Device for artificial intelligence-based cardiovascular examination based on skin-reflected waves

Abstract

The present invention relates to a device for on-device artificial intelligence-based cardiovascular risk level measurement based on skin-reflected waves, the device comprising: a collection unit for collecting photoplethysmography (PPG) of a specific wavelength from a wearable device; a storage unit for storing the collected wavelength and previous data; a wavelength generation unit for generating one or more multi-wavelengths by inputting the collected photoplethysmography to a first artificial intelligence model; a detection unit for extracting a biomarker value by inputting the generated multi-wavelengths and the previous data to a second artificial intelligence model; and a diagnosis unit for providing diagnosis based on the detected biomarker value.

Description

AI-based cardiovascular screening device based on skin reflection waves

[0001] The present invention was carried out under the support of the Ministry of Science and ICT under project unique number 2710068397 and project number 00265393, the research management agency for the said project is the National Research Foundation of Korea, the research project name is "Brain Science Leading Convergence Technology Development Project", the research project name is "Development and Verification of Deep Learning-based Cerebrovascular Disease in Slilico Model Using Multimodal Database", the lead institution is Samsung Seoul Hospital, and the research period is 2025.01.01 ~ 2026.04.30. (2023.07.01. ~ 2026.04.30.).

[0002] The present invention relates to an on-device artificial intelligence-based cardiovascular screening device using photoplethysmography.

[0003]

[0004] With the recent advancement of wearable devices, interest in personal health monitoring is increasing. In particular, technology that measures biosignals such as heart rate and blood oxygen saturation using photoplethysmography (PPG) signals through devices like smartwatches is being widely utilized.

[0005] Traditionally, red or infrared light was used for PPG signal measurement; however, these wavelengths are sensitive to motion, making it difficult to acquire accurate signals. To address this, PPG measurement utilizing green wavelengths has been introduced, which provides a more stable signal due to its lower sensitivity to motion.

[0006] However, there are limitations to measurements using only green wavelengths. While utilizing multi-wavelength data allows for obtaining more precise biometric information through signals reflected from various depths of the skin, it is difficult to incorporate LEDs and sensors of multiple wavelengths due to the physical constraints of wearable devices.

[0007] Furthermore, conventional biosignal analysis typically involves transmitting collected data to external servers for processing, which raises concerns regarding personal information protection and presents difficulties in real-time processing.

[0008] Therefore, there is a growing need for an on-device solution that utilizes artificial intelligence technology based on single-wavelength PPG signals to generate multi-wavelength data and thereby monitor cardiovascular biomarkers in real time.

[0009]

[0010] Accordingly, the present invention has one objective of providing an on-device artificial intelligence-based cardiovascular risk measurement device capable of generating multi-wavelength data using a single wavelength photoplethysmography (particularly a green wavelength) and monitoring cardiovascular biomarkers in real time based thereon.

[0011] In addition, the present invention has another objective of enabling movement-robust biosignal measurement in a wearable device without equipping it with LEDs and sensors of various wavelengths, thereby providing accurate monitoring of the body condition even with user movement.

[0012]

[0013] To achieve the above objective, the present invention comprises, in one aspect, a device for measuring cardiovascular risk based on on-device artificial intelligence according to skin reflection waves, a collection unit for collecting photoplethysmography (ppg) of a specific wavelength from a wearable device; a storage unit for storing the collected wavelength and storing prior data; a wavelength generation unit for generating one or more multiple wavelengths by inputting the collected photoplethysmography into a first artificial intelligence model; a detection unit for extracting biomarker values ​​by inputting the generated multiple wavelengths and the prior data into a second artificial intelligence model; and a diagnosis unit for providing a diagnosis based on the detected biomarker values.

[0014] Preferably, the collecting unit may be characterized by collecting green LED reflected waves in the 510 to 540 nm wavelength band using the photoplethysmography of the specific wavelength.

[0015] Preferably, the storage unit may be characterized by storing the photoplethysmography of another person at the same point in time as the prior data.

[0016] Preferably, the wavelength generating unit may be characterized by generating at least one wavelength among an infrared LED reflected wave in the 850 to 950 nm wavelength range, a red LED reflected wave in the 620 to 750 nm wavelength range, a yellow LED reflected wave in the 570 to 590 nm wavelength range, a green LED reflected wave in the 510 to 540 nm wavelength range, a blue LED reflected wave in the 450 to 495 nm wavelength range, a purple LED reflected wave in the 380 to 450 nm wavelength range, and an ultraviolet LED reflected wave in the 100 to 400 nm wavelength range.

[0017] Preferably, the first artificial intelligence model may be characterized by performing a forward process that adds noise to the photoplethysmography of the specific wavelength, and a reverse process that restores the photoplethysmography of the specific wavelength to which noise was added during the forward process.

[0018] Preferably, the first artificial intelligence model may be characterized by performing a back-diffusion process using a photoplethysmography of a specific wavelength with added noise and a machine learning-based artificial intelligence model trained on the specific wavelength.

[0019] Preferably, the detection unit may be characterized by extracting the value of at least one biomarker among Troponin T / I, CK-MB (Creatine Kinase-MB Isoenzyme) / BB, Myoglobin, and NT-Pro-BNP (Brain Natriuretic Peptide).

[0020] Preferably, the detection unit may be characterized by defining the collected wavelength as an anchor, the prior data as a negative, and the generated multiple wavelengths as a positive, and inputting them into the second artificial intelligence model.

[0021] Preferably, the second artificial intelligence model may be a contrastive learning model that analyzes the similarity of input data, and may be characterized by extracting first features of the anchor, the negative, and the positive based on a learned CNN, extracting local features and global features of the second feature based on multiple descriptors that extract local features and global features, and generating a third feature by combining local features and global features to extract the value of the biomarker.

[0022] In addition, the present invention is further characterized by a method for measuring cardiovascular risk based on on-device artificial intelligence according to skin reflection waves, comprising: a collection step of collecting photoplethysmography (ppg) of a specific wavelength from a wearable device; a storage step of storing the collected wavelength and storing prior data; a wavelength generation step of inputting the collected photoplethysmography into a first artificial intelligence model to generate one or more multiple wavelengths; a detection step of inputting the generated multiple wavelengths and the prior data into a second artificial intelligence model to extract the value of a biomarker; and a diagnosis step of providing a diagnosis based on the value of the detected biomarker.

[0023]

[0024] According to the present invention, by generating multi-wavelength data using only green photoplethysmography signals, it is possible to monitor cardiovascular biomarkers in real time, which has the advantage of reducing hardware complexity of wearable devices and improving measurement accuracy.

[0025] In addition, the present invention has the advantage of enabling real-time data processing by running an artificial intelligence model on a device and providing robust biosignal measurement even with user movement.

[0026]

[0027] Figure 1 shows a configuration diagram of an artificial intelligence-based cardiovascular risk measurement device according to an embodiment of the present invention.

[0028] FIG. 2 shows a structural diagram of a first artificial intelligence model according to an embodiment of the present invention.

[0029] Figure 3 shows a structural diagram of a second artificial intelligence model according to an embodiment of the present invention.

[0030] Figure 4 shows a flowchart of an artificial intelligence-based cardiovascular risk measurement device according to an embodiment of the present invention.

[0031]

[0032] An AI-based cardiovascular risk measuring device based on on-device AI according to skin reflection waves, comprising: a collecting unit that collects photoplethysmography (ppg) of a specific wavelength from a wearable device; a storage unit that stores the collected wavelength and stores prior data; a wavelength generating unit that inputs the collected photoplethysmography into a first AI model to generate one or more multiple wavelengths; a detection unit that inputs the generated multiple wavelengths and the prior data into a second AI model to extract the value of a biomarker; and a diagnosis unit that provides a diagnosis based on the value of the detected biomarker.

[0033]

[0034] The present invention will be described in detail below with reference to the contents described in the attached drawings. However, the present invention is not limited or restricted by exemplary embodiments. Identical reference numerals in each drawing indicate components that perform substantially the same function.

[0035] The purpose and effects of the present invention may be naturally understood or become clearer through the following description, and the purpose and effects of the present invention are not limited solely to the description below. Furthermore, in describing the present invention, if it is determined that a detailed description of known technology related to the present invention may unnecessarily obscure the essence of the present invention, such detailed description will be omitted.

[0036] The terms used in this invention are used merely to describe specific embodiments and are not intended to limit the invention. The singular expression includes the plural expression unless the context clearly indicates otherwise. In this application, terms such as "comprising" or "having" are intended to specify the presence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the description of the invention, 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.

[0037] 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.

[0038] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art to which this invention pertains. 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 should not be interpreted in an ideal or overly formal sense unless explicitly defined in this invention.

[0039] In interpreting the components, they are interpreted to include a margin of error even without a separate explicit indication. In the case of descriptions regarding temporal relationships, for example, where the temporal sequence is described using 'after,' 'following,' 'next,' 'before,' etc., cases that are not continuous are included unless 'immediately' or 'directly' is used.

[0040] Hereinafter, the technical configuration of the present invention will be described in detail with reference to the attached drawings.

[0041] FIG. 1 shows a configuration diagram of an artificial intelligence-based cardiovascular risk measurement device (100) according to an embodiment of the present invention. Referring to FIG. 1, the artificial intelligence-based cardiovascular risk measurement device (100) may include a collection unit (110), a storage unit (130), a wavelength generation unit (150), a detection unit (170), and a diagnosis unit (190).

[0042] Specifically, in the AI-based cardiovascular risk measurement device (100), the collection unit (110) collects photoplethysmography (ppg) of a specific wavelength from a wearable device and sends it to the storage unit (130), stores the collected wavelength, and can store prior data. In the AI-based cardiovascular risk measurement device (100), the wavelength generation unit (150) can input the collected photoplethysmography into a first AI model to generate one or more multiple wavelengths. The detection unit (170) can input the generated multiple wavelengths and the prior data into a second AI model to extract the values ​​of biomarkers. Additionally, the detection unit (170) can provide a diagnosis based on the values ​​of biomarkers detected by the diagnosis unit (190).

[0043] In one embodiment, the artificial intelligence-based cardiovascular risk measurement device (100) may be a wearable device or a smartwatch. In addition, it may be applied to the field of biosignal measurement, and in particular, it may be used to collect physiological data through the skin, such as blood oxygen saturation (SpO2) analysis and skin condition monitoring in digital healthcare wearable devices.

[0044] In another embodiment, the artificial intelligence-based cardiovascular risk measurement device (100) can be applied to material development (efficient light-transmitting material), component design (optimized LED and light detection sensor), final product (smartwatch, healthcare monitoring device), and system integration (real-time health monitoring system) using body penetration measurement technology utilizing an intermediate wavelength (green LED).

[0045] In another embodiment, the artificial intelligence-based cardiovascular risk measurement device (100) can also be used for real-time cardiovascular monitoring technology based on the protein detection technology of capillaries through multi-optical monitoring technology.

[0046] The collection unit (110) can collect photoplethysmography (ppg) of a specific wavelength from a wearable device, and the photoplethysmography of a specific wavelength may include green LED reflected waves in a wavelength range of 510 to 540 nm. In one embodiment, the wearable device may include at least one of an infrared LED in a wavelength range of 850 to 950 nm, a red LED in a wavelength range of 620 to 750 nm, a green LED in a wavelength range of 510 to 540 nm, and a blue LED in a wavelength range of 450 to 495 nm.

[0047] In one embodiment, the collection unit (110) can collect a PPG signal of a specific wavelength from a wearable device. The collection unit (110) can collect green LED reflected waves in the wavelength band of 510 nm to 540 nm to receive a stable signal that is robust to movement. The collection unit (110) can collect at a sampling rate of 25 Hz for 12 seconds using the green wavelength input value provided in the batching mode of the Galaxy Watch.

[0048] The storage unit (130) can store collected green wavelength PPG signals and prior data. The prior data includes PPG signals collected from others at the same time, which can be used for training and performance improvement of an artificial intelligence model, and can be used as data for wavelength generation and detection.

[0049] The wavelength generation unit (150) can input the collected PPG signal into the first artificial intelligence model to generate one or more multi-wavelength PPG signals.

[0050] The wavelength generating unit (150) can generate at least one wavelength among an infrared LED reflected wave in the 850 to 950 nm wavelength range, a red LED reflected wave in the 620 to 750 nm wavelength range, a yellow LED reflected wave in the 570 to 590 nm wavelength range, a green LED reflected wave in the 510 to 540 nm wavelength range, a blue LED reflected wave in the 450 to 495 nm wavelength range, a purple LED reflected wave in the 380 to 450 nm wavelength range, and an ultraviolet LED reflected wave in the 100 to 400 nm wavelength range. The meanings of the generated wavelengths are as shown in Table 1 below.

[0051] LED Wavelength Band | Skin Penetration | Characteristic Output Value Infrared 850-950nm: Deep skin layers and muscles, measures deep blood flow or oxygen saturation Generated Signal Red 620-750nm: Subcutaneous tissue, measures blood oxygen levels, heart rate, etc. Generated Signal Yellow 570-590nm: Part of subcutaneous tissue Generated Signal Green 510-540nm: Dermis Primary Signal Blue 450-495nm: Part of dermis layer, monitors skin surface condition or cellular activity Generated Signal Purple 380-450nm: Epidermis Generated Signal Ultraviolet 100-400nm: Part of epidermal layer Generated Signal

[0052] Infrared LEDs penetrate deep skin layers and muscles, and can measure deep blood flow or oxygen saturation. Red LEDs penetrate subcutaneous tissue and can measure blood oxygen levels and heart rate. Yellow LEDs penetrate part of subcutaneous tissue and can measure blood oxygen levels and heart rate. Green LEDs penetrate the dermis and can measure blood oxygen levels and heart rate. Blue LEDs penetrate part of the dermis and can monitor the condition of the skin surface or cellular activity. Purple LEDs penetrate the epidermis and can monitor the condition of the skin surface or cellular activity. Ultraviolet LEDs penetrate the epidermis and can monitor the condition of the skin surface or cellular activity. The generated wavelengths can be multiple wavelengths; the wavelength bands reflect the various depths and characteristics of the skin, and correlations with biomarkers may exist depending on how each wavelength band detects optical absorption and reflection characteristics according to protein properties.

[0053] FIG. 2 shows a structural diagram of a first artificial intelligence model according to an embodiment of the present invention. Referring to FIG. 2, the first artificial intelligence model can perform a forward process of adding noise to the photoplethysmography of the specific wavelength and a reverse process of restoring the photoplethysmography of the specific wavelength to which noise has been added during the forward process. The first artificial intelligence model can perform the reverse process using the photoplethysmography of the specific wavelength to which noise has been added and a machine learning-based artificial intelligence model trained on the specific wavelength.

[0054] Specifically, referring to FIG. 2, the first artificial intelligence model is characterized by generating a multi-wavelength signal based on a green PPG signal. In the process of training the first artificial intelligence model, a forward process and a reverse process may be included, and it may be a diffusion model.

[0055] In one embodiment, during the process of training a first artificial intelligence model, the forward diffusion process is the initial input data, the green PPG signal ( time step( By gradually adding Gaussian noise according to ), the noisy data ( It generates ), and the amount of noise is covariance ( It is controlled by ), and the forward diffusion probability distribution is expressed by the following mathematical formula 1. In the step of the forward diffusion process, it can be converted into almost pure noise.

[0056]

[0057] In the process of training the first artificial intelligence model, the inverse diffusion process is noisy data ( It is possible to reconstruct the original data or generate new data by reversing the process. A neural network trained as a method to remove noise from noisy data ( It can be trained to remove noise through ), and can return to the previous step using a Gaussian distribution. The inverse diffusion probability distribution is expressed by the following Equation 2.

[0058]

[0059] Each wavelength generated by the wavelength generation unit (150) can be generated into six wavelengths through a diffusion model from a green PPG signal. The generated wavelengths are output as waveforms in a total of six wavelength regions, including infrared, red, yellow, blue, violet, and ultraviolet, at the same 25Hz sampling rate, so that a 6X25 array shape can be generated.

[0060] The detection unit (170) can extract the value of an intravascular biomarker using the generated multi-wavelength signal and the prior data stored in the storage unit (130). The detection unit (170) can detect the biomarker through a second artificial intelligence model. The detection unit (170) can extract the value of the biomarker by inputting the generated multi-wavelength and prior data into the second artificial intelligence model. The detection unit (170) can extract the value of at least one biomarker among Troponin T / I, CK-MB (Creatine Kinase-MB Isoenzyme) / BB, Myoglobin, and NT-Pro-BNP (Brain Natriuretic Peptide).

[0061] The detection unit (170) inputs the generated multi-wavelength signal and the initial data stored in the storage unit into the second artificial intelligence model. At this time, the initial data consists of the PPG signal of another person at the same time point and is used for the model's learning. The collected wavelength is defined as the anchor, the prior data as the negative, and the generated multi-wavelength as the positive, and inputs them into the second artificial intelligence model.

[0062] FIG. 3 shows a structural diagram of a second artificial intelligence model according to an embodiment of the present invention. Referring to FIG. 3, the second artificial intelligence model is a model that learns using three inputs, and the input data may consist of an anchor, a positive, and a negative. The second artificial intelligence model can extract unique features by integrating various wavelengths. Based on this, changes in cardiovascular biomarkers can be detected by analyzing features obtained from each wavelength through a method of comparing data. The learning method of the second artificial intelligence model learns change patterns by comparing existing data with new data, and utilizes an anchor for this purpose. An anchor refers to a unique feature value derived from existing data, and the model is trained by comparing the feature values ​​of other data (positive and negative) through this.

[0063] The input to the second artificial intelligence model consists of a wavelength collected from the collection unit as an anchor, prior data from the storage unit as a negative, and multiple wavelengths generated from the wavelength generation unit (150) as a positive.

[0064] In one embodiment, a contrastive learning model that analyzes the similarity of input data may be characterized by extracting first features of the anchor, the negative, and the positive based on a trained CNN, extracting local features and global features of the second feature based on multiple descriptors that extract local features and global features, and generating a third feature by combining local features and global features to extract the value of the biomarker.

[0065] Specifically, the second artificial intelligence model can start learning by using data input from the wavelength collection unit, storage unit, and generation unit, using the anchor as the collected green wavelength PPG reflected wave, the negative as the other person's PPG signal at the same time point, and the positive as the generated multi-wavelength signal as inputs to the model.

[0066] First, the second AI model can extract anchor, negative, and positive features, respectively, using an AI model, for example, a CNN model. These extracted features can be the first features.

[0067] Next, the second AI model can extract second features from the first feature in two ways using multiple descriptors. The multiple descriptors may include descriptors that extract local and global features. The second AI model can be trained by concatenating the descriptors after applying the Orthogonal Fusion Module (OFM) to the local and global features extracted using the multiple descriptors.

[0068] In one embodiment, the second artificial intelligence model may be a CGD (Combination of Multiple Global Descriptors), and the second artificial intelligence model inputs seven wavelengths generated by the wavelength generation unit (150) in 12-second intervals, and the data may be input in the form of 7X300. In addition, specific individual numerical values ​​of biomarkers such as Troponin T / I, CK-MB / BB, Myoglobin, and NT-Pro-BNP (a total of six types, with T / I and MB / BB being separate biomarkers) can be output. The description of the biomarkers output here is as shown in Table 2 below.

[0069] Biomarker Description Troponin T / I: Biomarker for predicting myocardial infarction by evaluating cardiovascular damage CK-MB / BB: Biomarker for predicting myocardial infarction by evaluating myocardial damage Myoglobin: Biomarker for predicting myocardial infarction by evaluating myocardial damage NT-Pro-BNP: Biomarker for evaluating symptoms of heart failure

[0070] The biomarker value extracted from the detection unit (170) is transmitted to the diagnosis unit (190) to provide a diagnosis based on the detected biomarker value.

[0071] The diagnostic unit (190) diagnoses a total of six biomarkers, including Troponin T / I, CK-MB / BB, Myoglobin, and NT-Pro-BNP, and can estimate the user's condition based on the amount of each biomarker. Troponin T / I is a risk indicator for myocardial infarction, and if the amount of the biomarker is 0.04 ng / mL or higher, it is considered myocardial damage. CK-MB / BB is an indicator of muscle damage (especially heart muscle damage), and if the amount of the biomarker is 5 ng / mL or higher, it is considered a suspected myocardial damage category; if it is 6-10 ng / mL or higher, it is considered myocardial damage; and if it increases by more than 50% within 3 hours, it is also considered a risk. Myoglobin is a rapid indicator for detecting muscle damage or myocardial damage; if the amount of the biomarker is 85 ng / mL or higher, it is considered a suspected muscle damage or myocardial damage category, and if it is 100 ng / mL or higher, it is considered that immediate additional testing is required. NT-Pro-BNP is an important indicator of heart failure, and a biomarker level of 125 pg / mL or higher is considered a warning sign of heart failure.

[0072] FIG. 4 shows a flowchart of an artificial intelligence-based cardiovascular risk measurement device according to an embodiment of the present invention. Referring to FIG. 4, the artificial intelligence-based cardiovascular risk measurement device (100) consists of four main steps. First, green PPG data is collected from a wearable device. This represents blood flow patterns and biosignals and begins with data collected at regular intervals for 12 seconds. Second, the collected PPG data is processed by a multi-wavelength generation AI model to generate reflected waves for various wavelength regions. These reflected waves reflect physiological characteristics according to the skin layer and blood vessel density and provide additional information related to cardiovascular health. Third, cardiovascular and cerebrovascular related biomarkers are detected based on the generated multi-wavelength data. The detected biomarkers include Troponin T / I, CK-MB / BB, Myoglobin, NT-Pro-BNP, etc., and each biomarker is used as an important indicator to evaluate the state of diseases such as myocardial infarction, heart failure, and stroke. Fourth, based on the detected biomarker data, cardiovascular risk, myocardial risk, heart failure risk, etc., can be calculated to evaluate the user's health status in real time and provide notifications regarding this.

[0073] A method for measuring cardiovascular risk based on on-device artificial intelligence according to skin reflection waves may include a collection step, a storage step, a wavelength generation step, a detection step, and a diagnosis step.

[0074] The collection step can collect photoplethysmography (ppg) of a specific wavelength from a wearable device. The collection step refers to an operation performed on the aforementioned collection unit (110).

[0075] The storage step stores the collected wavelengths, and prior data may be stored. The storage step refers to the operation performed in the aforementioned storage unit (130).

[0076] The wavelength generation step can generate one or more multiple wavelengths by inputting the collected photoplethysmography into the first artificial intelligence model. The wavelength generation step refers to the operation performed in the aforementioned wavelength generation unit (150).

[0077] The detection step can extract the value of the biomarker by inputting the generated multiple wavelengths and the prior data into the second artificial intelligence model. The detection step refers to the operation performed in the aforementioned detection unit (170).

[0078] The diagnostic step can provide a diagnosis based on the value of the detected biomarker. The diagnostic step refers to an operation performed in the diagnostic unit (190).

[0079] Although the present invention has been described in detail above through representative embodiments, those skilled in the art will understand that various modifications can be made to the above-described embodiments within the scope of the present invention. Therefore, the scope of the present invention should not be limited to the described embodiments, but should be determined by the claims set forth below as well as all modifications or variations derived from the claims and equivalent concepts.

[0080]

[0081] Accordingly, the present invention has one objective of providing an on-device artificial intelligence-based cardiovascular risk measurement device capable of generating multi-wavelength data using a single wavelength photoplethysmography (particularly a green wavelength) and monitoring cardiovascular biomarkers in real time based thereon.

[0082] In addition, the present invention has another objective of enabling movement-robust biosignal measurement in a wearable device without equipping it with LEDs and sensors of various wavelengths, thereby providing accurate monitoring of the body condition even with user movement.

Claims

1. A device for measuring cardiovascular risk based on on-device artificial intelligence according to skin reflection waves, A collecting unit that collects photoplethysmography (ppg) of a specific wavelength from a wearable device; A storage unit that stores collected wavelengths and stores prior data; A wavelength generation unit that inputs collected photoplethysmography into a first artificial intelligence model to generate one or more multiple wavelengths; A detection unit that inputs the generated multiple wavelengths and the prior data into a second artificial intelligence model to extract the value of the biomarker; and A diagnostic unit that provides a diagnosis based on the numerical value of a detected biomarker; An artificial intelligence-based cardiovascular risk measurement device including 2. In Paragraph 1, The above collection unit is, An artificial intelligence-based cardiovascular risk measurement device characterized by collecting green LED reflected waves in the 510 to 540 nm wavelength band using the above-mentioned specific wavelength photoplethysmography.

3. In Paragraph 1, The above storage unit is, An artificial intelligence-based cardiovascular risk measurement device characterized by storing photoplethysmography of another person at the same time point as the above prior data.

4. In Paragraph 1, The above wavelength generating unit is, An artificial intelligence-based cardiovascular risk measurement device characterized by generating at least one wavelength among an infrared LED reflected wave in the 850 to 950 nm wavelength range, a red LED reflected wave in the 620 to 750 nm wavelength range, a yellow LED reflected wave in the 570 to 590 nm wavelength range, a green LED reflected wave in the 510 to 540 nm wavelength range, a blue LED reflected wave in the 450 to 495 nm wavelength range, a purple LED reflected wave in the 380 to 450 nm wavelength range, and an ultraviolet LED reflected wave in the 100 to 400 nm wavelength range.

5. In Paragraph 1, The above-mentioned first artificial intelligence model is, An artificial intelligence-based cardiovascular risk measurement device characterized by performing a forward process to add noise to a photoplethysmogram of a specific wavelength and a reverse process to restore the photoplethysmogram of a specific wavelength to which noise has been added during the forward process.

6. In Paragraph 5, The above-mentioned first artificial intelligence model is, An AI-based cardiovascular risk measurement device characterized by performing a reverse diffusion process using a photoplethysmography of a specific wavelength with added noise and a machine learning-based AI model trained on the specific wavelength.

7. In Paragraph 1, The above detection unit is, An artificial intelligence-based cardiovascular risk measurement device characterized by extracting the values ​​of at least one biomarker among Troponin T / I, CK-MB (Creatine Kinase-MB Isoenzyme) / BB, Myoglobin, and NT-Pro-BNP (Brain Natriuretic Peptide).

8. In Paragraph 1, The above detection unit is, An AI-based cardiovascular risk measurement device characterized by defining the collected wavelength as an anchor, the prior data as a negative, and the generated multiple wavelengths as a positive, and inputting them into the second AI model.

9. In Paragraph 8, The above second artificial intelligence model is, A contrastive learning model that analyzes the similarity of input data, extracts first features of the anchor, the negative, and the positive based on a trained CNN, and An artificial intelligence-based cardiovascular risk measurement device characterized by extracting the local and global features of the second feature based on multiple descriptors that extract local and global features, and generating a third feature by combining the local and global features to extract the value of a biomarker.

10. A method for measuring cardiovascular risk based on on-device artificial intelligence according to skin reflection waves, A collection step for collecting photoplethysmography (ppg) of a specific wavelength from a wearable device; A storage step that stores collected wavelengths and stores prior data; A wavelength generation step of inputting collected photoplethysmography into a first artificial intelligence model to generate one or more multiple wavelengths; A detection step of inputting the generated multiple wavelengths and the prior data into a second artificial intelligence model to extract the value of the biomarker; and A diagnostic step that provides a diagnosis based on the values ​​of detected biomarkers; A method including

Images