Artificial intelligence model-based automated flap state evaluation method and analysis device
An AI-driven image analysis system automates flap condition monitoring, addressing the time-consuming manual evaluation of flap surgery outcomes and enhancing surgical success rates by providing timely intervention insights.
Patent Information
- Application Number
- PCT/KR2024/006480
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2024-05-13
- Filing Date
- 2024-05-13
- Publication Date
- 2024-11-28
AI Technical Summary
Current methods for monitoring the condition of flaps post-surgery require extensive manual effort and time, leading to a potential decrease in the success rate of flap surgery if timely intervention is not made, especially when blood supply issues arise.
An automated image-based monitoring system using an AI model that classifies and predicts the condition of flaps through image analysis, reducing the need for continuous manual monitoring by medical personnel.
The system quickly provides information on the need for reoperation, thereby increasing the success rate of flap surgery by enabling timely interventions and reducing the workload on medical staff.
Smart Images

Figure KR2024006480_28112024_PF_FP_ABST
Abstract
Description
Automated evaluation method and analysis device for the condition of the cut based on an artificial intelligence model
[0001] The technique described below relates to a technique for evaluating the condition of a flap after flap surgery.
[0002] A flap is a surgical procedure that grafts a flap from another area to a skin tissue defect caused by skin cancer excision, trauma, or infection. Free flaps, in particular, involve connecting blood vessels and nerves with the skin tissue. Therefore, monitoring the blood supply of the grafted flap is crucial. If the blood supply to the flap becomes compromised, a reoperation may be necessary. Failure to perform a reoperation within this golden timeframe drastically reduces the success rate of the flap surgery.
[0003] Currently, medical staff continuously monitor the condition and progress of the flap for a considerable period of time. Therefore, significant time and manpower are spent on post-surgical care.
[0004] The technology described below aims to provide an image-based monitoring system for flaps transplanted using flap surgery. The technology described below aims to provide a system for assessing the current status of the flap. Furthermore, the technology described below aims to provide a system for predicting the future status of the flap.
[0005] A method for evaluating a flap condition using an image of a flap area includes a step of an analysis device receiving a source image including a flap area of a subject, a step of the analysis device distinguishing a flap area from the source image, and a step of the analysis device inputting an image of the distinguished flap area into a classification model to produce information indicating the condition of the flap area.
[0006] An analysis device for evaluating a flap condition includes an interface device that receives a source image including a flap area of a subject, a storage device that stores a segmentation model for distinguishing an area of interest and a classification model for classifying a flap condition by receiving an image of the flap area, and a calculation device that uses the segmentation model to distinguish the flap area from the source image and inputs the image of the distinguished flap area into the classification model to calculate the condition of the flap area.
[0007] The technology described below can automatically monitor the status of flaps transplanted using image-based techniques. Therefore, the technology described below can reduce the medical staff required to monitor flap status.
[0008] The techniques described below can quickly provide information on the need for revision flap surgery. In other words, the techniques described below contribute to increasing the success rate of flap surgery.
[0009] Figure 1 is an example of a plate condition monitoring system.
[0010] Figure 2 is an example of a process for evaluating the state of a plate.
[0011] Figure 3 is an example of the process of building a data set.
[0012] Figure 4 is an example of the learning process of a learning model for evaluating the state of the vertex.
[0013] Figure 5 is an example of an analysis device for evaluating or predicting a plate condition.
[0014] The technology described below is susceptible to various modifications and embodiments. Specific embodiments are illustrated and described in detail in the drawings. However, this does not limit the technology described below to specific embodiments, and it should be understood that all modifications, equivalents, and alternatives fall within the spirit and scope of the technology described below.
[0015] Terms such as first, second, A, and B may be used to describe various components, but these components are not limited by these terms and are used solely to distinguish one component from another. For example, without departing from the scope of the technology described below, a first component may be referred to as a second component, and similarly, a second component may also be referred to as a first component. The term "and / or" includes a combination of a plurality of related described items or any of a plurality of related described items.
[0016] As used herein, the singular expressions should be understood to include the plural expressions unless the context clearly dictates otherwise, and the term "comprises" and the like should be understood to mean the presence of a described feature, number, step, operation, component, part, or combination thereof, but not to exclude the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.
[0017] Before going into a detailed description of the drawings, it should be made clear that the division of components in this specification is merely a division based on the main function of each component. In other words, two or more components described below may be combined into a single component, or a single component may be further subdivided into two or more components with more detailed functions. In addition to its own main function, each component described below may additionally perform some or all of the functions of other components, and of course, some of the main functions of each component may be exclusively performed by other components.
[0018] Additionally, in performing a method or method of operation, each process constituting the method may occur in a different order than the stated order, unless the context clearly indicates a specific order. That is, each process may occur in the same order as the stated order, may be performed substantially simultaneously, or may be performed in the opposite order.
[0019] The technique described below is a technique for evaluating the flap condition using images of the patient's flap area.
[0020] The source image below refers to an image containing the flap area where a free flap procedure was performed. The flap area refers to the flap area where the original skin tissue was replaced by a free flap procedure. The source image refers to an image such as a photograph or a video frame.
[0021] The following describes how the analysis device uses images containing the lesion area to evaluate or predict the lesion condition. The analysis device can be implemented using various devices capable of data processing. For example, the analysis device can be implemented using a PC, a server on a network, a smart device, or a chipset embedded with a dedicated program.
[0022] The analysis device can use a learning model to identify a region of interest in the source image. The region of interest refers to the flap area. Furthermore, the analysis device can use the learning model to classify or evaluate the flap status.
[0023] The training data construction and training model construction processes can be performed on a separate training device. A training device refers to a computer capable of controlling specific image processing and model training processes. A training device can be implemented in the form of a PC, a networked server, a smart device, or a chipset embedded with a dedicated program.
[0024] Figure 1 is an example of a plate condition monitoring system (100).
[0025] Fig. 1(A) is an example of a system (100) composed of multiple devices. In Fig. 1, the analysis devices are illustrated as a computer terminal (130) and a server (140).
[0026] The camera (110) or image sensor acquires a source image including the flap area of the subject (patient). The source image may generally include the flap area and the surrounding skin area. The source image may also include patient clothing, medical equipment, gauze for protecting the affected area, a background, etc.
[0027] The camera (110) can acquire source images continuously or at regular intervals.
[0028] The camera (110) may be a device built into a portable device such as a smartphone.
[0029] The camera (110) can store the acquired image of the lesion area in an EMR (Electronic Medical Record, 120) or a separate database (DB) via a wired or wireless network.
[0030] User (A) may be a medical professional monitoring a patient's flap condition. User (A) may analyze source images using a computer terminal (130). The computer terminal (130) may receive source images from a camera (110) or EMR (120) via a wired or wireless network.
[0031] In some cases, the computer terminal (130) may be a device physically connected to the camera (110). The computer terminal (130) may preprocess the source image in a predetermined manner. The computer terminal (130) may distinguish a flap region, which is an area of interest, from the source image. The computer terminal (130) may analyze the distinguished flap region using a learning model. The computer terminal (130) may classify the flap state based on the value output by the learning model. In some cases, the computer terminal (130) may also predict the future flap state based on the value output by the learning model. User A may check the analysis results on the computer terminal (130).
[0032] The server (140) can receive a flap area image from the camera (110) or EMR (120). The server (140) can preprocess the source image in a predetermined manner. The server (140) can distinguish the flap area, which is a region of interest, from the source image. The server (140) can analyze the distinguished flap area using a learning model. The server (140) can classify the flap status based on the value output by the learning model. In some cases, the server (140) can also predict the future flap status based on the value output by the learning model. The server (140) can transmit the analysis result to the terminal of user A. Alternatively, if the server (140) determines that the flap status is abnormal (e.g., re-surgery is required), it can alarm the medical staff terminal (50) about the flap status.
[0033] The computer terminal (130) and / or server (140) may also store the analysis results in the EMR (120).
[0034] Figure 1(B) is an example of a judgment status monitoring system implemented in the form of a user terminal (150) such as a smartphone. A user (A) can check the judgment status using the user terminal (150).
[0035] The user terminal (150) can capture a source image including a target's flap area using a built-in camera. The user terminal (150) can analyze the source image using an installed application. The user terminal (150) can preprocess the source image in a predetermined manner. The user terminal (150) can distinguish the flap area from the source image. The user terminal (150) can analyze the distinguished flap area using a learning model. The user terminal (150) can classify the flap state based on the value output by the learning model. In some cases, the user terminal (150) can also predict the future flap state based on the value output by the learning model. The user terminal (150) can output information about the flap state.
[0036] Figure 2 is an example of a trial state evaluation process (200).
[0037] The analysis device receives a source image containing a cut region (210). The analysis device can preprocess the input cut region image in a predetermined manner. The preprocessing may include noise removal, image resizing, and image cropping.
[0038] The analysis device can segment the cut region from the source image (220). The analysis device can segment the cut region from the source image using a specific image processing technique. Alternatively, the analysis device can segment the cut region from the source image using a learned segmentation model. For example, the segmentation model can be any of a U-net-based model, a Fully Convolutional Network (FCN), etc.
[0039] The analysis device analyzes the extracted flap area using a learning model (230). The model that analyzes the status of the flap area can be referred to as a classification model. The classification model can utilize any of various types of models.
[0040] The classification model can be a regression model that analyzes trends in the color values of the image. The color of the skin changes depending on the blood flow status of the flap. If blood flow within the flap is poor, blood clots accumulate within the blood vessels, causing the skin to turn red. Therefore, regression analysis can classify the condition of the flap based on the currently input image by analyzing the correlation between the proportion of red areas in the skin and the condition of the flap.
[0041] Alternatively, the classification model can be a deep learning model that interprets images. For example, the classification model can be based on a convolutional neural network (CNN). A trained CNN can extract features from the input image of the flap region and output a probability value for the flap state based on these extracted features. Of course, the classification model can be implemented not only using a CNN-based model, but also using any of a variety of deep learning models, such as a vision transformer.
[0042] The analysis device can evaluate the current flap state of the subject of analysis based on the flap state classification result or flap state probability value of the classification model (240). Alternatively, the analysis device can predict the future flap state based on the flap state classification result or flap state probability value (240). The flap state prediction is to predict the flap state after a certain point in time from the time when the flap area image is captured. During the learning process, the learning model can infer the current flap state or future flap state based on the input data and the correct answer value (label value).
[0043] The researcher describes the model-building process. The researcher collected a data set from patients who underwent free flap surgery at their affiliated medical institution from January 2020 to August 2023. The data set includes retrospective and prospective data. The retrospective data were collected from patients who underwent free flap surgery from March 2020 to February 2022. The retrospective data were collected by capturing 15 to 20 photos for each patient using a smartphone. The retrospective data consisted of photos without normalization for image size, distance, and lighting conditions. The prospective data were collected for data normalization. The prospective data were additionally collected from patients who underwent free flap surgery from March 2022 to August 2023. The prospective data were collected using the same smartphone under constant lighting conditions and at a constant distance.
[0044] Table 1 below presents demographic and flap information for the population used to construct the dataset. Photographs of the flap area were collected from a total of 305 patients. Prospective data were collected from 203 patients, and retrospective data were collected from 102 patients.
[0045]
[0046] The body parts where free flaps were performed were classified into the head and neck, trunk, and extremities. The specific flap sites included the anterolateral thigh (ALT), thoracodorsal artery perforator (TDAP), latissimus dorsi myocutaneous (LDMC), superficial circumflex iliac artery perforator (SCIP), deep inferior epigastric perforator (DIEP), and dorsal scapular artery perforator (DSAP).
[0047] Figure 3 is an example of a process (300) for constructing a data set. It explains that a learning device performs the process of constructing a data set.
[0048] The learning device receives an initial data set (310). The researcher collected 12,395 images from the 305 patients described above as the initial data set.
[0049] The learning device can remove images that may negatively affect learning from the initial data set (320). The learning device can remove images containing discoloration due to hematomas, etc., and images containing multiple flaps. Through this process, the researcher removed 1,283 images, creating a data set of 11,112 images. This 11,112-image data set consisted of 10,115 normal flaps (91%) and 997 flaps with abnormal perfusion (9%).
[0050] The learning device can build a dataset for a segmentation model from images with clearly visible flap boundaries in the dataset (330). The researcher extracted 2,068 images with clearly visible flap boundaries from among 11,112 images.
[0051] The dataset for the segmentation model contains original images and correct images with the region of interest (criteria) manually identified (manually annotated) for the original images.
[0052] The learning device can build a data set for learning and validating a classification model from the data set.
[0053] The learning device can select initial images for learning a classification model from the dataset. The researcher selected 3,538 images of 43 patients from among 11,112 images as training images. The learning device can use the trained segmentation model to identify regions of interest (flap regions) in the input images. The learning device can use the images with identified regions of interest as a training dataset for the classification model (341). Furthermore, the learning device can augment the training dataset. The learning device can augment the images by flipping the images, adjusting the brightness (delta random adjustment), and adjusting the contrast (342). Furthermore, the learning device matches the correct answers (flap status information) for the images in the training dataset and stores them (342). Therefore, the training dataset consists of flap region images and the correct answers for the corresponding images. The researcher augmented the 3,583 images to create a training dataset of 5,606 images. Additionally, the researchers determined the correct answer by assessing the condition through multiple medical professionals.
[0054] The learning device can select images from the dataset for validation of the classification model. The researcher selected 5,506 images of 258 patients from among 11,112 images as validation images. The learning device can use the trained segmentation model to identify regions of interest (flap regions) in the input images. The learning device can use the images with identified regions of interest as a validation dataset for the classification model (351). Furthermore, the learning device can augment the validation dataset. However, the researcher used the 5,506 images as-is without augmenting the validation data. In addition, the learning device matches the correct answers (flap status information) for the images in the validation dataset and stores them (352). Therefore, the validation dataset consists of flap region images and the correct answers for the corresponding images.
[0055] Figure 4 illustrates an example of a learning process (400) for evaluating a judgment condition. The learning model construction process can be divided into a learning data construction process (410), a segmentation model learning process (420), and a classification model learning process (430).
[0056] The learning device constructs a learning data set (410). The image database (DB) stores source images containing the flap area of the population. The learning device can select source images to be used for learning from the image DB. The learning device can segment the flap area from the source images. The segmentation of the flap area can be performed manually. Alternatively, the learning device can segment the flap area using a learned, separate segmentation model. The learning device can segment the flap area and generate a mask for the flap area. In addition, the learning device can receive status evaluation information for the flap area. The status evaluation can be performed by a medical expert as described above. For example, the flap status can be classified into a normal status (normal, grade 1) and an abnormal status (abnormal, grade 3). The learning device can store the final learning data set in the learning DB. The learning data set can include the entire image, an image or mask of the flap area of the entire image, and status information for the flap area.
[0057] The learning device can perform a process of learning a segmentation model (420). The segmentation model is a model that distinguishes a flap region, which is a region of interest, from a source image. The segmentation model may be any of various structural models. The learning device builds a segmentation model using the entire image (source image) and a flap region mask distinguished from the image in a learning database. The learning device inputs the entire image into the segmentation model. The learning device compares the flap region distinguished by the segmentation model with an answer and updates the learning parameters of the model. Through this process, the segmentation model is trained to distinguish the flap region corresponding to the answer. The answer may be a flap region image or a flap region image with a mask applied to the entire image.
[0058] The learning device can perform a process of learning a classification model (430). The learning device can build a classification model using the segmented image of the cut area and the correct cut state information. The classification model can be any of various structural models.
[0059] The learning device inputs images of the flap region into the classification model. The learning device compares the output of the classification model (flap state classification) with the correct answer and updates the model's learning parameters. Through this process, the classification model learns to output state information about the input flap region images.
[0060] The learning device can repeatedly perform the learning process using the learning data of the population.
[0061] Below, the results of the performance of the model constructed by the researcher are explained.
[0062] Describes the validation results of the segmentation model.
[0063] The researchers built and verified three CNN-based segmentation models. The three segmentation models were U-Net (see O. Ronneberger et al., U-net: Convolutional networks for biomedical image segmentation, in Medical Image Computing and Computer-Assisted Intervention-MICCAI, 2015), VGG16-UNet (see A. A. Pravitasari et al., UNet-VGG16 with transfer learning for MRI-based brain tumor segmentation, TELKOMNIKA Telecommunication, Computing, Electronics and Control, vol. 18, no. 3, pp. 1310-1318, 2020), and FS-Net (Flap Segmentation Network).
[0064] FS-Net is a CNN-based network model developed by researchers for flap segmentation. FS-Net has an encoder-decoder architecture. The encoder receives a 256×256 pixel image and compresses it into a 16×16 representation. The encoder includes a series of 2D (dimension) convolutional layers with batch normalization and ReLU activation functions. The decoder has a structure symmetrical to the encoder, but includes 2D transposed convolutional layers for upsampling. The decoder upsamples the 16×16 representation to generate a 256×256 pixel image. The decoder scales each image pixel value through min-max normalization. FS-Net receives a source image containing the flap region as input and outputs a mask corresponding to the flap region. A learning device and / or an analysis device can segment the flap region using the mask output by the segmentation model.
[0065] The researchers performed a five-fold cross-validation on the developed segmentation model. The performance of the segmentation model was evaluated by comparing the segmentation model's classification results with the results manually classified by an expert (correct answers). Table 2 below presents the results of the segmentation model performance evaluation. Performance was evaluated using Jaccard Similarity and Dice Coefficient. The three segmentation models showed no significant differences in performance, with FS-Net performing slightly better.
[0066]
[0067] Describes the performance verification results of the classification model.
[0068] The researchers built and validated five CNN-based classification models. Four segmentation models used VGG16, ResNet50, InceptionV3, and DenseNet121. Furthermore, the researchers additionally used a custom CNN model. The custom CNN model consists of three convolutional blocks: a dense layer and a softmax layer. The convolutional block includes a 2D convolutional layer with a ReLU activation function and a max pooling layer. The researchers verified the performance of the classification models by comparing the inference results with the correct answers. The performance of the classification models is shown in Table 3 below. All classification models showed high performance without significant differences.
[0069]
[0070] Additionally, the researchers verified the performance of the classification model by differentiating the site and flap type where the glass flap was performed, as shown in Table 4 below. Regardless of the site and type of flap, the classification model demonstrated high performance.
[0071]
[0072] The classification model that verified the performance was a model that binary classified the status of the plate.
[0073] Furthermore, the classification model can be constructed as a model that performs multiple classifications of the papillary condition. The multiple classifications can be three or more grades or scores for the flap condition.
[0074] Additionally, classification models can be built to predict future states of the flap, rather than its current state. In this case, the training data can include the current flap region and the correct values for the future state of that flap region.
[0075] Figure 5 illustrates an example of an analysis device (500) for evaluating or predicting a plate condition. The analysis device (500) corresponds to the aforementioned analysis device (130, 140, or 150 of Figure 1). The analysis device (500) may be physically implemented in various forms. For example, the analysis device (500) may take the form of a computer device such as a PC, a network server, a data processing chipset, or the like.
[0076] The analysis device (500) may include a storage device (510), a memory (520), a computing device (530), an interface device (540), a communication device (550), and an output device (560).
[0077] The storage device (510) can store a source image including a cut area captured from a camera.
[0078] The storage device (510) can store a segmentation model that distinguishes a cut area from a source image. The segmentation model is a model learned through the aforementioned process.
[0079] The storage device (510) can store an image processing algorithm or program that distinguishes a cut area from a source image.
[0080] The storage device (510) can store a cut area image that separates the cut area from the source image.
[0081] The storage device (510) can store a code or program that analyzes the image of the plate area and classifies the plate state.
[0082] The storage device (510) can receive images of the segmented plate area and store a classification model that analyzes the plate state. The classification model is a model learned through the aforementioned process.
[0083] The storage device (510) can store the results of classifying or predicting the state of the plate included in the input source image.
[0084] The memory (520) can store data and information generated during the process of the analysis device analyzing the image of the cut area.
[0085] The interface device (540) is a device that receives certain commands and data from the outside.
[0086] The interface device (540) can receive a source image including a cut area from a physically connected input device or an external storage device.
[0087] The interface device (540) can also transmit the results of classifying the judgment status to an external object.
[0088] The interface device (540) may be a device that transmits data or information received through a communication device (550) to the inside of the analysis device.
[0089] A communication device (550) refers to a configuration that receives and transmits certain information through a wired or wireless network.
[0090] The communication device (550) can receive a source image including a cut area from an external object.
[0091] The communication device (550) may also transmit the result of classifying the judgment status to an external object such as a user terminal.
[0092] The output device (560) is a device that outputs certain information.
[0093] The output device (560) can output interfaces, plate areas, plate status analysis results, etc. required for the data processing process.
[0094] Meanwhile, the analysis device (500) may further include a camera (570). The camera (570) can capture the flap area of a patient who has undergone a glass flap procedure. In this case, the interface device (540) can transmit the source image captured by the camera to the storage device (510).
[0095] The computational device (530) can distinguish a cut area from a source image using an image processing algorithm or program.
[0096] Alternatively, the computational device (530) can input the source image into a segmentation model to segment the cut region. The computational device (530) can segment the cut region in the source image using a mask output by the segmentation model.
[0097] The computational device (530) can analyze the segmented flap region image to classify the flap state. The computational device (530) can input the segmented flap region image into a classification model to classify the flap state.
[0098] Alternatively, the computational device (530) may input a segmented flap region image into a classification model to predict the prognosis of the flap. In this case, the classification model is a model trained to predict the flap status at a future point in time.
[0099] The computing device (530) may be a device such as a processor, AP, or chip embedded with a program that processes data and performs certain operations.
[0100] Additionally, the aforementioned method for processing image data, constructing a learning model, and classifying a state of the skin can be implemented as a program (or application) including an executable algorithm that can be executed on a computer. The program can be stored and provided on a temporary or non-transitory computer-readable medium.
[0101] A non-transitory readable medium refers to a medium that stores data semi-permanently and can be read by a device, rather than a medium that stores data for a short period of time, such as a register, cache, or memory. Specifically, the various applications or programs described above may be stored and provided on a non-transitory readable medium, such as a CD, DVD, hard disk, Blu-ray disk, USB, memory card, ROM (read-only memory), PROM (programmable read only memory), EPROM (Erasable PROM, EPROM), EEPROM (Electrically EPROM), or flash memory.
[0102] Temporarily readable media refers to various types of RAM, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced SDRAM (ESDRAM), Synchronous DRAM (Synclink DRAM, SLDRAM), and Direct Rambus RAM (DRRAM).
[0103] The present embodiment and the drawings attached to the present specification only clearly illustrate a part of the technical idea included in the above-described technology, and it is obvious that all modified examples and specific embodiments that can be easily inferred by a person skilled in the art within the scope of the technical idea included in the specification and drawings of the above-described technology are included in the scope of the rights of the above-described technology.
Claims
1. A step in which an analysis device receives a source image including a target area; The step of the above analysis device distinguishing the cut area from the source image; and A method for evaluating the condition of a flap using an image of a flap region, including a step of inputting an image of the flap region classified by the above analysis device into a classification model and calculating the condition of the flap region.
2. In paragraph 1, The above analysis device is a method for evaluating the condition of a flap using a flap area image that segments the flap area from the source image using a segmentation model.
3. In paragraph 1, The above classification model is a flap condition evaluation method using a flap area image that evaluates the condition of the flap area based on the red area ratio of the flap area.
4. In paragraph 1, A method for evaluating the condition of a flap using an image of a flap region, wherein the classification model is a deep learning model that receives an image of the flap region and classifies the condition of the flap region.
5. In paragraph 1, The above classification model is a deep learning model that receives an image of the above-mentioned flap area and predicts the state of the above-mentioned flap area at a future point in time, and is a method for evaluating the flap state using an image of the flap area.
6. An interface device for receiving a source image including a target's cut area; A storage device storing a segmentation model that distinguishes an area of interest and a classification model that classifies a flap state by receiving an image of a flap area; and An analysis device for evaluating a flap state, including a calculation device that segments the flap area from the source image using the segmentation model, inputs an image of the segmented flap area into the classification model, and calculates the state of the flap area.
7. In paragraph 6, The above classification model is an analysis device for evaluating the condition of the flap area based on the red area ratio of the flap area.
8. In paragraph 6, The above classification model is an analysis device for evaluating the condition of the flap, which is a deep learning model that receives an image of the flap area and classifies the condition of the flap area.
9. In paragraph 6, The above classification model is an analysis device for evaluating the condition of the flap, which is a deep learning model that receives an image of the flap area and predicts the condition of the flap area at a future point in time.
Citation Information
Patent Citations
Flap image acquisition method and device, storage medium and electronic equipment
CN110393508A
3D bio-printer monitoring system and repairing system and method for correction of shape defects
KR1020230030710A
Composite separator, manufacturing method thereof, and electrochemical device comprising same
KR1020230065150A
Glass ceramic and cooktop using same
KR1020240082995A
Target RNA detection method using isothermal nucleic acid amplification without reverse transcription
KR1020240162641A
