI), Pittsburgh Sleep Quality Index (PSQI), Patient Health Questionnaire-9 (PHQ-9), Generalized Anxiety Disorder-7 (GAD-7), and Positive and Negative Syndrome Scale (PANSS) may correspond to clinical scale data. However, clinical scale data may be calculated by various forms of assessment tools and is not limited to these forms.

[0048] Past medical data may include the patient's past medical history, family history, hospitalization history, previous diagnoses, etc.

[0049] Drug treatment history data may include the type, dosage, duration of use, and history of changes of medications that the patient is currently taking or has taken in the past.

[0050] MRI image data may be magnetic resonance imaging data capturing the patient's brain structure. Such MRI image data can be obtained through structural MRI or functional MRI (fMRI).

[0051] Electroencephalogram (EEG) data can be electrical signal data of the brain measured through electrodes attached to a patient's scalp. Such EEG data can be utilized for analysis by specific frequency bands, event-related potential (ERP) analysis, and quantitative EEG analysis.

[0052] The local server (200) can train an artificial intelligence model (300) based on psychiatric clinical multimodal data stored in the storage (100), and can infer and output a psychiatric treatment plan for a target patient using the trained artificial intelligence model (300). Such a local server (200) can be implemented as a computing device equipped with a GPU (Graphics Processing Unit) and can be installed and operated within a medical institution.

[0053] The artificial intelligence model (300) may be a computational model that is learned and executed by the local server (200). In this case, the artificial intelligence model (300) may be loaded and executed inside the local server (200), and in some cases, may be loaded and executed on an external server connected to the local server (200) via a network.

[0054] Such an artificial intelligence model (300) can be trained to receive psychiatric clinical multimodal data as input and to produce a psychiatric treatment plan for a target patient as output.

[0055] The artificial intelligence model (300) can be implemented as a deep learning-based neural network model, for example, a multilayer perceptron (MLP), a convolutional neural network (CNN), a recurrent neural network (RNN), a transformer structure, etc., can be applied alone or in combination. However, the artificial intelligence model (300) can be implemented as various forms of machine learning or deep learning architectures, and is not limited to only such forms.

[0056] The artificial intelligence model (300) may be formed with a structure including a plurality of submodules that process different types of psychiatric clinical multimodal data, and an integrated inference module that infers psychiatric treatment guidelines by integrating feature values ​​output from the plurality of submodules.

[0057] As illustrated in FIG. 4, in this embodiment, the local server (200) may include a data processing unit (210), a learning execution unit (220), an inference execution unit (230), and a report generation unit (240).

[0058] The data processing unit (210) can receive and preprocess psychiatric clinical multimodal data from the storage (100). Such preprocessing may include converting different types of data into a form that can be input into an artificial intelligence model (300), processing missing values, or performing normalization of the data.

[0059] The learning execution unit (220) can train an artificial intelligence model (300) using psychiatric clinical multimodal data preprocessed by the data processing unit (210). Such a learning execution unit (220) can be configured to train the artificial intelligence model (300) in a batch manner during a pre-set time period outside of treatment hours. That is, psychiatric clinical multimodal data collected during the treatment process during the day and accumulated in the storage (100) can be collectively reflected in the training of the artificial intelligence model (300) during times when treatment is not performed, such as at night. The reason for doing this is to prevent the inference performance of the inference execution unit (230) from being affected during treatment hours, as the computational resources of the local server (200) are intensively consumed during the training process of the artificial intelligence model (300).

[0060] The inference execution unit (230) can infer a psychiatric treatment policy by inputting psychiatric clinical multimodal data corresponding to the target patient into the learned artificial intelligence model (300). Additionally, the inference execution unit (230) can process the artificial intelligence model (300) to infer a psychiatric treatment policy by referring to a previously stored clinical paper database (400). Such a clinical paper database (400) may be stored inside a local server (200) or provided to be accessible from an external database. The inference execution unit (230) can derive supporting literature corresponding to the inferred psychiatric treatment policy by comparing the inference result of the artificial intelligence model (300) with the literature information contained in the clinical paper database (400).

[0061] As illustrated in FIG. 3, in this embodiment, the psychiatric treatment policy may include at least one of the following: a drug prescription policy, a neuromodulation application policy including tDCS (transcranial direct current stimulation), TMS (transcranial magnetic stimulation), TUS (transcranial ultrasound stimulation), and tACS (transcranial alternating current stimulation), and a psychosocial treatment policy including psychological counseling, psychotherapy, and cognitive behavioral therapy (CBT).

[0062] The drug prescription policy may include inference results regarding whether to change the currently prescribed drug for the patient, the type of drug, and the direction of dosage adjustment.

[0063] The application policy for tDCS (transcranial direct current stimulation) may include inference results regarding whether to apply transcranial direct current stimulation treatment and the stimulation conditions for the target patient. tDCS is a non-invasive brain stimulation technique that modulates the activity of the cerebral cortex by applying a minute direct current through electrodes placed on the scalp.

[0064] The application policy for TMS (transcranial magnetic stimulation) may include inference results regarding whether to apply transcranial magnetic stimulation treatment and the stimulation conditions for the target patient. TMS is a technique that non-invasively stimulates specific areas of the brain using a magnetic field generated by an electric current flowing through a coil.

[0065] The application policy for TUS (transcranial ultrasound stimulation) may include inference results regarding whether to apply transcranial ultrasound stimulation to the target patient and the stimulation conditions. TUS is a technique that uses ultrasound to non-invasively stimulate specific areas of the brain.

[0066] The application policy for tACS (Transcranial Alternating Current Stimulation) may include inference results regarding whether to apply tACS to the target patient and the stimulation conditions. tACS is a technique that non-invasively stimulates brain activity using alternating current.

[0067] Psychosocial treatment guidelines, including psychological counseling, psychotherapy, and cognitive behavioral therapy (CBT), may include inference results regarding whether to apply psychosocial treatment, including psychological counseling, to the patient and the direction of counseling. Such guidelines for applying psychological counseling can be generated using a large-scale language model (LLM).

[0068] The report generation unit (240) can generate and output a report based on the inference results of the inference execution unit (230). This report may include detailed information on psychiatric treatment policies, as well as source literature information derived by the inference execution unit (230) by referring to the clinical paper database (400). The report generated by the report generation unit (240) can be output so that a medical professional can view it through a medical professional terminal (10).

[0069] Additionally, the local server (200) can receive medical feedback data input from a medical terminal (10) owned by a medical professional and reflect it in the retraining of the artificial intelligence model (300). Such medical feedback data may be data input by a medical professional through the medical terminal (10) after performing actual medical treatment on a target patient by referring to the inference results of the artificial intelligence model (300).

[0070] As illustrated in FIG. 5, in this embodiment, the medical feedback data may include at least one of the following: data on a doctor's drug prescription decision, data on drug side effects, data on neuromodulation prescription, data on side effects after neuromodulation, and data on the determination of treatment effect.

[0071] The doctor's drug prescription decision data may be data including the type, dosage, and changes of the drug that the medical professional ultimately decided on for the patient. Such doctor's drug prescription decision data may be used for the artificial intelligence model (300) to learn the relationship between the drug prescription policy inferred by the artificial intelligence model (300) and the medical professional's actual prescription decision.

[0072] Drug side effect data may be data recording the types and severity of side effects experienced by the subject patient during the course of taking the drug.

[0073] Neuromodulation prescription data may include the type of neuromodulation, stimulation site, stimulation intensity, stimulation frequency, and number of procedures finally determined by a medical professional for the target patient.

[0074] Data on adverse effects following neuromodulation may be data recording the types and severity of adverse effects experienced by the subject patient after the neuromodulation procedure. Such data on adverse effects following neuromodulation may include details of adverse reactions reported after the procedure, such as headache, dizziness, skin irritation, and tinnitus.

[0075] Treatment efficacy assessment data may be data recording the degree of treatment response assessed by medical professionals evaluating the course of treatment for a patient. Such treatment efficacy assessment data can be calculated based on changes in clinical scale data performed by the patient, changes in brain waves, LLM data during patient consultation, and clinical global impression scales (CGI) used by the attending physician to evaluate the patient's side effects and degree of symptom improvement.

[0076] The learning execution unit (220) can incorporate medical feedback data into the retraining of the artificial intelligence model (300) along with psychiatric clinical multimodal data. That is, the artificial intelligence model (300) can be trained in a supervised learning manner using psychiatric clinical multimodal data as input and medical feedback data as labels. With this structure, the artificial intelligence model (300) can be continuously updated to reflect the actual prescription decision patterns of medical professionals and the treatment progress accordingly.

[0077] Additionally, the local server (200) may provide a psychiatric clinical application capable of inputting and viewing psychiatric clinical multimodal data and viewing psychiatric treatment guidelines output by the artificial intelligence model (300). Such a psychiatric clinical application may be installed on a medical professional terminal (10) or provided in a form accessible through a web browser. Through the psychiatric clinical application, the medical professional can input or view the patient's psychiatric clinical multimodal data and view the psychiatric treatment guidelines inferred by the artificial intelligence model (300) in the form of a report. Additionally, the medical professional can input medical feedback data through the psychiatric clinical application and transmit it to the local server (200).

[0078] As described above, the AI-assisted mental health medical decision support system according to the present invention can train an artificial intelligence model (300) based on psychiatric clinical multimodal data stored in a storage storage (100), and use the trained artificial intelligence model (300) to infer and output a psychiatric treatment policy for a target patient.

[0079] In addition, by reflecting the medical feedback data received from the medical terminal (10) into the retraining of the artificial intelligence model (300), the artificial intelligence model (300) can be continuously updated to reflect the medical doctor's prescription decision patterns and treatment progress. Furthermore, by the local server (200) providing a psychiatric clinical application, the medical doctor can perform input and retrieval of psychiatric clinical multimodal data and view psychiatric treatment policies through a single interface.

[0080] The AI-assisted mental health medical decision support system according to one embodiment of the present invention has been described in detail above. Below, various features additionally applicable to the present invention will be described.

[0081] First, the inference execution unit (230) can refer to the psychiatric clinical multimodal data of existing patients stored in the storage (100) during the process of inferring a psychiatric treatment policy for the target patient.

[0082] Specifically, the inference execution unit (230) can calculate the similarity between the psychiatric clinical multimodal data of a target patient and the psychiatric clinical multimodal data of each of a plurality of existing patients already stored in the storage (100). Such similarity may include item-specific similarity calculated individually for each of the plurality of data items included in the psychiatric clinical multimodal data, and a comprehensive similarity calculated by weighted summing the item-specific similarities.

[0083] The method for calculating item-specific similarity may be applied differently depending on the type of data item. For items corresponding to numerical data, such as clinical scale data and drug treatment history data, item-specific similarity may be calculated by dividing the difference between the numerical value of the target patient and the numerical value of the existing patient by the value range of each item to normalize it, and then subtracting the normalized difference value from 1. For items corresponding to text-type data, such as medical conversation data, item-specific similarity may be calculated by converting the medical conversation data of the target patient and the medical conversation data of the existing patient into fixed-length numerical arrays using a natural language processing model, respectively, and then calculating the degree of directional agreement between the two converted numerical arrays. For items corresponding to signal or image-type data, such as electroencephalogram (EEG) data and MRI image data, item-specific similarity may be calculated by calculating the degree of directional agreement between fixed-length numerical arrays extracted from the intermediate layer of the artificial intelligence model (300). However, the method for calculating item-specific similarity may be implemented by various distance calculation methods or similarity calculation methods, and is not limited to only such methods.

[0084] In the weighted summation process for calculating the overall similarity, pre-set item-specific weights may be applied to each data item. Such item-specific weights may be set manually by a medical professional, or may be automatically determined during the learning process of the artificial intelligence model (300).

[0085] The inference execution unit (230) can compare the calculated total similarity with a preset similarity threshold value and select existing patients whose total similarity is greater than or equal to the similarity threshold value as similar patients. Such a similarity threshold value may be set by a medical professional or may be automatically calculated by the local server (200) based on the number of existing patients and data distribution stored in the storage (100).

[0086] Furthermore, in the process of selecting similar patients, the inference execution unit (230) may consider not only the similarity at a single point in time but also the temporal change trajectory of psychiatric clinical multimodal data collected over multiple points in time. The inference execution unit (230) may generate a treatment trajectory of a target patient by arranging psychiatric clinical multimodal data collected at multiple points in time for the target patient in a time-series order. Such a treatment trajectory may include the trend of change over time of clinical scale data, the sequence of change history of drug treatment history data, and the temporal fluctuation pattern of brainwave (EEG) data.

[0087] The inference execution unit (230) can calculate the trajectory similarity between the treatment trajectory of the target patient and the treatment trajectory of each existing patient. Specifically, the inference execution unit (230) corresponds the numerical item values ​​at each point in time constituting the two treatment trajectories in order, but if the time series lengths of the two treatment trajectories are different or the intervals between medical visits are uneven, it can establish a correspondence relationship by non-linearly stretching the time axis. Such non-linear stretching of the time axis can be performed by searching for a corresponding path where the cumulative sum of the differences in numerical item values ​​between each point in time of the two time series is minimized. The inference execution unit (230) can convert the cumulative difference value according to the searched corresponding path into trajectory similarity.

[0088] The inference execution unit (230) can calculate a final similarity by combining the previously calculated comprehensive similarity and trajectory similarity. In the process of calculating such a final similarity, a first weight set for the comprehensive similarity and a second weight set for the trajectory similarity may be applied, respectively. The ratio of the first weight and the second weight may be manually set by a medical professional, or the local server (200) may automatically calculate it based on the characteristics of the data accumulated in the storage (100). The inference execution unit (230) can select existing patients whose final similarity is greater than or equal to a pre-set final similarity threshold as final similar patients.

[0089] The inference execution unit (230) can retrieve the treatment progress and prescription history of the selected final similar patient from the storage (100). Such treatment progress may include treatment effect assessment data and drug side effect data included in the medical feedback data collected for the final similar patient, and the prescription history may include the doctor's drug prescription decision data collected for the final similar patient.

[0090] The inference execution unit (230) can extract numerical item values ​​from clinical scale data, drug treatment history data, and electroencephalogram (EEG) data of each final similar patient, and calculate an item-specific weighted average by applying the final similarity between the corresponding final similar patient and the target patient as a weight to each extracted item value. The inference execution unit (230) can generate a similar patient reference array by arranging the calculated item-specific weighted average values ​​in a row according to the item order.

[0091] Subsequently, the inference execution unit (230) can combine a numeric array extracted from the target patient's own psychiatric clinical multimodal data in the same order of items with a similar patient reference array. This combination can be performed by concatenating the two arrays in order, or by arranging pairs of the target patient's value and the weighted average value of the similar patient reference array for each item. The inference execution unit (230) can pass the combined array as input to an artificial intelligence model (300) to infer a psychiatric treatment plan for the target patient.

[0092] Additionally, the inference execution unit (230) can distinguish between final similar patients whose treatment effect judgment data is above a preset treatment response threshold and final similar patients whose treatment response judgment data is below the preset treatment response threshold by referring to the treatment effect judgment data included in the treatment feedback data within the final similar patient group. The inference execution unit (230) can separately extract the physician's drug prescription decision data for final similar patients whose treatment response threshold is above the threshold and can calculate a recommended drug candidate for a drug in which the number of prescriptions for the same drug is above a preset frequency threshold from the physician's drug prescription decision data. Such recommended drug candidates can be included in a report by the report generation unit (240) and provided to medical professionals along with the inference results of the artificial intelligence model (300).

[0093] When a similar patient-based inference process is performed, the report generation unit (240) can output a report including the number of final similar patients, the distribution of final similarity, a summary of the treatment progress of the final similar patients, a summary of prescription history, and recommended drug candidates.

[0094] Next, the inference execution unit (230) can infer a drug prescription policy as a psychiatric treatment policy, and then verify whether there is a risk of interaction between the drugs included in the inferred drug prescription policy and the drugs currently being taken by the target patient.

[0095] A local server (200) may be provided to store a drug interaction reference table or to access an external drug interaction reference table. Such a drug interaction reference table may be data in a table structure that assigns a unique drug identification code to each of a plurality of drugs and maps interaction information corresponding to a combination of two drug identification codes. The interaction information may include an interaction type code and a risk class code. The risk class code may be classified into a first class corresponding to contraindication for concomitant use, a second class corresponding to caution for concomitant use, and a third class corresponding to permission for concomitant use. The interaction type code may be classified into pharmacokinetic interactions and pharmacodynamic interactions. A pharmacokinetic interaction corresponds to a type where one drug affects the absorption, distribution, metabolism, or excretion of another drug, and a pharmacodynamic interaction corresponds to a type where two drugs act on the same or related receptors, thereby changing the pharmacological effect.

[0096] Additionally, the drug interaction reference table may include pharmacological classification codes corresponding to each drug identification code. Such pharmacological classification codes may be based on a classification system assigned according to the drug's mechanism of action; for example, classifications such as selective serotonin reuptake inhibitors (SSRIs), serotonin-norepinephrine reuptake inhibitors (SNRIs), atypical antipsychotics, and benzodiazepines may apply.

[0097] The inference execution unit (230) can generate a current drug list by extracting the drug identification code of each drug currently being taken from the drug treatment history data of the target patient. Subsequently, the inference execution unit (230) can generate pairs of drugs to be verified by combining the drug identification code of the drug included in the inferred drug prescription policy with each drug identification code included in the current drug list. For each generated pair of drugs to be verified, the inference execution unit (230) can query a drug interaction reference table to verify the corresponding risk grade code and interaction type code.

[0098] If there is a pair of drugs among the pairs of drugs to be verified that corresponds to a first-grade risk grade code, the inference execution unit (230) may generate contraindication warning information regarding the pair of drugs. Such contraindication warning information may include the drug identification code, risk grade code, interaction type code, and expected adverse reaction details of the two drugs constituting the pair of drugs.

[0099] In addition, the inference execution unit (230) can search for alternative drug candidates that can replace a drug corresponding to Grade 1 from the drug interaction reference table. Such search for alternative drug candidates can be performed through the following process. The inference execution unit (230) can verify the pharmacological classification code of the drug corresponding to Grade 1 in the drug interaction reference table. Subsequently, the inference execution unit (230) can extract drugs corresponding to Grade 3 in risk grade codes with all drugs currently included in the drug list from among drugs having the same pharmacological classification code in the drug interaction reference table. If there are multiple extracted drugs, the inference execution unit (230) can refer to the drug prescription decision data of existing patients stored in the storage (100) and select alternative drug candidates in order of highest prescription frequency among the extracted drugs, up to a pre-set number of alternative candidates.

[0100] If there is a pair of drugs among the pairs of drugs subject to verification whose risk class code corresponds to the second class, the inference execution unit (230) may generate caution information regarding the pair of drugs. Such caution information may include the drug identification code, risk class code, interaction type code, and recommended monitoring items for combined use of the two drugs constituting the pair of drugs. The monitoring items may be information previously stored in a drug interaction reference table corresponding to the interaction type code of the pair of drugs. For example, if the blood concentration of a specific drug may fluctuate due to pharmacokinetic interactions, the measurement of the blood concentration of the drug may be included as a monitoring item.

[0101] The inference execution unit (230) can correct the drug prescription policy based on the results of the drug interaction verification. If a drug corresponding to Grade 1 is included, the inference execution unit (230) can generate a corrected drug prescription policy by replacing the drug with one of the alternative drug candidates. If a drug corresponding to Grade 2 is included, the inference execution unit (230) can maintain the drug prescription policy but add caution information.

[0102] The report generation unit (240) can output a report containing contraindication warning information, caution information, alternative drug candidates, and corrective drug prescription guidelines generated during the drug interaction verification process. By viewing the report through the medical professional terminal (10), the medical professional can determine the final prescription by referring to the verification results regarding drug interactions along with the inferred drug prescription guidelines. In addition, it can also guide the determination of the location and intensity of neuromodulation in this manner.

[0103] Preferred embodiments according to the present invention have been described above, and it is obvious to those skilled in the art that, in addition to the embodiments described above, the present invention may be embodied in other specific forms without departing from the spirit or scope thereof. Therefore, the embodiments described above should be regarded as illustrative rather than restrictive, and accordingly, the present invention is not limited to the description above but may be modified within the scope of the appended claims and their equivalents. Explanation of the symbols

[0104] 10: Medical Personnel Terminal 100: Storage 200: Local Server 210: Data processing unit 220: Learning Execution Unit 230: Inference Execution Unit 240: Report Generation Section 300: Artificial Intelligence Model 400: Clinical paper database

Claims

Claim 1 An AI-assisted mental health medical decision support system comprising: a storage storage for storing psychiatric clinical multimodal data collected about a patient; and a local server that trains an artificial intelligence model based on the psychiatric clinical multimodal data stored in the storage storage and infers and outputs a psychiatric treatment policy for a target patient using the trained artificial intelligence model; wherein the local server comprises: a learning execution unit that trains the artificial intelligence model using the psychiatric clinical multimodal data; and an inference execution unit that inputs the psychiatric clinical multimodal data corresponding to the target patient into the artificial intelligence model to infer the psychiatric treatment policy; wherein the artificial intelligence model comprises: a plurality of sub-modules that process different types of the psychiatric clinical multimodal data respectively; and an integrated inference module that infers the psychiatric treatment policy by integrating feature values ​​output from the plurality of sub-modules; and wherein the local server receives medical feedback data input from a medical terminal owned by a medical professional and reflects it in the retraining of the artificial intelligence model, and wherein the learning execution unit retrains the artificial intelligence model using the psychiatric clinical multimodal data as input and the medical feedback data as labels in a supervised learning manner. Claim 2 In claim 1, the psychiatric clinical multimodal data comprises at least one of patient symptom checkbox data, medical conversation data, medical record text data, clinical scale data, medical history data, drug treatment history data, MRI image data, and electroencephalogram (EEG) data, an AI-assisted mental health medical decision support system. Claim 3 An AI-assisted mental health decision support system according to claim 1, wherein the psychiatric treatment policy comprises at least one of a drug prescription policy, a neuromodulation application policy including tDCS (transcranial direct current stimulation), TMS (transcranial magnetic stimulation), TUS (transcranial ultrasound stimulation), and tACS (transcranial alternating current stimulation), and a psychosocial treatment policy including psychological counseling, psychotherapy, and cognitive behavioral therapy (CBT). Claim 4 delete Claim 5 In claim 1, the local server further comprises a report generation unit that generates and outputs a report based on the inference result of the inference execution unit, an AI-assisted mental health medical decision support system. Claim 6 In claim 1, the local server further comprises a data processing unit that receives and preprocesses the psychiatric clinical multimodal data from the storage, an AI-assisted mental health medical decision support system. Claim 7 delete Claim 8 An AI-assisted mental health medical decision support system according to claim 1, wherein the medical feedback data comprises at least one of the following: data on a physician's drug prescription decision, data on drug side effects, data on neuromodulation prescription, data on side effects after neuromodulation, changes in clinical scale data performed by the patient and changes in brain waves, LLM data during patient treatment, and treatment effect assessment data including a clinical global impression scale (CGI) in which the physician evaluates the patient's side effects and degree of symptom improvement. Claim 9 In claim 1, the local server processes the artificial intelligence model to infer the psychiatric treatment policy by referring to a previously stored clinical paper database, an AI-assisted mental health medical decision support system. Claim 10 In claim 1, the local server provides an AI-assisted mental health medical decision support system that enables the input and retrieval of psychiatric clinical multimodal data and a psychiatric clinical application capable of retrieving psychiatric treatment policies output by the artificial intelligence model.

Images