System for detecting and analyzing intubation position in human body by using deep learning-based dual-step model

Abstract

The present invention relates to a system for detecting and analyzing the intubation position in a human body, the system comprising: a segmentation deep learning module for receiving an X-ray image and generating an intubation mask regarding the position of an intubation in a human body included in the X-ray image; a connecting module for connecting the intubation mask and the X-ray image to generate a mask-coupled X-ray image; and a classification deep learning module for receiving the mask-coupled X-ray image and classifying the appropriateness of the position of the intubation in the human body.

Description

A system for detecting and analyzing intra-human intubation locations using a deep learning-based two-stage model.

[0001] The present invention relates to a system for detecting and analyzing a position of an intubation site within a human body using a deep learning-based two-stage model.

[0002] Intubation is a medical procedure essential to patient care and survival. It is used for various purposes, including nutrition, medication administration, and fluid removal. Examples of intubation devices include nasogastric tubes, central venous catheters, endotracheal tubes, and thoracic drainage tubes. These various intubation devices must be inserted precisely for their intended purpose, and incorrect placement can lead to serious complications.

[0003] For example, nasogastric tubes are mainly used to provide nutrition or administer medication to patients, and if they are not positioned correctly and are incorrectly inserted into the airway or do not reach the correct location, they can cause serious complications such as pneumonia, pneumothorax, and tracheal perforation.

[0004] Conventional methods for nasogastric tube positioning rely on manual analysis of chest X-rays by medical staff. Errors arising during this process can have a critical impact on patient safety. In particular, factors such as medical staff fatigue, lack of experience, or low-resolution chest X-rays can lead to misdiagnosis, potentially leading to failure to detect incorrect nasogastric tube insertion. The limitations of these manual methods have raised the need for more accurate and automated nasogastric tube positioning technology. An automated system can reduce the burden on medical staff, increase the accuracy of positioning, and significantly improve patient safety.

[0005] Recently, various methods utilizing deep learning have been introduced for nasogastric tube positioning. Attempts have been made to automatically detect the positions of nasogastric tubes, central venous catheters, and endotracheal tubes using neural network architectures such as DenseNet, Inception-v3, and EfficientNet.

[0006] However, these approaches primarily focus on classification, limiting their ability to visually describe or interpret nasogastric tube locations. Classification models typically probabilistically indicate the presence of specific features (e.g., misplacement) but do not provide spatial information about the tube's precise location or path. These limitations can make it difficult for clinicians to clinically interpret and utilize the model's results. For example, classification results alone cannot accurately identify the exact location of a nasogastric tube misplacement, making them insufficient for clinical decision-making.

[0007] Ultimately, there is a need for new methods that can help determine whether nasogastric tubes and other intubations within the body have been properly inserted.

[0008] One object of the present invention is to provide a system that can visually clearly indicate the exact location of an intubation within the human body and automatically determine whether it has been inserted incorrectly, thereby assisting medical staff in making clinical decisions regarding whether intubation within the human body has been performed appropriately.

[0009] Meanwhile, other unspecified purposes of the present invention will be additionally considered within the scope that can be easily inferred from the detailed description and effects thereof below.

[0010] To achieve the purpose proposed above, the following solutions are proposed.

[0011] A system for detecting and analyzing a position of an intubation within a human body according to one embodiment of the present invention includes: a segmentation deep learning module for receiving an X-ray image as input and generating an intubation mask for a position of an intubation within a human body included in the X-ray image; a linking module for connecting the intubation mask and the X-ray image to generate a mask-combined X-ray image; and a classification deep learning module for receiving the mask-combined X-ray image as input and classifying the appropriateness of the position of an intubation within a human body.

[0012] In one embodiment, the intubation may be characterized by at least one selected from the group consisting of a nasogastric tube, a central venous catheter, an endotracheal tube, and a thoracic drainage tube.

[0013] In one embodiment, the intubation mask may be characterized as a mask regarding the shape and position of the line of intubation.

[0014] In one embodiment, the intubation mask may further comprise a mask regarding the shape of the tip of the intubation and the body.

[0015] In one embodiment, the mask-coupled X-ray image may be characterized in that it is generated by stacking the intubation mask and the X-ray image in a channel form.

[0016] In one embodiment, the method further includes a visualization module for visualizing the judgment result of the classification deep learning module, wherein the visualization module generates a heat map indicating a basis area considered important as a basis for judgment by the classification deep learning module, and displays the heat map by overlaying it on the mask-combined X-ray image.

[0017] In one embodiment, the method further includes a verification module for verifying the appropriateness of the judgment result of the classification deep learning module, and the verification module may be characterized in that, if the basis area and the intubation mask do not overlap, the judgment result of the classification deep learning module is output as inappropriate.

[0018] According to one embodiment of the present invention, a system for detecting and analyzing a location of an intubation within a human body first receives an X-ray image as input by a segmentation deep learning module, generates an intubation mask including the location of the intubation, and combines this with the original X-ray image to generate a mask-combined X-ray image. The mask-combined X-ray image is analyzed by a classification deep learning module to determine the appropriateness of the intubation. In other words, by inputting an image that combines spatial information on the location and shape of the intubation rather than simply inputting an X-ray image, the accuracy of location determination is greatly improved compared to a simple classification model.

[0019] In addition, the system for detecting and analyzing the position of intubation in the human body according to one embodiment of the present invention can further improve the reliability of the judgment result by using a verification module to determine whether the basis for judging the appropriateness of intubation is related to intubation.

[0020] Finally, the system for detecting and analyzing the position of an intubation site within the human body according to one embodiment of the present invention can contribute to medical staff making more accurate clinical judgments by visually providing the results determined by the system for detecting and analyzing the position of an intubation site within the human body.

[0021] Meanwhile, even if the effect is not explicitly mentioned herein, it is added that the effect and its provisional effect described in the following specification expected by the technical features of the present invention are treated as described in the specification of the present invention.

[0022] FIG. 1 is a schematic diagram of a system for detecting and analyzing a position of an intubation site within a human body according to one embodiment of the present invention.

[0023] FIG. 2 is a schematic flowchart illustrating a workflow of segmentation, connection, and classification of a system for detecting and analyzing intra-human intubation positions according to one embodiment of the present invention.

[0024] FIG. 3 is a reference diagram for explaining a process of detecting and analyzing an intubation position within a human body from an external verification dataset by a system for detecting and analyzing an intubation position within a human body according to one embodiment of the present invention.

[0025] FIG. 4 is a prototype user interface of a system for detecting and analyzing intra-human intubation position according to one embodiment of the present invention.

[0026] It is to be understood that the attached drawings are provided for reference only to help understand the technical concept of the present invention, and the scope of the present invention is not limited thereby.

[0027] Hereinafter, with reference to the drawings, the configuration of the present invention, guided by various embodiments thereof, and the effects resulting from such configurations will be examined. In describing the present invention, detailed descriptions of related, well-known functions that are obvious to those skilled in the art and that may unnecessarily obscure the gist of the present invention will be omitted.

[0028] The term "module" as used in this document may include a unit implemented in hardware, software, or firmware, and may be used interchangeably with terms such as logic, logic block, component, or circuit. A module may be an integral component, or a minimum unit or part of such a component that performs one or more functions.

[0029] In this document, "modules" or "nodes" use computing devices, such as CPUs and APs, to perform tasks such as moving, storing, and converting data. For example, "modules" or "nodes" may be implemented as devices such as servers, PCs, tablet PCs, and smartphones.

[0030] Hereinafter, the structure, operation, and effect of a system for detecting and analyzing a position of an intubation site in a human body according to one embodiment of the present invention will be described with reference to the drawings.

[0031] FIG. 1 is a schematic diagram of a system for detecting and analyzing a position of an intubation within a human body according to one embodiment of the present invention, and FIG. 2 is a schematic flowchart illustrating a workflow of segmentation, concatenation, and classification of a system for detecting and analyzing a position of an intubation within a human body according to one embodiment of the present invention.

[0032] A system (100) for detecting and analyzing a position of an intubation within a human body according to one embodiment of the present invention is composed of three main modules and has an integrated structure capable of accurately detecting and analyzing the position of the intubation by receiving an X-ray image of a patient who has undergone intubation within the human body. The three main modules may include a segmentation deep learning module (10), a connection module (20), and a classification deep learning module (30). Furthermore, the system (100) for detecting and analyzing a position of an intubation within a human body according to one embodiment of the present invention may further include a visualization module (40) and a verification module (50).

[0033] The segmentation deep learning module (10) identifies and separates the location and shape of the intubation, more specifically, the line of the intubation, in the X-ray image to generate an intubation mask. Furthermore, it is also possible to identify and separate the tip (end) of the intubation and additionally generate an intubation mask related to the tip. The mask refers to a binary image created to visually identify the location of the intubation in the X-ray image. For example, the mask displays the pixel where the intubation is located as 1 (or a color that is visually easy to confirm, such as white or red), and the other parts as 0 (or black), thereby visually distinguishing the location and shape of the intubation. In the system (100) for detecting and analyzing the location of an intubation in the human body according to one embodiment of the present invention, the intubation mask may indicate the location and shape of the line of the intubation in order to increase the accuracy of determining whether a three-dimensional intra-abdominal intubation has been properly inserted.

[0034] The segmentation deep learning module (10) can utilize a deep learning model trained on a pre-trained dataset to determine whether each pixel corresponds to an intubation or not. For example, the segmentation deep learning module (10) can utilize nnU-Net, but the present invention is not limited thereto.

[0035] Pre-training can be performed to improve the performance of the segmentation deep learning module (10). For example, pre-training of the segmentation deep learning module (10) can be performed using the RANZCR CLiP dataset, which contains various information related to chest X-ray images.

[0036] In addition, k-fold cross-validation can be used to improve the performance of the segmented deep learning module (10). K-fold cross-validation divides the dataset into k folds, uses one of the folds as a validation dataset, and the remaining (k-1) folds as a training dataset. This process is repeated k times, so that each fold is used for validation once. Each model is trained in each iteration. The k trained models are evaluated on a test set, and the results of the k models are combined to calculate the final performance. The method for combining the final results can use the average, but the present invention is not limited thereto. This allows for the most efficient use of the entire data, reduces the overfitting problem of the model, and improves generalization performance.

[0037] Finally, an ensemble strategy can be used to improve the performance of the segmentation deep learning module (10). Ensemble combines multiple nnU-Net models to produce a final prediction, effectively improving prediction accuracy by compensating for weaknesses between models.

[0038] The connection module (20) receives the intubation mask generated by the segmentation deep learning module (10) and the original X-ray image. The connection module (20) stacks the intubation mask and the original X-ray image in the form of a channel to generate a mask-combined X-ray image.

[0039] Existing intubation location confirmation technologies have primarily focused on classification. For example, neural networks such as DenseNet, Inception-v3, and EfficientNet have been used to probabilistically classify whether the intubation is in the wrong location in X-ray images. In order to increase classification accuracy in deep learning models, contextual information and clarity of features are very important. However, if probabilistically classifying whether the intubation is in the wrong location using only X-ray images as in the past, the contextual information becomes ambiguous and the features inevitably become unclear. Masks are binary images processed to clearly distinguish a specific region of interest, that is, the location of the intubation. In other words, the mask allows the classification deep learning module (30) in the combined X-ray image to focus on the area where the intubation is located, thereby reducing the interference of unnecessary information. Furthermore, since the mask-combined X-ray image provides a detailed intubation mask combined with the contextual information of the original image (surrounding tissue, positional relationship, etc.), the model can determine the correct location of the intubation together with the surrounding structures, thereby deriving a more accurate classification result than a simple mask or a simple original image.

[0040] These mask-combined X-ray images are transmitted to the classification deep learning module (30) to determine whether the intubation has been inserted correctly. More specifically, the classification deep learning module (30) classifies the appropriateness of the intubation through input that combines the contextual information of the X-ray image and the spatial information of the intubation mask. The classification deep learning module (30) determines whether the intubation has been inserted correctly based on the features learned from the input image. This process follows the principle that the deep learning model learns the pattern of each class (correctly inserted and incorrectly inserted) through a large amount of data and predicts which class a new image belongs to. The model analyzes various features of the image (e.g., shape, location, texture, etc.) to determine whether the intubation is in a normal position and classifies it probabilistically.

[0041] The classification deep learning module (30) can utilize a deep learning model trained on a pre-trained dataset to determine whether the tube has been inserted correctly. For example, ResNet50 can be used as the classification deep learning module (30), but the present invention is not limited thereto.

[0042] Pre-training can be performed to improve the performance of the classification deep learning module (30). For example, pre-training can be performed using MedCLIP, a state-of-the-art medical vision-text pre-training model.

[0043] Furthermore, the human body intubation position detection and analysis system (100) according to one embodiment of the present invention may further include a visualization module (40) and a verification module (50).

[0044] The visualization module (40) generates a heat map indicating the basis area considered important as a basis for the judgment of the above classification deep learning module, and displays the heat map by overlaying it on the mask-combined X-ray image. On the heat map, the area considered important as a basis for judgment is displayed in red (bright area), and the area that is not is displayed in blue (dark area). The basis area can be selected above a user-specified threshold value, and for example, it is possible to set a green to red area as the basis area. At this time, the heat map can be generated using the Grad-CAM (Gradient-weighted Class Activation Mapping) technique. Grad-CAM operates by visually emphasizing the area of ​​the image that the model considers most important when performing classification. Specifically, Grad-CAM combines the feature map of the last convolutional layer with the gradient of the corresponding class to emphasize the part of the image that the model focuses on when predicting a specific class. The mask-combined X-ray image with the heat map overlaid allows medical staff to intuitively understand the basis for the model's predictions.

[0045] The verification module (50) outputs the classification deep learning module's judgment result as inappropriate if the evidence region and the intubation indicated in the intubation mask do not overlap. If the critical evidence region used by the model for prediction does not match the actual location of the intubation, there is a significant possibility that the model made a judgment based on incorrect criteria. Therefore, by verifying whether the evidence region and the intubation mask overlap, it is possible to confirm whether the model's prediction is reliable, i.e., whether the model's judgment is based on correct information, and to filter out incorrect judgments, thereby increasing the reliability of the prediction.

[0046] Below, we will examine in more detail a position detection and analysis system according to one embodiment of the present invention through actual position detection and analysis of a gastric tube.

[0047] The study on nasogastric tube position detection and analysis was conducted in compliance with the Declaration of Helsinki and was approved by the ethics committees of Chuncheon Sacred Heart Hospital, Gangneung Asan Medical Center, and Kangwon National University Hospital. Subject to ethics committee approval, consent for the anonymous use of clinical data for research purposes was not required from participants.

[0048] Data were collected as follows. In this study, a comprehensive dataset containing 2,627 anonymized patient X-ray images was collected from three major medical institutions: Chuncheon Sacred Heart Hospital, Gangneung Asan Medical Center, and Kangwon National University Hospital. Data from Kangwon National University Hospital were used for external verification purposes. The systematic collection of these records was conducted in strict compliance with the approval of each hospital's Institutional Review Board (IRB). All images were carefully extracted and stored in DICOM format using a Picture Archiving and Communication System (PACS). Patient consent was not required because the data underwent thorough de-identification. The characteristics of the patients included in the dataset are summarized in Table 1 below.

[0049]

[0050] [Table 1]

[0051]

[0052]

[0053] To obtain labeled data, clinical experts labeled normal NG tube (nasogastric tube) insertion as “complete” and abnormal status as “incomplete.” This was based on the opinions of four qualified experts (MKM, KHG, HJY, and HJW). Through this process, 1,957 complete samples and 292 incomplete samples were used for model training and testing. Three respiratory specialists (MKM, HJY, and HJW) and one radiologist (KHS) with more than 10 years of clinical experience collected a dataset of 2,249 chest X-rays with labels for NG tube positions. All cases were classified as “complete” or “incomplete.”

[0054] When drug administration or enteral nutrition was possible clinically through the NG tube, it was classified as “complete,” and this was defined as normal NG tube position. Conversely, when drug administration or enteral nutrition was not possible, it was classified as “incomplete,” and this was defined as abnormal NG tube position. Normal tube position was when the NG tube did not enter the airway and the tip was located at least 3 cm below the diaphragm on chest X-ray PA or AP view. If it was difficult for a single clinical expert to classify the NG tube position as “complete” or “incomplete,” four experts simultaneously reviewed the NG tube position on chest X-ray. Labels were assigned based on the classification that all agreed upon.

[0055] First, an intubation mask for the nasogastric tube was generated. To do this, a dedicated segmentation model fθ, designed to identify the intubation mask including the line and tip of the intubation in the X-ray image, was introduced. The model fθ is a preprocessed and normalized image (x) that is derived from the line (m l ∈ {0, 1} HХW ) and tip (m t ∈ {0, 1} HХW) is generated. Here, it is calculated as in Equation 1 below according to the sample index (i).

[0056]

[0057] [Formula 1]

[0058] m s ∈ {l, t} = argmax(Φ(fθ(x i )))

[0059]

[0060] Here, Φ represents the postprocessing of nnU-Net. This process extracts relevant information related to the nasogastric line, preparing it for improved shape recognition in subsequent steps.

[0061] We achieved precise nasogastric tube detection using nnU-Net, a state-of-the-art framework based on the U-Net architecture for biomedical segmentation, as the segmentation model fθ. To improve the accuracy of the segmentation model, we strategically employed pretraining techniques. Specifically, model fθ was pretrained on RANZCR CLiP.

[0062] The original X-ray image and the intubation mask were combined (⊙) to create a mask-combined X-ray image containing more information (see Equation 2).

[0063]

[0064] [Formula 2]

[0065] z i = x i ⊙ m s ∈ {l, t}

[0066]

[0067] Here x i is an X-ray image, and m s ∈ {l, t} is a segmented mask. The input z, which is a mask-joined X-ray image, greatly improves the classification deep learning module's understanding of the anatomical structures and nasogastric mask in the X-ray image.

[0068] The classification deep learning module trained a model (gθ) using the mask-combined X-ray image as input to predict whether the nasogastric tube was abnormally positioned. The input z, which is the mask-combined X-ray image, was input to the model gθ to calculate the probability p of the nasogastric tube's incorrect position. This is p i = σ(gθ(z i )), where σ is an activation function. These probabilities are optimized by the cross-entropy loss function, so that the model classifies a location as either 'complete' or 'incomplete'.

[0069] Here, we used a ResNet-based architecture, one of the most widely used neural networks in image analysis, as the model (gθ). For more accurate performance, we also used a pre-trained model from MedCLIP, a state-of-the-art medical vision-text pre-training model, as the classification network (gθ).

[0070] Furthermore, visualization and validation modules were introduced to enhance the model's interpretability, understand the decision-making process behind predictions, and enhance reliability through validation. The visualization module utilized Gradient-weighted Class Activation Mapping (Grad-CAM). Grad-CAM provides a visual explanation of the model's decisions and highlights specific areas of the X-ray image that the model focuses on when determining the nasogastric tube position. This approach not only helps validate the model's accuracy and reliability, but also identifies potential areas requiring additional training or improvement, contributing to continuous improvement of the model's performance.

[0071] Both the network and experimental setup were implemented using the PyTorch framework. The segmentation deep learning module was trained using five-fold cross-validation to increase robustness, and an ensemble strategy was used to ensure accurate segmentation results. Due to the large size difference in X-ray images, the input patch size was set to 512 х 512. Similar implementation details, such as optimization methods and data augmentation techniques, were applied to nnU-Net. In the classification step, all networks were trained on a single NVIDIA RTX 3090 GPU, and all network parameters were optimized using the SGD optimizer via stratified five-fold cross-validation. The learning rate of the feature extractor in the classification model was set to 0.0005, which is 10 times lower than the learning rate of a classifier consisting of a single MLP layer. All networks were trained for 100 epochs with a batch size of 16. The weights of the feature extractor were initialized with MedClip's ResNet-50, and the weights of the classifier were initialized randomly. The inputs of all models were uniformly resized to 512 х 512, and a random horizontal flip was applied with probability 0.5 during training. The training dataset was split into training data (80%) and validation data (20%), and all hyperparameters were tuned during the validation phase.

[0072] The training dataset contained 1,799 images, and the test dataset contained 450 images. In the segmentation step, the training dataset was used for five-fold cross-validation, and the test set was used to evaluate model performance. Additionally, 378 images provided by Kangwon National University Hospital were used for external validation to further evaluate the model's robustness and generalizability.

[0073] Since the present invention includes two deep learning models for segmentation and classification, six indicators were used to evaluate each model.

[0074] To evaluate the segmentation deep learning module, we first used the Dice coefficient and Jaccard coefficient, which represent the similarity between predicted and actual values. These indices are calculated using true positives, false positives, and false negatives (denoted TP, FP, and FN, respectively).

[0075]

[0076] [Formula 3]

[0077] Dice = 2TP / (2TP + FP + FN)

[0078] Jaccard = TP / (TP + FP + FN)

[0079]

[0080] Considering the imbalanced class distribution, the macro F1 score, balanced accuracy (B-ACC), area under the receiver operating characteristic curve (AUC-ROC), and area under the precision-recall curve (AUC-PR) were also evaluated to evaluate the classification model. The F1 score and B-ACC are calculated as follows.

[0081]

[0082] [Formula 4]

[0083] F1 score = 2 × (precision * recall) / (precision + recall)

[0084] Balanced accuracy = (sensitivity + specificity) / 2

[0085]

[0086] We used exponential moving averages (EMA) to track the performance metrics of the segmentation deep learning module. To ensure robust and accurate predictions, we ensembled the logic of five-layer trained models when evaluating the test dataset. The segmentation deep learning module achieved a Dice coefficient of 65.35% and a Jaccard coefficient of 57.49% on the test dataset (see Table 2).

[0087]

[0088] [Table 2]

[0089]

[0090]

[0091] As shown in Table 3, when the X-ray image combined with the intubation mask (line or tip) was used as input, the classification ability of the classification deep learning module was consistently improved in all indices such as F1-score, B-ACC, AUC-ROC, and AUC-PR. Compared with the method using only the original X-ray image, the combination of the X-ray image and the line mask (x + ml) showed performance improvements of 1.74%, 3.71%, 1.79%, and 2.09%, respectively. Similarly, the combination of the X-ray image and all masks (x + ml + mt) showed performance improvements of 0.66%, 2.14%, 1.55%, and 1.5%, respectively. Although all mask combinations helped improve model performance, using all masks (x + ml + mt) resulted in 1.08%, 1.57%, 0.24%, and 0.59% lower performance than using only the line mask, respectively.

[0092]

[0093] [Table 3]

[0094]

[0095]

[0096] According to Table 4, ResNet-50 outperformed all other architectures, except for ResNet-50 pre-trained with MedCLIP. These results confirm that ResNet-50 is the most suitable model. ResNet-50 pre-trained with MedCLIP improved performance by 2.11%, 1.79%, 0.38%, and 0.85%, respectively, compared to the baseline ResNet-50. These results suggest the potential of deep learning-based methods to detect incorrect insertions of NGT.

[0097]

[0098] [Table 4]

[0099]

[0100]

[0101] As shown in Table 5 below, the present invention achieved a Dice Similarity Coefficient of 65.35±0.25, an Area Under the Curve (AUC) of 99.72, and an Area Up to the Center of Ratio (AUPRC) of 99.30. Among these, AUC is a key indicator for evaluating a model's predictive performance, specifically indicating how well the model distinguishes between positive and negative classes. A high AUC value indicates that the model can distinguish between normal and incorrect locations with high accuracy. While the comparative example achieved an AUC of only 92, the present invention achieved a whopping 99.72, demonstrating significantly higher reliability.

[0102]

[0103] [Table 5]

[0104]

[0105]

[0106] In addition, five-fold crossover models were evaluated using an ensemble method on external datasets, and the results are shown in Table 6. The F1-score, B-ACC, AUC-ROC, and AUC-PR values ​​were 70.70%, 82.41%, 88.14%, and 79.77%, respectively. This suggests that the model can be applied to a wider range of clinical settings.

[0107]

[0108] [Table 6]

[0109]

[0110]

[0111] Next, visualization was performed. Fig. 3 is a reference diagram for explaining the process of detecting and analyzing the intubation location within the human body from an external verification dataset by the system for detecting and analyzing the intubation location within the human body according to one embodiment of the present invention. Fig. 3 relates to (a) an X-ray image, (b) an intubation mask labeled with the correct answer learned by the segmentation deep learning module, (c) a result of combining the intubation mask generated by the segmentation deep learning module and the original X-ray image, and (d) a heat map result.

[0112] Referring to Figure 3, the top two rows represent incorrectly predicted cases (incomplete, complete), while the bottom two rows represent correctly predicted cases (incomplete, complete). That is, as seen in the heatmaps in the top two rows, the classification deep learning module sometimes highlights areas unrelated to the insertion, such as inappropriate text or peripheral areas. In such cases, the verification module determines that the evidence area and the insertion mask do not overlap, and outputs the classification deep learning module's judgment as inappropriate.

[0113] Finally, Fig. 4 is a prototype user interface of a system for detecting and analyzing the location of an intubation within a human body according to one embodiment of the present invention. As shown in Fig. 4, when medical staff needs to determine the location and analysis of an intubation within a human body, they input a related X-ray image. The location of the intubation and the basis for judgment are displayed and output on the X-ray image, and the appropriateness of the intubation and the verification results are also provided. Medical staff can visually confirm the appropriateness or inappropriateness of the intubation, rather than simply being provided with a probability, which can assist in clinical judgment.

[0114] The system for detecting and analyzing the position of an intra-human intubation device as described above 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 non-transitory computer-readable medium. Here, the non-transitory computer-readable medium refers to a medium that permanently stores data 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 can be stored and provided on a non-transitory computer-readable medium, such as a CD, DVD, hard disk, Blu-ray disc, USB, memory card, or ROM.

[0115] The scope of protection of the present invention is not limited to the description and expression of the embodiments explicitly described above. Furthermore, it should be noted that the scope of protection of the present invention may not be limited by obvious modifications or substitutions within the technical field to which the present invention pertains.

[0116]

[0117] This invention was supported by the following national research and development projects:

[0118] [Project Number] S1402-23-1001

[0119] [Ministry Name] NIPA

[0120] [Name of Project Management (Specialist) Agency] KCA

[0121] [Research Project Name] AI-Based Medical Solution Commercialization Demonstration Support Project

[0122] [Research Project Name] Detection of the type and location of tubes and catheters in chest X-ray images.

[0123] Development of medical AI software to improve patient safety through evaluation

[0124] [Contribution rate] 1 / 1

[0125] [Name of Project Performing Organization] Geovision Co., Ltd.

[0126] Research Period: July 1, 2023 - December 31, 2025

Claims

1. A segmentation deep learning module that receives an X-ray image as input and generates an intubation mask for the location of the intubation inside the human body included in the X-ray image; A connection module for connecting the above-mentioned intubation mask and the X-ray image to generate a mask-combined X-ray image; and A system for detecting and analyzing a location of an intubation within a human body, comprising a classification deep learning module that receives the above mask-combined X-ray image as input and classifies the appropriateness of the location of the intubation within the human body.

2. In paragraph 1, A system for detecting and analyzing the position of an intubation within a human body, characterized in that the intubation is at least one selected from the group consisting of a nasogastric tube, a central venous catheter, an endotracheal tube, and a thoracic drainage tube.

3. In paragraph 1, A system for detecting and analyzing the position of an intubation site within a human body, characterized in that the above intubation mask is a mask regarding the shape and position of the line of intubation.

4. In paragraph 3, A system for detecting and analyzing an intubation position within a human body, characterized in that the intubation mask further includes a mask regarding the shape of the tip of the intubation and the body.

5. In paragraph 1, A system for detecting and analyzing an intubation position within a human body, characterized in that the above mask-combined X-ray image is generated by stacking the intubation mask and the X-ray image in a channel form.

6. In paragraph 1, A system for detecting and analyzing a location of an intubation within a human body, further comprising a visualization module for visualizing the judgment result of the classification deep learning module, wherein the visualization module generates a heat map indicating a basis area considered important as a basis for the judgment of the classification deep learning module and displays the heat map by overlaying it on the mask-combined X-ray image.

7. In paragraph 4, In one embodiment, a system for detecting and analyzing a location of an intubation within a human body, further comprising a verification module for verifying the appropriateness of the judgment result of the classification deep learning module, wherein the verification module outputs that the judgment result of the classification deep learning module is inappropriate when the basis area and the intubation mask do not overlap.

Images