PICC positioning review method and related device based on chest X-ray

Abstract

The present invention discloses a PICC positioning review method and related device based on an X-ray chest film, wherein the method comprises: obtaining an X-ray chest film image of a target patient; establishing an X-ray chest film PICC positioning model to identify the anatomical structure and PICC catheter path in the X-ray chest film image; the X-ray chest film PICC positioning model comprises an anatomical structure recognition unit, a catheter path recognition unit, and a position evaluation unit; the anatomical structure recognition unit is used to recognize the anatomical structure in the X-ray chest film image; the catheter path recognition unit is used to recognize the PICC catheter path in the X-ray chest film image, the PICC catheter path including the starting point, running path, and terminal position of the catheter; the position evaluation unit is used to evaluate the catheter terminal position based on the anatomical structure and the PICC catheter path to generate a position evaluation result; and reviewing the PICC catheter positioning based on the PICC catheter tip position information text and the position evaluation result. The present invention improves the accuracy and efficiency of PICC positioning review by using the X-ray chest film PICC positioning model.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a PICC positioning review method and device based on chest X-rays, and a computing device. Background Art

[0002] Peripherally inserted central catheters (PICCs) are widely used clinically. However, after PICC placement, the position of the catheter tip must be confirmed before use. Currently, the gold standard for PICC tip positioning is chest X-rays, and clinically, chest X-rays are generally used for PICC tip positioning in hospitalized patients. Chest X-rays are convenient and quick, but clinical physicians have difficulty identifying and interpreting the results, often requiring consultation with PICC specialists or dedicated personnel to determine whether the PICC tip has reached the ideal position. The entire process is time-consuming and complex.

[0003] To solve the above problems, the present invention proposes a PICC positioning review method and device based on chest X-ray, as well as a computing device, to improve the accuracy and efficiency of PICC positioning review. Summary of the Invention

[0004] In view of the above problems, the present invention provides a PICC positioning review method and device based on chest X-ray, and a computing device.

[0005] According to one aspect of the present invention, a method for PICC positioning review based on chest X-ray is provided, comprising:

[0006] Obtain chest X-ray images of target patients through the hospital information system;

[0007] Establishing an X-ray chest film PICC positioning model to identify anatomical structures and PICC catheter paths in X-ray chest film images; wherein the X-ray chest film PICC positioning model includes an anatomical structure recognition unit, a catheter path recognition unit, and a position evaluation unit; the anatomical structure recognition unit is used to recognize anatomical structures in X-ray chest film images, wherein the anatomical structures include trachea, clavicle, ribs, spine, heart outline, and mediastinal anatomical landmarks; the catheter path recognition unit is used to recognize the PICC catheter path in X-ray chest film images, wherein the PICC catheter path includes the starting point, running path, and terminal position of the catheter; the position evaluation unit is used to evaluate the terminal position of the catheter based on the anatomical structure and the PICC catheter path, and generate a position evaluation result;

[0008] Generate PICC catheter tip position information text that integrates visual features and text coding features according to the anatomical structure and PICC catheter path; and review PICC catheter positioning based on the PICC catheter tip position information text and position assessment results.

[0009] In an optional manner, the anatomical structure recognition unit utilizes skip connections to fuse features of corresponding levels in the encoder and outputs a pixel-level segmentation mask for each anatomical structure; wherein the pixel values ​​of the segmentation mask are all 0 or 1, 0 represents background and 1 represents anatomical structure;

[0010] The precision and recall of the segmentation mask of each anatomical structure are calculated, and the F1 score is used as the evaluation indicator for each anatomical structure.

[0011] In an optional manner, the catheter path identification unit uses a multi-scale edge response algorithm to identify the catheter edge;

[0012] The straight line segments in the catheter edge are detected by Hough transform algorithm and optimized by dynamic program method.

[0013] The bidirectional LSTM network is used to process each straight line segment to obtain the coordinates of the starting point, intermediate nodes, and end position of the PICC catheter.

[0014] According to the coordinates of the starting point, middle node and end position of the PICC catheter, the PICC catheter path is connected and generated, and the PICC catheter path is smoothed by B-spline curve.

[0015] In an optional manner, the assessing of the catheter tip position based on the anatomical structure and the PICC catheter path further comprises:

[0016] Calculate the distance between the PICC catheter tip and the entrance of the superior vena cava;

[0017] If the distance is less than a preset threshold, the end position is located in the superior vena cava; wherein the superior vena cava entry point is determined by the relative positions of the trachea mask and the clavicle mask;

[0018] If the catheter tip is located in the right atrium, the tip is located in the right atrium; if the catheter tip is located in the right ventricle, the tip is located in the right ventricle; wherein the ranges of the right atrium and the right ventricle are determined by the heart contour mask.

[0019] In an optional manner, generating the PICC catheter tip position information text that integrates visual features and text coding features further includes:

[0020] Extract the local image of the PICC catheter end position in the X-ray chest image, and use the ResNet network to extract the visual features of the local image to obtain the visual feature vector;

[0021] Generate location description text based on the location evaluation results, and encode the location description text through the BERT model to obtain a text encoding vector;

[0022] The visual feature vector and the text encoding vector are spliced ​​together to generate a fusion vector and obtain the PICC catheter tip position information text.

[0023] In an optional manner, the reviewing of the PICC catheter positioning according to the PICC catheter tip position information text and the position assessment result further includes:

[0024] Assign different confidence scores to the classification of each evaluation result to obtain the confidence score of the evaluation result;

[0025] The PICC catheter tip location information text and location evaluation results are input into the multi-layer perceptron network, which outputs the location category of the PICC catheter tip and the corresponding confidence score;

[0026] The confidence score of the assessment result and the confidence score of the PICC catheter tip position category were weighted and summed to obtain the review confidence score.

[0027] In an optional manner, extracting the visual features of the local image using the ResNet network to obtain the visual feature vector further includes:

[0028] Use the ResNet network as the basic feature extractor and remove the fully connected layer of the ResNet network;

[0029] Introducing an attention mechanism module to weight the extracted feature maps, wherein the attention mechanism module includes at least one convolutional layer and a sigmoid activation layer;

[0030] The weighted feature map is subjected to global average pooling to obtain a visual feature vector.

[0031] In an optional manner, a Transformer encoder layer is introduced before the input layer of the multi-layer perceptron to capture the contextual dependency between the PICC catheter tip position information text and the position assessment result;

[0032] A cross-modal fusion module is introduced between the output of the Transformer encoder layer and the input layer of the multi-layer perceptron network to concatenate the fused feature vector of the text output and the evaluation result of the Transformer encoder with the global visual feature vector.

[0033] The Transformer encoder layer includes:

[0034] The embedding layer maps the word vectors of the PICC catheter tip location information text and the category one-hot encoding vectors of the location assessment results into a high-dimensional space;

[0035] Position encoding layer, which adds position information to the embedded vector to preserve the sequence order;

[0036] Multiple interleaved and stacked multi-head self-attention sublayers and feedforward neural network sublayers, wherein the multi-head self-attention sublayer uses a self-attention mechanism to calculate the correlation between input vectors, and the feedforward neural network sublayer performs nonlinear transformation.

[0037] According to another aspect of the present invention, a PICC positioning and review device based on chest X-ray is provided, comprising:

[0038] A data acquisition module is used to obtain the chest X-ray images of target patients through the hospital information system;

[0039] a model building module for establishing an X-ray chest film PICC positioning model to identify anatomical structures and PICC catheter paths in X-ray chest film images; wherein the X-ray chest film PICC positioning model includes an anatomical structure recognition unit, a catheter path recognition unit, and a position evaluation unit; the anatomical structure recognition unit is used to recognize anatomical structures in X-ray chest film images, wherein the anatomical structures include trachea, clavicle, ribs, spine, heart outline, and mediastinal anatomical landmarks; the catheter path recognition unit is used to recognize the PICC catheter path in X-ray chest film images, wherein the PICC catheter path includes the starting point, running path, and terminal position of the catheter; and the position evaluation unit is used to evaluate the terminal position of the catheter based on the anatomical structure and the PICC catheter path, and generate a position evaluation result;

[0040] The positioning review module is used to generate PICC catheter tip position information text that integrates visual features and text coding features according to the anatomical structure and PICC catheter path; and to review the PICC catheter positioning based on the PICC catheter tip position information text and the position evaluation results.

[0041] According to another aspect of the present invention, there is provided a computing device, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus;

[0042] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned PICC positioning review method based on chest X-ray.

[0043] According to the solution provided by the present invention, an X-ray chest image of a target patient is obtained through a hospital information system; a PICC positioning model for the X-ray chest image is established to identify the anatomical structure and PICC catheter path in the X-ray chest image; wherein the X-ray chest image PICC positioning model includes an anatomical structure recognition unit, a catheter path recognition unit, and a position evaluation unit; the anatomical structure recognition unit is used to recognize the anatomical structure in the X-ray chest image, the anatomical structure including the trachea, clavicle, ribs, spine, heart outline, and mediastinal anatomical landmarks; the catheter path recognition unit is used to recognize the PICC catheter path in the X-ray chest image, the PICC catheter path including the starting point, running path, and end position of the catheter; the position evaluation unit is used to evaluate the catheter end position based on the anatomical structure and the PICC catheter path to generate a position evaluation result; based on the anatomical structure and the PICC catheter path, a PICC catheter tip position information text is generated that integrates visual features and text coding features; and the PICC catheter positioning is reviewed based on the PICC catheter tip position information text and the position evaluation result. The present invention improves the accuracy and efficiency of PICC positioning review through the X-ray chest image PICC positioning model. Specifically, by identifying key anatomical structures such as the trachea, clavicle, ribs, spine, heart outline, and mediastinal anatomical landmarks, a reliable reference is provided for the positioning of PICC catheters. By identifying the starting point, path, and end position of the catheter, it helps to determine whether the catheter has abnormal course. An objective position assessment result is generated based on the anatomical structure and catheter path, reducing the bias of subjective judgment. The generated PICC catheter tip position information text, which integrates visual features and text coding features, facilitates doctors to quickly understand the catheter position information.

[0044] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are specifically listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0046] Figure 1 A schematic diagram showing a flow chart of a PICC positioning review method based on chest X-ray according to an embodiment of the present invention is shown;

[0047] Figure 2The schematic diagram of the process of the PICC positioning review method based on X-ray chest film of the embodiment of the present invention is shown. Figure 2 ;

[0048] Figure 3 A schematic diagram of a framework of a PICC positioning and review device based on chest X-ray according to an embodiment of the present invention is shown;

[0049] Figure 4 A schematic structural diagram of a computing device according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0050] Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0051] Figure 1 FIG. 1 is a flow chart showing a method for PICC positioning review based on chest X-ray according to an embodiment of the present invention. Specifically, Figure 1 As shown, the following steps are included:

[0052] Step S101: Obtain a chest X-ray image of a target patient through a hospital information system.

[0053] For example, a chest X-ray image list of a target patient is obtained in a PACS system.

[0054] Step S102, establishing an X-ray chest film PICC positioning model to identify the anatomical structure and PICC catheter path in the X-ray chest film image; wherein the X-ray chest film PICC positioning model includes an anatomical structure recognition unit, a catheter path recognition unit and a position evaluation unit; the anatomical structure recognition unit is used to identify the anatomical structure in the X-ray chest film image, and the anatomical structure includes the trachea, clavicle, ribs, spine, heart contour and mediastinal anatomical landmarks; the catheter path recognition unit is used to identify the PICC catheter path in the X-ray chest film image, and the PICC catheter path includes the starting point, running path and end position of the catheter; the position evaluation unit is used to evaluate the catheter end position according to the anatomical structure and the PICC catheter path to generate a position evaluation result.

[0055] In this embodiment, the X-ray chest PICC positioning model includes three units: an anatomical structure recognition unit, a catheter path recognition unit, and a position assessment unit. The anatomical structure recognition unit processes the X-ray chest image as input and outputs the outline or position information of the anatomical structure. The catheter path recognition unit processes the image and outputs the catheter's starting point, path, and terminal position information. The position assessment unit integrates the anatomical structure and catheter path information to generate a catheter terminal position estimation result and outputs the position estimation result and a visually marked image.

[0056] Among them, the anatomical structure recognition unit is responsible for identifying key anatomical structures in X-ray chest films. For example, the trachea helps determine the midline of the chest and serves as a reference for the entrance of the superior vena cava. The clavicle provides a skeletal framework for the upper chest, assisting in determining the location of the subclavian vein. The ribs provide a framework for the thorax, making it easier to determine the boundaries of the thoracic cavity and locate the position of the catheter in the thoracic cavity. The spine serves as a midline reference to assist in determining whether the catheter deviates from the midline. The heart outline is used to determine whether the catheter enters the heart or pericardium. Mediastinal anatomical landmarks (such as the aortic arch and pulmonary artery) are used to determine whether the catheter is located within the blood vessel. A large number of X-ray chest images are required for training data, and each image needs to be manually annotated with the outlines of the above anatomical structures.

[0057] The catheter path recognition unit is responsible for identifying the PICC catheter path in chest X-rays. The catheter starting point is the point where the catheter is inserted into the skin, the path is the path the catheter takes within the blood vessel, and the end position is the location of the catheter tip. Each image in the training data requires manual annotation of the catheter's starting point, path, and end position.

[0058] The position assessment unit evaluates the position of the catheter tip based on anatomical structures and catheter path information and generates an assessment result. For example, an ideal position is when the catheter tip is at the junction of the superior vena cava and right atrium. An abnormal position is when the catheter tip enters the heart, pericardium, internal jugular vein, or subclavian vein. Potential risks include risks caused by excessively deep or shallow catheter tip placement.

[0059] In an optional manner, the anatomical structure recognition unit utilizes skip connections to fuse features of corresponding levels in the encoder and outputs a pixel-level segmentation mask for each anatomical structure; wherein the pixel values ​​of the segmentation mask are all 0 or 1, 0 represents background and 1 represents anatomical structure;

[0060] The precision and recall of the segmentation mask of each anatomical structure are calculated, and the F1 score is used as the evaluation indicator for each anatomical structure.

[0061] In this embodiment, skip connections are used to fuse features from different levels in the encoder with features from corresponding levels in the decoder. This preserves image details and improves segmentation accuracy, which is particularly important for identifying small objects in medical images. A pixel-level segmentation mask is output for each anatomical structure, where each pixel has a value of either 0 or 1, more accurately demarcating the boundaries of each anatomical structure.

[0062] Specifically, the encoder uses a convolutional neural network to extract features from the input image, gradually reducing the size of the feature map and increasing the number of channels. The decoder uses deconvolution or upsampling operations to gradually increase the size of the feature map and reduce the number of channels, gradually restoring the low-resolution abstract features to high-resolution pixel-level features. Skip connections directly transfer the feature map corresponding to each downsampling layer in the encoder to the corresponding upsampling layer in the decoder, connecting the feature maps along the channel dimension.

[0063] In an optional manner, the catheter path identification unit uses a multi-scale edge response algorithm to identify the catheter edge;

[0064] The straight line segments in the catheter edge are detected by Hough transform algorithm and optimized by dynamic program method.

[0065] The bidirectional LSTM network is used to process each straight line segment to obtain the coordinates of the starting point, intermediate nodes, and end position of the PICC catheter.

[0066] According to the coordinates of the starting point, middle node and end position of the PICC catheter, the PICC catheter path is connected and generated, and the PICC catheter path is smoothed by B-spline curve.

[0067] In this example, a sequence of straight line segments is fed into a bidirectional long short-term memory (Bi-LSTM) network, capturing the contextual relationships within the sequence data and predicting the coordinates of the catheter's starting point, intermediate nodes, and end points. Based on the coordinates obtained from the Bi-LSTM network, the PICC catheter path is constructed. To achieve a smoother and more natural path, a B-spline curve is used to smooth the path and reduce jitter or uneven sections.

[0068] In an optional manner, the assessing of the catheter tip position based on the anatomical structure and the PICC catheter path further comprises:

[0069] Calculate the distance between the PICC catheter tip and the entrance of the superior vena cava;

[0070] If the distance is less than a preset threshold, the end position is located in the superior vena cava; wherein the superior vena cava entry point is determined by the relative positions of the trachea mask and the clavicle mask;

[0071] If the catheter tip is located in the right atrium, the tip is located in the right atrium; if the catheter tip is located in the right ventricle, the tip is located in the right ventricle; wherein the ranges of the right atrium and the right ventricle are determined by the heart contour mask.

[0072] In this embodiment, a preset threshold is used to determine whether the catheter end is located in the superior vena cava. If the distance between the catheter end and the entrance of the superior vena cava is less than this threshold, the catheter end is considered to be located in the superior vena cava. The determination of the entrance point of the superior vena cava can be achieved by analyzing the relative positions of the trachea and clavicle, because these anatomical structures usually have a fixed spatial relationship with the superior vena cava in the image. If the position of the catheter end is not in the superior vena cava, it is further determined whether it is located in the heart. The range of the right atrium is determined by the heart contour mask. If the catheter end is within this range, it is determined to be located in the right atrium. By using the relative position of the anatomical structure and the mask to determine the position of the catheter, it is possible to accurately evaluate whether the position of the catheter end meets medical requirements and ensure the safe placement of the catheter.

[0073] Step S103: generating a PICC catheter tip position information text that integrates visual features and text coding features according to the anatomical structure and the PICC catheter path; and reviewing the PICC catheter positioning according to the PICC catheter tip position information text and the position evaluation result.

[0074] In this embodiment, features such as brightness, contrast, and surrounding vascular structure of the catheter can also be extracted. For text encoding features, a series of text labels are predefined and relevant anatomical terms are extracted, such as "inside the superior vena cava", "inside the right atrium", "inside the right ventricle", etc. Position description text is automatically generated based on the catheter path and anatomical structure, such as "the end of the catheter is located 2 cm above the entrance of the superior vena cava", "the catheter passes through the subclavian vein and enters the superior vena cava", etc. The extracted visual features (coordinates, mask information) and text encoding features (position labels, anatomical terms) are fused to generate structured catheter tip position information text. For example:

[0075]

[0076] The generated PICC catheter tip position information text, catheter path visualization results, and position assessment results (distance, whether it is located in the lumen) are displayed to the doctor or other reviewers. Based on the above information, a manual review is performed to determine whether the catheter tip position meets the requirements. Pass: If the catheter tip position is correct, it is marked as passed. If the catheter tip position is incorrect, it is marked as unqualified and the catheter position needs to be adjusted or further checked. If there is uncertainty or doubt in the information, it is marked as requiring review.

[0077] Key points to review include: whether the catheter tip is located at the intended target location (usually the superior vena cava or its junction with the right atrium); whether the catheter path is smooth, with no significant bends or kinks; and whether the catheter position poses a potential risk to the patient (for example, whether it enters the right ventricle).

[0078] Then, automated review assistance is implemented based on pre-set rules (e.g., the superior vena cava entrance distance threshold), and situations that do not comply with the rules are automatically marked for manual review, thereby reducing the workload of manual review.

[0079] In an optional manner, generating the PICC catheter tip position information text that integrates visual features and text coding features further includes:

[0080] Extract the local image of the PICC catheter end position in the X-ray chest image, and use the ResNet network to extract the visual features of the local image to obtain the visual feature vector;

[0081] Generate location description text based on the location evaluation results, and encode the location description text through the BERT model to obtain a text encoding vector;

[0082] The visual feature vector and the text encoding vector are spliced ​​together to generate a fusion vector and obtain the PICC catheter tip position information text.

[0083] In this embodiment, the local image includes the catheter tip and the surrounding anatomical structures, so as to provide sufficient context information.

[0084] In an optional manner, the reviewing of the PICC catheter positioning according to the PICC catheter tip position information text and the position assessment result further includes:

[0085] Assign different confidence scores to the classification of each evaluation result to obtain the confidence score of the evaluation result;

[0086] The PICC catheter tip location information text and location evaluation results are input into the multi-layer perceptron network, which outputs the location category of the PICC catheter tip and the corresponding confidence score;

[0087] The confidence score of the assessment result and the confidence score of the PICC catheter tip position category were weighted and summed to obtain the review confidence score.

[0088] In this embodiment, the position assessment results are classified, for example, within the superior vena cava (SVC): the catheter tip is located in the ideal position in the superior vena cava. Within the right atrium (RA): the catheter tip enters the right atrium, which may not be ideal. Within the right ventricle (RV): the catheter tip enters the right ventricle, which is usually not ideal and needs to be repositioned. Other locations: other situations where the position of the catheter tip is not within the expected range. A different confidence score is assigned to each classification result, and the confidence score reflects the reliability of the position assessment result. For example, within the superior vena cava (SVC): 0.95 (indicating high confidence, the catheter position is good); within the right atrium (RA): 0.75 (indicating moderate confidence, further evaluation is required); within the right ventricle (RV): 0.2 (indicating low confidence).

[0089] The generated PICC tip location text (a vector representation that fuses visual features and text-encoded features) and the location assessment result (a numerical or categorical value) are fed into a multi-layer perceptron (MLP) network, which outputs the PICC tip location category (SVC, RA, RV, or other) and the corresponding confidence score. The confidence score of the assessment result and the confidence score output by the MLP are weighted and summed to obtain the final review confidence score.

[0090] In an optional manner, extracting the visual features of the local image using the ResNet network to obtain the visual feature vector further includes:

[0091] Use the ResNet network as the basic feature extractor and remove the fully connected layer of the ResNet network;

[0092] Introducing an attention mechanism module to weight the extracted feature maps, wherein the attention mechanism module includes at least one convolutional layer and a sigmoid activation layer;

[0093] The weighted feature map is subjected to global average pooling to obtain a visual feature vector.

[0094] In this example, the goal is to extract general visual features rather than perform specific classification or regression tasks. Retaining the fully connected layers biases the features learned by the network toward specific classification or regression targets, reducing the generalization ability of the features. Removing the fully connected layers preserves the more primitive, general features extracted by the convolutional layers.

[0095] In an optional manner, a Transformer encoder layer is introduced before the input layer of the multi-layer perceptron to capture the contextual dependency between the PICC catheter tip position information text and the position assessment result;

[0096] A cross-modal fusion module is introduced between the output of the Transformer encoder layer and the input layer of the multi-layer perceptron network to concatenate the fused feature vector of the text output and the evaluation result of the Transformer encoder with the global visual feature vector.

[0097] The Transformer encoder layer includes:

[0098] The embedding layer maps the word vectors of the PICC catheter tip location information text and the category one-hot encoding vectors of the location assessment results into a high-dimensional space;

[0099] Position encoding layer, which adds position information to the embedded vector to preserve the sequence order;

[0100] Multiple interleaved and stacked multi-head self-attention sublayers and feedforward neural network sublayers, wherein the multi-head self-attention sublayer uses a self-attention mechanism to calculate the correlation between input vectors, and the feedforward neural network sublayer performs nonlinear transformation.

[0101] In this example, the PICC catheter tip location text refers to a textual description of the catheter tip location, such as "located in the superior vena cava" or "near the atrium." The location assessment result refers to a preliminary assessment of the catheter tip location, such as "correct," "too deep," or "too shallow." The Transformer encoder aims to understand the relationship between the textual description and the location assessment result, for example, that the two pieces of information, "near the atrium" and "too deep," are contextually related.

[0102] According to the solution provided by the present invention, an X-ray chest image of a target patient is obtained through a hospital information system; a PICC positioning model for the X-ray chest image is established to identify the anatomical structure and PICC catheter path in the X-ray chest image; wherein the X-ray chest image PICC positioning model includes an anatomical structure recognition unit, a catheter path recognition unit, and a position evaluation unit; the anatomical structure recognition unit is used to recognize the anatomical structure in the X-ray chest image, the anatomical structure including the trachea, clavicle, ribs, spine, heart outline, and mediastinal anatomical landmarks; the catheter path recognition unit is used to recognize the PICC catheter path in the X-ray chest image, the PICC catheter path including the starting point, running path, and end position of the catheter; the position evaluation unit is used to evaluate the catheter end position based on the anatomical structure and the PICC catheter path to generate a position evaluation result; based on the anatomical structure and the PICC catheter path, a PICC catheter tip position information text is generated that integrates visual features and text coding features; and the PICC catheter positioning is reviewed based on the PICC catheter tip position information text and the position evaluation result. The present invention improves the accuracy and efficiency of PICC positioning review through the X-ray chest image PICC positioning model. Specifically, by identifying key anatomical structures such as the trachea, clavicle, ribs, spine, heart outline, and mediastinal anatomical landmarks, a reliable reference is provided for the positioning of PICC catheters. By identifying the starting point, path, and end position of the catheter, it helps to determine whether the catheter has abnormal course. An objective position assessment result is generated based on the anatomical structure and catheter path, reducing the bias of subjective judgment. The generated PICC catheter tip position information text, which integrates visual features and text coding features, facilitates doctors to quickly understand the catheter position information.

[0103] Figure 3 The schematic diagram of the framework of the PICC positioning and review device based on chest X-ray according to an embodiment of the present invention is shown. The PICC positioning and review device based on chest X-ray comprises:

[0104] A data acquisition module is used to obtain the chest X-ray images of target patients through the hospital information system;

[0105] a model building module for establishing an X-ray chest film PICC positioning model to identify anatomical structures and PICC catheter paths in X-ray chest film images; wherein the X-ray chest film PICC positioning model includes an anatomical structure recognition unit, a catheter path recognition unit, and a position evaluation unit; the anatomical structure recognition unit is used to recognize anatomical structures in X-ray chest film images, wherein the anatomical structures include trachea, clavicle, ribs, spine, heart outline, and mediastinal anatomical landmarks; the catheter path recognition unit is used to recognize the PICC catheter path in X-ray chest film images, wherein the PICC catheter path includes the starting point, running path, and terminal position of the catheter; and the position evaluation unit is used to evaluate the terminal position of the catheter based on the anatomical structure and the PICC catheter path, and generate a position evaluation result;

[0106] The positioning review module is used to generate PICC catheter tip position information text that integrates visual features and text coding features according to the anatomical structure and PICC catheter path; and to review the PICC catheter positioning based on the PICC catheter tip position information text and the position evaluation results.

[0107] Figure 4 The schematic diagram of the structure of the computing device embodiment of the present invention is shown. The specific embodiment of the present invention does not limit the specific implementation of the computing device.

[0108] like Figure 4 As shown, the computing device may include: a processor 402 , a communications interface 404 , a memory 406 , and a communication bus 408 .

[0109] Processor 402, communication interface 404, and memory 406 communicate with each other via communication bus 408. Communication interface 404 is used to communicate with other devices, such as client devices or other server network elements. Processor 402 is used to execute program 410, which specifically performs the steps described in the aforementioned embodiment of the PICC location and review method based on chest X-rays.

[0110] Specifically, the program 410 may include program codes, which include computer operation instructions.

[0111] Processor 402 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The one or more processors included in a computing device may be processors of the same type, such as one or more CPUs, or processors of different types, such as one or more CPUs and one or more ASICs.

[0112] The memory 406 is used to store the program 410. The memory 406 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0113] According to the solution provided by the present invention, an X-ray chest image of a target patient is obtained through a hospital information system; a PICC positioning model for the X-ray chest image is established to identify the anatomical structure and PICC catheter path in the X-ray chest image; wherein the X-ray chest image PICC positioning model includes an anatomical structure recognition unit, a catheter path recognition unit, and a position evaluation unit; the anatomical structure recognition unit is used to recognize the anatomical structure in the X-ray chest image, the anatomical structure including the trachea, clavicle, ribs, spine, heart outline, and mediastinal anatomical landmarks; the catheter path recognition unit is used to recognize the PICC catheter path in the X-ray chest image, the PICC catheter path including the starting point, running path, and end position of the catheter; the position evaluation unit is used to evaluate the catheter end position based on the anatomical structure and the PICC catheter path to generate a position evaluation result; based on the anatomical structure and the PICC catheter path, a PICC catheter tip position information text is generated that integrates visual features and text coding features; and the PICC catheter positioning is reviewed based on the PICC catheter tip position information text and the position evaluation result. The present invention improves the accuracy and efficiency of PICC positioning review through the X-ray chest image PICC positioning model. Specifically, by identifying key anatomical structures such as the trachea, clavicle, ribs, spine, heart outline, and mediastinal anatomical landmarks, a reliable reference is provided for the positioning of PICC catheters. By identifying the starting point, path, and end position of the catheter, it helps to determine whether the catheter has abnormal course. An objective position assessment result is generated based on the anatomical structure and catheter path, reducing the bias of subjective judgment. The generated PICC catheter tip position information text, which integrates visual features and text coding features, facilitates doctors to quickly understand the catheter position information.

[0114] Those skilled in the art will appreciate that modules in the devices of the embodiments may be adaptively modified and deployed in one or more devices different from the embodiments. Modules, units, or components in the embodiments may be combined into a single module, unit, or component, and furthermore, they may be divided into multiple submodules, subunits, or subcomponents. All features disclosed in this specification (including the accompanying claims, abstract, and drawings), as well as all processes or units of any method or device disclosed therein, may be combined in any combination, except where at least some of such features and / or processes or units are mutually exclusive. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that provides the same, equivalent, or similar purpose. Furthermore, those skilled in the art will appreciate that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of the present invention and to form different embodiments. For example, in the claims below, any of the claimed embodiments may be used in any combination. The present invention may be implemented using hardware comprising a number of different elements and using a suitably programmed computer. In a unit claim that lists several means, several of these means may be embodied by the same hardware item. Unless otherwise specified, the steps in the above embodiments should not be understood as limiting the order of execution.

Claims

1. A PICC positioning review method based on chest X-ray, characterized in that: include: Obtain chest X-ray images of target patients through the hospital information system; Establishing an X-ray chest film PICC positioning model to identify anatomical structures and PICC catheter paths in X-ray chest film images; wherein the X-ray chest film PICC positioning model includes an anatomical structure recognition unit, a catheter path recognition unit, and a position evaluation unit; the anatomical structure recognition unit is used to recognize anatomical structures in X-ray chest film images, wherein the anatomical structures include trachea, clavicle, ribs, spine, heart outline, and mediastinal anatomical landmarks; the catheter path recognition unit is used to recognize the PICC catheter path in X-ray chest film images, wherein the PICC catheter path includes the starting point, running path, and terminal position of the catheter; the position evaluation unit is used to evaluate the terminal position of the catheter based on the anatomical structure and the PICC catheter path, and generate a position evaluation result; Based on the anatomical structure and PICC catheter path, a local image of the PICC catheter tip position in the X-ray chest image is extracted, and the visual features of the local image are extracted using the ResNet network to obtain a visual feature vector; a position description text is generated based on the position evaluation result, and the position description text is encoded using the BERT model to obtain a text encoding vector; the visual feature vector and the text encoding vector are spliced ​​together to generate a fusion vector to obtain the PICC catheter tip position information text; and the PICC catheter positioning is reviewed based on the PICC catheter tip position information text and the position evaluation result.

2. The PICC positioning review method based on chest X-ray according to claim 1, characterized in that: The anatomical structure recognition unit uses skip connections to fuse features of corresponding levels in the encoder and outputs a pixel-level segmentation mask for each anatomical structure; wherein the pixel values ​​of the segmentation mask are all 0 or 1, 0 represents background and 1 represents anatomical structure; The precision and recall of the segmentation mask of each anatomical structure are calculated, and the F1 score is used as the evaluation indicator for each anatomical structure.

3. The PICC positioning review method based on chest X-ray according to claim 1, characterized in that: The catheter path identification unit uses a multi-scale edge response algorithm to identify the catheter edge; The straight line segments in the catheter edge are detected by Hough transform algorithm and optimized by dynamic program method. The bidirectional LSTM network is used to process each straight line segment to obtain the coordinates of the starting point, intermediate nodes, and end position of the PICC catheter. According to the coordinates of the starting point, middle node and end position of the PICC catheter, the PICC catheter path is connected and generated, and the PICC catheter path is smoothed by B-spline curve.

4. The PICC positioning review method based on chest X-ray according to claim 1, characterized in that: The assessing of the position of the catheter tip according to the anatomical structure and the PICC catheter path further comprises: Calculate the distance between the PICC catheter tip and the entrance of the superior vena cava; If the distance is less than a preset threshold, the end position is located in the superior vena cava; wherein the superior vena cava entry point is determined by the relative positions of the trachea mask and the clavicle mask; If the catheter tip is located in the right atrium, the tip is located in the right atrium; if the catheter tip is located in the right ventricle, the tip is located in the right ventricle; wherein the ranges of the right atrium and the right ventricle are determined by the heart contour mask.

5. The PICC positioning review method based on chest X-ray according to claim 1, characterized in that: The review of PICC catheter positioning based on the PICC catheter tip position information text and position assessment results further includes: Assign different confidence scores to the classification of each evaluation result to obtain the confidence score of the evaluation result; The PICC catheter tip location information text and location evaluation results are input into the multi-layer perceptron network, which outputs the location category of the PICC catheter tip and the corresponding confidence score; The confidence score of the assessment result and the confidence score of the PICC catheter tip position category were weighted and summed to obtain the review confidence score.

6. The PICC positioning review method based on chest X-ray according to claim 1, characterized in that: The step of extracting the visual features of the local image using the ResNet network to obtain the visual feature vector further includes: Use the ResNet network as the basic feature extractor and remove the fully connected layer of the ResNet network; Introducing an attention mechanism module to weight the extracted feature maps, wherein the attention mechanism module includes at least one convolutional layer and a sigmoid activation layer; The weighted feature map is subjected to global average pooling to obtain a visual feature vector.

7. The PICC positioning review method based on chest X-ray according to claim 5, characterized in that: Before the input layer of the multi-layer perceptron, a Transformer encoder layer is introduced to capture the contextual dependency between the PICC catheter tip position information text and the position assessment result; A cross-modal fusion module is introduced between the output of the Transformer encoder layer and the input layer of the multi-layer perceptron network to concatenate the fused feature vector of the text output and the evaluation result of the Transformer encoder with the global visual feature vector. The Transformer encoder layer includes: The embedding layer maps the word vectors of the PICC catheter tip location information text and the category one-hot encoding vectors of the location assessment results into a high-dimensional space; Position encoding layer, which adds position information to the embedded vector to preserve the sequence order; Multiple interleaved and stacked multi-head self-attention sublayers and feedforward neural network sublayers, wherein the multi-head self-attention sublayer uses a self-attention mechanism to calculate the correlation between input vectors, and the feedforward neural network sublayer performs nonlinear transformation.

8. A PICC positioning and review device based on chest X-ray, characterized in that: include: A data acquisition module is used to obtain the chest X-ray images of target patients through the hospital information system; A model building module is used to establish an X-ray chest film PICC positioning model to identify the anatomical structure and PICC catheter path in the X-ray chest film image; wherein the X-ray chest film PICC positioning model includes an anatomical structure recognition unit, a catheter path recognition unit, and a position evaluation unit; the anatomical structure recognition unit is used to recognize the anatomical structure in the X-ray chest film image, wherein the anatomical structure includes the trachea, clavicle, ribs, spine, heart outline, and mediastinal anatomical landmarks; the catheter path recognition unit is used to recognize the PICC catheter path in the X-ray chest film image, wherein the PICC catheter path includes the starting point, running path, and terminal position of the catheter; the position evaluation unit is used to evaluate the terminal position of the catheter based on the anatomical structure and the PICC catheter path, and generate a position evaluation result; The positioning review module is used to extract a local image of the PICC catheter tip position in the X-ray chest image based on the anatomical structure and the PICC catheter path, and use the ResNet network to extract the visual features of the local image to obtain a visual feature vector; generate a position description text based on the position evaluation result, and encode the position description text through the BERT model to obtain a text encoding vector; splice the visual feature vector and the text encoding vector to generate a fusion vector to obtain the PICC catheter tip position information text; and review the PICC catheter positioning based on the PICC catheter tip position information text and the position evaluation result.

9. A computing device comprising: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the PICC positioning review method based on chest X-ray according to any one of claims 1 to 7.

Images