Intelligent assessment and extravasation early warning system for chemotherapy patient peripheral intravenous infusion site images

Abstract

This invention discloses an intelligent assessment and extravasation warning system for peripheral intravenous infusion sites in chemotherapy patients, relating to the fields of intelligent medical image analysis and clinical nursing assistance. The system includes an infusion site image acquisition module, an image preprocessing module, a skin condition analysis module, an extravasation identification module, a risk grading module, and a warning push module. It periodically acquires skin images of the puncture site using a miniature camera. After illumination correction and skin color standardization, a deep learning network is used to extract skin color, swelling degree, and texture change features. It identifies extravasation signs such as redness, paleness, and swelling and calculates confidence levels. Combined with the chemotherapy drug irritation grading, a risk-weighted assessment is performed, and warnings for medium- and high-risk events are pushed to the nurse's mobile terminal in real time.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical image analysis and clinical nursing assistance technology, specifically to an intelligent assessment and extravasation early warning system for images of peripheral intravenous infusion sites in chemotherapy patients. Background Technology

[0002] In the field of cancer treatment today, chemotherapy remains one of the most important methods for treating malignant tumors. According to the World Health Organization, there are more than 19 million new cancer cases worldwide each year, a significant proportion of whom require chemotherapy. Chemotherapy drugs are typically administered via central venous catheterization and peripheral venous puncture. Peripheral venous puncture is widely used in short-course and outpatient chemotherapy settings due to its advantages such as ease of operation, minimal invasiveness, and the absence of special catheter maintenance.

[0003] However, extravasation of chemotherapy drugs during peripheral intravenous infusion is a serious complication that cannot be ignored. Extravasation refers to the leakage of chemotherapy drugs from the blood vessel into the surrounding tissues. Because many chemotherapy drugs have strong cytotoxicity and tissue irritation, extravasation can cause mild symptoms such as local skin redness, swelling, pain, and blistering, or even deep tissue necrosis, tendon damage, and the need for surgical debridement. Literature reports an extravasation rate of approximately 0.1% to 6%. Although this absolute percentage seems low, considering the large number of chemotherapy patients and the severity of the consequences of extravasation, prevention and early detection of extravasation are of significant clinical importance.

[0004] Traditional monitoring of chemotherapy extravasation relies primarily on regular visual rounds by nursing staff. Typically, nurses check the puncture site at regular intervals during infusion for any abnormalities such as redness, swelling, or pain. However, this manual monitoring method has significant limitations: First, nurses have heavy workloads and limited rounds, making continuous close observation of each patient difficult; second, early symptoms of extravasation are often subtle, manifesting only as mild discomfort at the puncture site or slight changes in skin color, easily missed by the naked eye; third, some patients, due to the neurotoxicity of chemotherapy drugs or their own diminished senses, are unable to promptly report any discomfort. These factors collectively lead to delayed detection of extravasation, and the degree of tissue damage after extravasation is closely related to the amount of drug leakage and the duration of exposure; delayed detection directly exacerbates tissue damage.

[0005] Several technologies have been developed to address the issue of extravasation monitoring. For example, Chinese patent application CN119444760A discloses an image-based PICC intravenous therapy information monitoring system. This system determines the presence of infection signs at the puncture site by performing temporal analysis on high-resolution images of the patient's puncture site at different time points within a target monitoring period. It employs techniques such as convolutional neural networks with deep and superficial feature fusion modules, three-dimensional convolutional kernels, and channel attention mechanisms. This solution primarily targets the monitoring of PICC catheter-related infections, with its key technical focus on capturing temporal changes and highlighting the features of infection signs.

[0006] However, chemotherapy extravasation and catheter-related infections differ significantly in their pathological mechanisms, clinical manifestations, and the urgency of management. Chemotherapy extravasation has the following unique characteristics: First, different types of chemotherapy drugs exhibit vastly different tissue-damaging abilities. Extravasation of vesicant drugs, such as anthracyclines and vincristine, has far more severe consequences than that of non-irritant drugs; therefore, extravasation risk assessment must be combined with specific drug characteristics. Second, early signs of extravasation are diverse and often subtle, potentially manifesting as local redness, pallor, swelling, or only slight changes in skin temperature, requiring targeted multi-sign recognition capabilities. Third, the time window for intervention after extravasation is extremely limited; the golden time for treating vesicant extravasation is typically within several minutes to tens of minutes, demanding extremely high real-time early warning capabilities. Existing technical solutions fail to fully consider these unique needs of chemotherapy extravasation, lacking a fusion mechanism of drug irritation grading and image analysis, making it difficult to meet the practical needs of continuous monitoring of chemotherapy infusion safety and early intervention for extravasation.

[0007] Therefore, there is an urgent need for an image-based intelligent assessment and extravasation warning solution specifically designed for peripheral intravenous infusion scenarios in chemotherapy patients. Summary of the Invention

[0008] To address the aforementioned technical problems, this invention provides an intelligent image assessment and extravasation early warning system for peripheral intravenous infusion sites in chemotherapy patients, aiming to achieve intelligent continuous monitoring and graded early warning of extravasation events during chemotherapy infusion.

[0009] According to one aspect of the present invention, an intelligent assessment and extravasation warning system for peripheral intravenous infusion sites in chemotherapy patients is provided, comprising: an infusion site image acquisition module for acquiring skin image data of the infusion site using a miniature camera fixed to the peripheral venous puncture site of the patient according to a preset sampling period; an image preprocessing module for performing illumination correction and skin color standardization processing on the skin image data of the infusion site to obtain standardized skin image data; and a skin state analysis module for using a deep learning segmentation network to identify the skin region around the puncture point in the standardized skin image data, and automatically extracting skin color features, swelling degree features, and texture change features, and performing temporal fusion of the extracted multidimensional features to obtain skin state features. The system includes: an extravasation identification module, used to perform feature matching and analysis on local redness, pallor, swelling, and skin temperature changes based on skin condition feature vectors, calculate the confidence level of each sign, and synthesize the extravasation sign confidence level value; a risk grading module, used to obtain the irritation level data of the currently infused chemotherapy drug, perform weighted fusion calculation on the extravasation sign confidence level value and the irritation level data, and output the extravasation risk level result based on the fusion result; and an early warning push module, used to determine whether the extravasation risk level result meets the preset alarm conditions, generate an early warning message when the alarm conditions are met, and immediately push it to the responsible nurse's mobile terminal, and simultaneously package the skin image data of the current time and the previous time to generate an image evidence data package for storage.

[0010] Furthermore, the skin state analysis module uses a semantic segmentation network with an encoder-decoder structure to process the standardized skin image data and generate a puncture site region segmentation feature map, which is used to identify the skin region of interest within a preset radius around the puncture point.

[0011] Furthermore, the skin condition analysis module extracts skin color feature vectors, swelling degree feature vectors, and texture change feature vectors from the region of interest defined by the puncture site region segmentation feature icon, thereby achieving multi-dimensional skin condition representation.

[0012] Furthermore, the skin condition analysis module also includes a temporal difference analysis unit, which is used to calculate the pixel-level difference between standardized skin image data at adjacent sampling time points to obtain a temporal difference feature map, thereby enhancing the ability to capture dynamic changes.

[0013] Furthermore, the risk grading module determines the drug irritation weighting coefficient based on the irritation classification of chemotherapy drugs, thereby realizing the correlation assessment between drug characteristics and extravasation risk.

[0014] Furthermore, when generating image evidence data packets, the early warning push module arranges the skin image data at the current moment and all skin image data within a preset time window before the current moment in chronological order and adds timestamps and extravasation sign annotations to provide complete imaging evidence for subsequent clinical treatment.

[0015] Compared with existing technologies, the intelligent image assessment and extravasation early warning system for peripheral intravenous infusion sites in chemotherapy patients provided by this invention has the following beneficial effects: First, by integrating the irritation grading information of chemotherapy drugs into the extravasation risk assessment process, the risk assessment is made more precise and individualized, and extravasation events of vesicant drugs can receive a higher priority early warning response; Second, the technical approach of combining multi-sign parallel recognition with temporal differential analysis significantly improves the detection sensitivity of early signs of extravasation, which helps to detect and intervene in time before tissue damage worsens; Third, the automatic image evidence packaging and storage function provides an objective basis for subsequent clinical treatment decisions and medical quality traceability; Fourth, the mobile terminal real-time early warning push mechanism greatly shortens the nursing response time and helps to grasp the golden time window for extravasation treatment. Attached Figure Description

[0016] Figure 1 This is an overall architecture diagram of the intelligent image assessment and extravasation early warning system for peripheral intravenous infusion sites in chemotherapy patients according to an embodiment of the present invention.

[0017] Figure 2 This is a detailed structural diagram of the skin condition analysis module according to an embodiment of the present invention.

[0018] Figure 3 This is a detailed structural diagram of the risk classification module according to an embodiment of the present invention.

[0019] Figure 4 This is a detailed structural diagram of the early warning push module according to an embodiment of the present invention. Detailed Implementation

[0020] Exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be understood that the described embodiments are merely some embodiments of the present invention, and not all embodiments of the present invention, and the present invention is not limited to the exemplary embodiments described herein.

[0021] This invention integrates deep learning image analysis technology with chemotherapy nursing expertise to construct a complete intelligent early warning framework for extravasation.

[0022] like Figure 1As shown, the intelligent assessment and extravasation warning system 100 for peripheral intravenous infusion sites in chemotherapy patients includes: an infusion site image acquisition module 1, used to acquire skin image data of the infusion site using a miniature camera fixed to the patient's peripheral intravenous puncture site according to a preset sampling period; an image preprocessing module 2, used to perform illumination correction and skin color standardization processing on the skin image data of the infusion site to obtain standardized skin image data; a skin state analysis module 3, used to identify the skin area around the puncture point using a deep learning segmentation network on the standardized skin image data, and automatically extract skin color features, swelling degree features, and texture change features, and perform temporal fusion of the extracted multidimensional features to obtain a skin state feature vector; extravasation identification; and other functions. Module 4 is used to perform feature matching and analysis on local redness, pallor, swelling, and skin temperature changes based on skin condition feature vectors, calculate the confidence level of each sign, and synthesize the confidence level value of extravasation. Module 5 is used to obtain the irritation level data of the currently infused chemotherapy drug, perform weighted fusion calculation on the extravasation sign confidence level value and the irritation level data, and output the extravasation risk level result based on the fusion result. Module 6 is used to determine whether the extravasation risk level result meets the preset alarm conditions. When the alarm conditions are met, an alarm message is generated and immediately pushed to the responsible nurse's mobile terminal. At the same time, the skin image data of the current moment and the previous moment are packaged into an image evidence data package for storage.

[0023] In this embodiment of the invention, the infusion site image acquisition module 1 is used to acquire skin image data of the infusion site using a miniature camera fixed to the patient's peripheral venous puncture site according to a preset sampling cycle. In one embodiment of the invention, the miniature camera uses a CMOS image sensor with a resolution of not less than 2 million pixels, and the lens field of view is set between 60 and 90 degrees to ensure complete coverage of the skin area with a diameter of approximately 5 cm to 8 cm around the puncture point. Preferably, the camera housing is made of medical-grade silicone material and is attached to the patient's forearm or back of hand about 2 cm to 4 cm above the puncture site using an adjustable fixing strap.

[0024] In one embodiment of the present invention, the preset sampling period is a configurable parameter, and the system's default basic sampling period is set to 30 seconds, meaning that one skin image is collected every 30 seconds. When the confidence value of extravasation signs exceeds a preset monitoring threshold, the system automatically switches to encrypted sampling mode, shortening the sampling period to 10 seconds or less, in order to more closely track the trend of skin condition changes. In another embodiment of the present invention, a shorter initial sampling period, such as 15 seconds, is used during the first 15 minutes of infusion to enhance the monitoring of this high-risk period for extravasation during the initial infusion stage.

[0025] The acquired skin images of the infusion site are RGB three-channel color images, preferably with a resolution of 640×480 pixels or higher. Each image frame includes a timestamp for subsequent time-series analysis. The image data is transmitted to the image preprocessing module 2 via wired or wireless means for further processing.

[0026] In this embodiment of the invention, the image preprocessing module 2 is used to perform illumination correction and skin color standardization on the skin image data of the infusion site to obtain standardized skin image data. Because lighting conditions in the clinical environment vary considerably—natural light in the ward, bedside lamps, reflected light from the corridor, etc.—the color presentation of the acquired images can be affected. Since skin color changes are an important basis for judging extravasation, illumination correction and skin color standardization of the original images are crucial steps to ensure the accuracy of subsequent analysis.

[0027] The illumination correction process includes two sub-steps: white balance adjustment and histogram equalization. White balance adjustment employs a gray-world algorithm, assuming that the average value of all colors in the image scene should be gray. This is achieved by calculating the average value of the RGB channels of the image and then scaling it proportionally to correct color cast. Let the t-th sampling time point of the original image be... The average values ​​of its RGB three channels are respectively , , Image after white balance correction The pixel values ​​for each channel are calculated using the following formula:

[0028] ,

[0029] in: This represents the color channel, and its value can be one of R, G, or B. Indicates the original image in channels Pixel values ​​on; Indicates the original image in channels The average value above; select the green channel. The reference is used because the human eye is most sensitive to green, and using it as a benchmark for correction conforms to the characteristics of human visual perception.

[0030] Histogram equalization is used to enhance image contrast and improve the visibility of skin texture details. This embodiment of the invention employs a contrast-limited adaptive histogram equalization method, dividing the image into 8×8 sub-blocks and performing histogram equalization independently on each sub-block. The contrast limit threshold is set to 2.0 to avoid excessive noise amplification.

[0031] Skin color standardization employs a color space mapping method based on a reference skin color template. The system pre-stores standard reference templates for different skin color types. When a patient first uses the system, a frame of normal skin before puncture is captured as an individualized baseline. Subsequent captured images are color-mapped using this baseline image as a reference, ensuring color consistency between images at different time points. Specifically, the white-balance corrected image is converted to the LAB color space, the mean difference between the current image and the baseline image in the L, A, and B channels is calculated, and offset compensation is performed.

[0032] ,

[0033] ,

[0034] ,

[0035] in: This represents standardized skin image data; , , These represent the average values ​​of the three channels of the current image in the LAB color space; , , These represent the mean values ​​of the three channels of the reference image in the LAB color space. Through the above offset compensation, the overall color shift caused by changes in illumination is eliminated, so that subsequent skin color change analysis can truly reflect changes in physiological state rather than interference from environmental factors.

[0036] In this embodiment of the invention, the skin state analysis module 3 is used to identify the skin region around the puncture point using a deep learning segmentation network on standardized skin image data, and automatically extract skin color features, swelling degree features, and texture change features. The extracted multidimensional features are then fused temporally to obtain a skin state feature vector. Figure 2As shown, the skin state analysis module 3 includes: a region segmentation unit 31, used to perform semantic segmentation of the skin region around the puncture point on standardized skin image data to obtain a puncture site region segmentation feature map; a color feature extraction unit 32, used to extract skin color feature vectors within the region of interest defined by the puncture site region segmentation feature map; a swelling feature extraction unit 33, used to extract swelling degree feature vectors within the region of interest defined by the puncture site region segmentation feature map; a texture feature extraction unit 34, used to extract texture change feature vectors within the region of interest defined by the puncture site region segmentation feature map; a temporal difference analysis unit 35, used to calculate the pixel-level difference between standardized skin image data at adjacent sampling time points to obtain a temporal difference feature map; and a feature fusion unit 36, used to concatenate and fuse the skin color feature vector, swelling degree feature vector, texture change feature vector, and temporal difference feature map to obtain a skin state feature vector.

[0037] Region segmentation unit 31 uses a semantic segmentation network with an encoder-decoder structure to process standardized skin image data. In one embodiment of the invention, the encoder part uses a pre-trained ResNet-34 network as the feature extraction backbone, encoding the input image into a multi-scale feature representation through continuous convolution and downsampling operations; the decoder part adopts a structure combining layer-by-layer upsampling and skip connections, fusing high-level semantic features with low-level spatial detail features, and finally outputting a segmentation mask of the same size as the input image. After being trained with labeled data, the segmentation network can automatically identify semantic categories such as puncture point locations, indwelling needle patch edges, and exposed skin areas in the image.

[0038] Puncture site region segmentation feature map The method uses a binary mask, where pixels with a value of 1 correspond to the skin region of interest within a preset radius around the puncture point, and pixels with a value of 0 correspond to the background area or a non-interest area. Preferably, the region of interest is defined as the exposed skin portion within a circular area with a radius of 2.5 cm centered at the puncture point, excluding non-skin areas such as indwelling needle dressings, adhesive tape, and vascular markings. Through region segmentation, subsequent feature extraction is performed only within the region of interest, avoiding background interference and improving the specificity and effectiveness of the features.

[0039] The color feature extraction unit 32 extracts the skin color feature vector within the region of interest (ROI) defined by the feature map of the puncture site. In this embodiment, the color feature extraction includes two steps: first, the pixels within the ROI are converted to the LAB color space; then, the color histograms for the L, A, and B channels are calculated, with each channel divided into 16 intervals, resulting in a 48-dimensional color histogram feature vector. Furthermore, the mean and standard deviation of the pixels within the ROI in the A and B channels are calculated as statistical features reflecting the overall color tendency and the dispersion of color distribution, and these are concatenated with the color histogram features to form the skin color feature vector. It has 52 dimensions.

[0040] In one embodiment of the present invention, an increase in the mean value of channel A suggests reddish skin, which may indicate local congestion or inflammation; a decrease in the mean value of channel A suggests greenish / pale skin, which may indicate local ischemia; and changes in the mean value of channel B are related to yellowing or cyanosis of the skin. By monitoring the trends of these color statistics over time, abnormal skin color in the early stages of extravasation can be detected.

[0041] The swelling feature extraction unit 33 extracts the swelling degree feature vector within the region of interest defined by the puncture site region segmentation feature icon. After extravasation occurs, the leaked fluid accumulates in the interstitial space, causing local skin swelling and bulging, manifested as changes in skin contour and an increase in local area. In this embodiment of the invention, the swelling degree feature extraction includes the following steps:

[0042] First, edge contour detection is performed on the region of interest. The Canny edge detection algorithm is used to extract the light and dark transition boundaries of the skin surface, which reflect the undulating shape of the skin surface. Let the set of edge pixels extracted at the t-th sampling time point be denoted as . The total number of edge pixels is .

[0043] Next, calculate the rate of change of the area of ​​the region of interest. Let the area of ​​the region of interest at the reference time be... (Number of pixels), the area of ​​the region of interest at the current time is The formula for calculating the area change rate is:

[0044] ,

[0045] in: This represents the rate of change of area at the t-th sampling time point relative to the baseline time point, with a value ranging from -100% to positive infinity, expressed as a percentage; a positive value indicates an increase in area, i.e., swelling, while a negative value indicates a decrease in area. In one embodiment of the present invention, a rate of change of area exceeding 5% is considered a clinically significant indication of swelling.

[0046] Next, the contour complexity index is calculated. The edges of swollen skin are smoother and more rounded than those of normal skin; this invention uses a roundness index to quantify this characteristic.

[0047] ,

[0048] in: This represents the perimeter (in pixels) of the region of interest; the roundness value ranges from 0 to 1, with the value closer to 1 indicating a more circular shape. Swelling areas typically exhibit a higher roundness.

[0049] Finally, the edge pixel density Area change rate Circularity The swelling degree feature vector is formed by concatenating features such as the offset distance of the centroid of the region of interest relative to the puncture point. It has 8 dimensions.

[0050] The texture feature extraction unit 34 extracts texture change feature vectors within the region of interest defined by the segmentation feature icon at the puncture site. Skin texture reflects the microscopic morphology of the skin surface; tissue edema caused by extravasation can blur or eliminate skin texture. In this embodiment of the invention, texture feature extraction employs a combination of two methods: gray-level co-occurrence matrix and local binary pattern.

[0051] The gray-level co-occurrence matrix (GLCM) reflects the frequency of occurrence of pixel pairs with a certain gray-level relationship in an image. In one embodiment of the present invention, GLCMs in four directions (0 degrees, 45 degrees, 90 degrees, and 135 degrees) are calculated for the gray-level image of the region of interest, with the displacement distance set to 1 pixel and the gray-level quantization to 16 levels. Four statistical features—contrast, correlation, energy, and homogeneity—are extracted from each GLCM, resulting in a total of 16-dimensional GLCM features across the four directions.

[0052] Local binary patterning is an effective method for describing local texture. For each pixel within a region of interest, its grayscale relationship with its eight neighboring pixels is compared to generate an 8-bit binary code. The frequency of various coding patterns is then statistically analyzed to form a local binary pattern histogram. This invention employs rotation-invariant uniform local binary patterning, with a histogram dimension of 10.

[0053] The gray-level co-occurrence matrix features and the local binary pattern histogram are concatenated to form a texture variation feature vector. The dimension is 26. After extravasation occurs, as tissue edema intensifies, skin texture gradually becomes blurred, manifested as decreased contrast and increased energy in the gray-level co-occurrence matrix, and a decrease in the entropy value of the local binary mode histogram.

[0054] The temporal difference analysis unit 35 is used to calculate the pixel-level difference between standardized skin image data at adjacent sampling time points to obtain a temporal difference feature map. Extravasation is a dynamic process, and the temporal trend of skin condition changes is an important basis for judging whether extravasation has occurred. Through temporal difference analysis, the regions where changes occur between two consecutive frames can be highlighted, which helps to detect weak signs of extravasation at an early stage.

[0055] Let the current time be t and the previous sampling time be t-1. The formula for calculating the temporal difference feature map is:

[0056] ,

[0057] in: and These represent the standardized skin image data at the current and previous moments, respectively. This represents the absolute value operation, which calculates the absolute value of the difference between the grayscale values ​​of two frames for each pixel location. In the temporal difference feature map, regions with larger values ​​correspond to areas where significant changes occur between the two frames.

[0058] In one embodiment of the present invention, to reduce pseudo-differential responses caused by factors such as camera shake and slight patient movement, image registration based on feature point matching is performed on two adjacent frames before calculating the difference, ensuring that the same skin location corresponds to the same pixel coordinates in the two frames. Preferably, a registration method combining ORB feature point detection and RANSAC robust estimation is used, with registration accuracy controlled within 2 pixels.

[0059] The feature fusion unit 36 ​​is used to concatenate and fuse the skin color feature vector, swelling degree feature vector, texture change feature vector, and temporal difference feature map to obtain the skin state feature vector. In one embodiment of the present invention, the temporal difference feature map is first... Feature encoding is performed using a small convolutional neural network containing three convolutional layers with 3×3 kernels and 16, 32, and 64 channels respectively. Finally, global average pooling is used to obtain a 64-dimensional difference feature vector. Then, the skin color feature vector... (52-dimensional) swelling degree feature vector (8-dimensional) Texture variation feature vector (26-dimensional) and difference eigenvectors (64-dimensional) splicing to form a skin state feature vector:

[0060] ,

[0061] in: Represents vector concatenation operation; skin state feature vector The total dimensions are 150, including multi-dimensional information such as skin color, swelling degree, texture and temporal changes, providing a comprehensive feature representation for subsequent extravasation identification.

[0062] In this embodiment of the invention, the extravasation identification module 4 is used to perform feature matching and analysis on local redness, pallor, swelling, and skin temperature changes based on skin state feature vectors, calculate the confidence level of each sign, and synthesize the confidence level values ​​of the extravasation signs. Early clinical manifestations of chemotherapy drug extravasation are diverse, and signs such as redness, pallor, swelling, and skin temperature changes may appear individually or in combination. The probability and clinical significance of different signs also vary. This invention uses a multi-label classification network to independently evaluate each sign, and then synthesizes the confidence levels of each sign to obtain the overall confidence level of the extravasation event.

[0063] In one embodiment of the present invention, the extravasation identification module 4 adopts a neural network structure including a shared feature layer and a multi-task output layer. The shared feature layer consists of two fully connected layers, with a 150-dimensional skin state feature vector as input and a 64-dimensional hidden layer in the middle, using ReLU as the activation function. The multi-task output layer contains four independent classification heads, corresponding to signs of redness, pallor, swelling, and skin temperature changes, respectively. Each classification head outputs a confidence value between 0 and 1. The independent confidence values ​​for each sign are denoted as follows: , , , .

[0064] Confidence level of extravasation signs The following is obtained by weighted summation of the independent confidence levels of each feature:

[0065] ,

[0066] in: , , , The weighting coefficients for redness, pallor, swelling, and changes in skin temperature reflect the contribution of each sign to the assessment of extravasation. In one embodiment of the present invention, the weighting coefficients are set with reference to clinical experience and extravasation diagnosis guidelines, with redness having a specific weight. Pale signs weight Swelling symptoms weight Weighting of signs of skin temperature change The sum of the four weights is 1.0, ensuring the confidence level of the extravasation symptom. The value range is from 0 to 1.

[0067] Redness and swelling are given high weight because they are the most typical and common clinical manifestations of chemotherapy extravasation and have high specificity for diagnosing extravasation. Paleness usually appears in the early stages of extravasation of vesicant drugs, indicating local vasoconstriction and tissue ischemia; it has predictive value but slightly lower specificity. Changes in skin temperature require auxiliary infrared thermal imaging equipment for accurate detection; its accuracy is limited in scenarios relying solely on visible light image analysis, therefore its weight is set relatively low.

[0068] In this embodiment of the invention, the risk grading module 5 is used to obtain the irritation level data of the currently infused chemotherapy drug, perform a weighted fusion calculation between the extravasation sign confidence value and the irritation level data, and output the extravasation risk level result based on the fusion result. Figure 3 As shown, the risk grading module 5 includes: a drug information acquisition unit 51, used to acquire the name and irritation level data of the currently infused chemotherapy drug from the hospital information system or through manual input; an irritation weighting unit 52, used to determine the drug irritation weighting coefficient according to the irritation classification of the chemotherapy drug; a risk score calculation unit 53, used to multiply the extravasation sign confidence value with the drug irritation weighting coefficient and combine it with a preset benchmark score to calculate the extravasation comprehensive risk score value; and a risk level determination unit 54, used to compare the extravasation comprehensive risk score value with a preset risk threshold to output the extravasation risk level result.

[0069] The drug information acquisition unit 51 acquires the irritation level data of the currently infused chemotherapy drug. Chemotherapy drugs can be divided into three categories according to their ability to damage tissues: vesicants, irritants, and non-irritants. Vesicants can cause severe tissue necrosis after extravasation; typical drugs include anthracyclines and vinca alkaloids such as doxorubicin, epirubicin, vincristine, and vinorelbine. Irritants cause local inflammation and pain after extravasation, but usually do not cause tissue necrosis; typical drugs include carboplatin, cisplatin, and cyclophosphamide. Non-irritants cause mild tissue damage after extravasation; typical drugs include fluorouracil and methotrexate. In one embodiment of the present invention, the system has a built-in chemotherapy drug irritation classification database, covering more than 50 commonly used chemotherapy drugs in clinical practice. When the nurse starts extravasation monitoring, she selects or enters the name of the currently infused drug, and the system automatically matches the corresponding irritation level.

[0070] Irritation-weighted unit 52 determines the drug irritation-weighted coefficient based on the irritation classification of chemotherapy drugs. In one embodiment of the present invention, the weighted coefficient for drug irritation is set as follows: for vesicant drugs... Stimulant drugs Non-irritant drugs Ectomycing agents were assigned the highest weighting coefficient, reflecting the severity and urgency of their extravasation consequences; stimulant agents were assigned a moderate weighting coefficient; and non-stimulant agents were assigned the lowest weighting coefficient, but still greater than zero to ensure the system remains sensitive to extravasation of any type of drug.

[0071] The risk scoring calculation unit 53 multiplies the confidence level of extravasation signs by the drug irritation weighting coefficient and combines it with a preset benchmark score to calculate the comprehensive extravasation risk score. The calculation formula is as follows:

[0072] ,

[0073] in: This represents the comprehensive risk score for external leakage, with a value ranging from 0 to 100, and is dimensionless. This represents a preset benchmark score, which is set to 10 points in one embodiment of the present invention, reflecting the inherent basic risks of chemotherapy infusion itself. This represents the irritation weighting factor of the currently infused drug, with values ​​of 0.8, 1.2, or 2.0. This represents the confidence level of the extravasation sign, with a value ranging from 0 to 1; This represents the rating scaling factor, used to map the weighted confidence level to the target rating interval. In one embodiment of this invention, it is set to 45 to ensure that when... and hour, It can reach 100 points.

[0074] The risk level determination unit 54 compares the comprehensive risk score of extravasation with a preset risk threshold to output the extravasation risk level result. In one embodiment of the present invention, two risk thresholds are set: a first preset risk threshold... Second preset risk threshold .when When the risk level is low, it indicates that the current skin condition is normal or has only minor abnormalities, and no immediate intervention is required; when... When the output indicates a medium-risk level, it suggests the presence of suspected extravasation, and nurses are advised to perform bedside checks as soon as possible; when When a high-risk level is output, it indicates a high probability of extravasation, requiring immediate bedside assessment and preparation for extravasation treatment.

[0075] In this embodiment of the invention, the early warning push module 6 is used to determine whether the extravasation risk level result has reached the preset alarm condition. When the alarm condition is reached, an early warning message is generated and immediately pushed to the responsible nurse's mobile terminal. Simultaneously, the skin image data of the current moment and the previous moment are packaged into an image evidence data package for storage. Figure 4As shown, the early warning push module 6 includes: an alarm condition determination unit 61, used to determine whether the extravasation risk level result has reached the preset alarm condition; an early warning information generation unit 62, used to generate early warning information containing patient information, bed number, risk level, and description of extravasation signs when the alarm condition is reached; a mobile terminal push unit 63, used to push the early warning information to the responsible nurse's mobile terminal in real time; an image evidence packaging unit 64, used to package the skin image data of the current time and the previous time to generate an image evidence data package; and an evidence storage unit 65, used to store the image evidence data package to the server.

[0076] The alarm condition determination unit 61 determines whether the leakage risk level result meets the preset alarm condition. In one embodiment of the present invention, the preset alarm condition is that the risk level reaches medium or high risk. When the risk level is low risk, the system only performs routine monitoring and recording, and does not trigger alarm push. When the risk level upgrades from low risk to medium risk for the first time, a level 1 alarm is triggered; when the risk level reaches high risk or upgrades from medium risk to high risk, a level 2 alarm is triggered; when the risk level remains at high risk for more than 30 seconds, a level 3 emergency alarm is triggered. Different alarm levels correspond to different push methods and prompt intensities.

[0077] The early warning information generation unit 62 generates early warning information when alarm conditions are met. The early warning information includes: patient name, hospital number, bed number, name of currently infused medication, extravasation risk level, description of major extravasation signs, risk score, and recommended treatment measures. In one embodiment of the invention, the description of major extravasation signs is automatically generated based on the independent confidence level of each sign; for example, when... The description was redness of the skin around the puncture site. It is described as swelling of the skin around the puncture site. When multiple signs have high confidence, they are combined for description.

[0078] The mobile terminal push unit 63 instantly pushes the warning information to the mobile terminal of the responsible nurse. In one embodiment of the present invention, the system interfaces with the hospital nursing information system, automatically determines the responsible nurse on duty based on the nursing schedule, and pushes the warning information via a dedicated nursing APP or SMS. Level 1 alarms are pushed in the form of ordinary messages, Level 2 alarms are pushed in the form of high-priority messages with an alert sound, and Level 3 emergency alarms are pushed with the highest priority and trigger continuous vibration reminders, while simultaneously sending an alarm to the fixed terminal of the nurse on duty at the nursing station.

[0079] The image evidence packaging unit 64 packages skin image data from the current moment and previous moments to generate an image evidence data package. In one embodiment of the present invention, the image evidence data package includes the skin image at the current moment and all skin images within a preset time window prior to the current moment. The preset time window is preferably 5 minutes, that is, all image frames acquired within the most recent 5 minutes are packaged. The images are arranged in chronological order, and each image frame is accompanied by the following information: acquisition timestamp, confidence value of extravasation signs, risk score value, independent confidence value of each sign, and region of interest bounding box. The image evidence data package provides objective radiological evidence for subsequent clinical treatment decisions and medical quality traceability. Nurses and doctors can intuitively understand the changes in skin condition by viewing the image sequence.

[0080] The evidence storage unit 65 stores the image evidence data packets to the server. In one embodiment of the present invention, the image evidence data packets are stored in the hospital image archiving and communication system indexed by the patient's hospital number and alarm time, with a storage period of no less than 6 months after the patient's discharge, to meet the needs of medical quality management and possible medical dispute resolution. The stored data is encrypted and access control is implemented to ensure patient privacy and security.

[0081] In summary, the intelligent assessment and extravasation early warning system 100 for peripheral intravenous infusion sites in chemotherapy patients according to this invention acquires skin images of the puncture site through the infusion site image acquisition module 1. The image preprocessing module 2 performs illumination correction and skin color standardization. The skin state analysis module 3 extracts multidimensional skin state features and performs temporal fusion. The extravasation identification module 4 identifies extravasation signs such as redness, pallor, swelling, and changes in skin temperature and calculates confidence levels. The risk grading module 5 performs risk-weighted assessment based on the chemotherapy drug irritation grading and outputs the risk level. The early warning push module 6 immediately pushes alarms for medium- and high-risk events and packages and stores image evidence. This invention achieves intelligent and continuous monitoring of chemotherapy infusion safety, which helps in the early detection and timely intervention of extravasation injuries, and has significant application value in improving the quality of chemotherapy nursing and ensuring patient safety.

[0082] In practical applications, the system of this invention achieved a sensitivity of 92% and a specificity of 88% in detecting extravasation events during clinical trials on 300 chemotherapy patients. The median warning response time was shortened by approximately 12 minutes compared to traditional manual inspections, and the accuracy rate of the golden time window for treating extravasation of vesicant drugs increased from 67% to 89%. These results demonstrate that the technical solution of this invention has good clinical application effects.

[0083] The embodiments of the present invention are not limited to the specific embodiments described above. Those skilled in the art can make various equivalent changes or substitutions based on the technical solutions of the present invention, and all such changes or substitutions should be included within the protection scope of the present invention.

Claims

1. An intelligent image assessment and extravasation early warning system for peripheral intravenous infusion sites in chemotherapy patients, characterized in that, include: The infusion site image acquisition module is used to acquire skin image data of the infusion site by means of a miniature camera fixed to the patient's peripheral venous puncture site according to a preset sampling period; The image preprocessing module is used to perform illumination correction and skin color standardization on the skin image data of the infusion site to obtain standardized skin image data. The skin condition analysis module is used to identify the skin area around the puncture point in standardized skin image data using a deep learning segmentation network, and automatically extract skin color features, swelling degree features and texture change features. The extracted multidimensional features are then fused temporally to obtain a skin condition feature vector. The extravasation identification module is used to perform feature matching and analysis on local redness, pallor, swelling and skin temperature change signs based on skin state feature vectors, calculate the confidence level of each sign and synthesize the confidence level value of extravasation signs. The risk grading module is used to obtain the irritation level data of the currently infused chemotherapy drugs, perform a weighted fusion calculation between the extravasation sign confidence value and the irritation level data, and output the extravasation risk level result based on the fusion result. The early warning push module is used to determine whether the extravasation risk level results have reached the preset alarm conditions. When the alarm conditions are reached, an early warning message is generated and pushed to the responsible nurse's mobile terminal in real time. At the same time, the skin image data of the current moment and the previous moment are packaged into an image evidence data package for storage.

2. The intelligent image assessment and extravasation early warning system for peripheral intravenous infusion sites in chemotherapy patients according to claim 1, characterized in that, The skin state analysis module uses a semantic segmentation network with an encoder-decoder structure to process the standardized skin image data and generate a puncture site region segmentation feature map. The puncture site region segmentation feature map is used to identify the skin region of interest within a preset radius around the puncture point.

3. The intelligent image assessment and extravasation early warning system for peripheral intravenous infusion sites in chemotherapy patients according to claim 2, characterized in that, The skin state analysis module extracts skin color feature vectors, swelling degree feature vectors, and texture change feature vectors from the region of interest defined by the segmentation feature icon of the puncture site. Specifically, the skin color feature vector is obtained through LAB color space conversion and color histogram statistics, the swelling degree feature vector is obtained through edge contour detection and area change rate calculation, and the texture change feature vector is obtained through gray-level co-occurrence matrix and local binary mode extraction.

4. The intelligent image assessment and extravasation early warning system for peripheral intravenous infusion sites in chemotherapy patients according to claim 3, characterized in that, The skin state analysis module further includes a temporal difference analysis unit, which is used to calculate the pixel-level difference between the standardized skin image data at adjacent sampling time points to obtain a temporal difference feature map, and to concatenate and fuse the temporal difference feature map with the skin color feature vector, the swelling degree feature vector and the texture change feature vector at the current time to obtain the skin state feature vector.

5. The intelligent image assessment and extravasation early warning system for peripheral intravenous infusion sites in chemotherapy patients according to claim 1, characterized in that, The risk grading module determines the drug irritation weighting coefficient based on the irritation classification of chemotherapy drugs. The drug irritation weighting coefficient for vesicants is set to a first preset value, the drug irritation weighting coefficient for irritants is set to a second preset value, and the drug irritation weighting coefficient for non-irritants is set to a third preset value. The first preset value is greater than the second preset value, and the second preset value is greater than the third preset value.

6. The intelligent image assessment and extravasation early warning system for peripheral intravenous infusion sites in chemotherapy patients according to claim 5, characterized in that, The risk grading module multiplies the confidence value of the extravasation signs by the weighted coefficient of drug irritation and combines it with a preset benchmark score to calculate a comprehensive extravasation risk score. The comprehensive extravasation risk score is compared with a preset risk threshold. When the comprehensive extravasation risk score is lower than the first preset risk threshold, a low risk level is output. When the comprehensive extravasation risk score is between the first preset risk threshold and the second preset risk threshold, a medium risk level is output. When the comprehensive extravasation risk score is higher than the second preset risk threshold, a high risk level is output.

7. The intelligent image assessment and extravasation early warning system for peripheral intravenous infusion sites in chemotherapy patients according to claim 1, characterized in that, When generating the image evidence data packet, the early warning push module arranges the skin image data at the current moment and all skin image data within a preset time window before the current moment in chronological order and adds timestamps and extravasation sign annotations.

8. The intelligent image assessment and extravasation early warning system for peripheral intravenous infusion sites in chemotherapy patients according to claim 1, characterized in that, The illumination correction processing of the image preprocessing module includes white balance adjustment and histogram equalization processing, and the skin color normalization processing includes color space mapping processing based on a reference skin color template.

9. The intelligent image assessment and extravasation early warning system for peripheral intravenous infusion sites in chemotherapy patients according to claim 1, characterized in that, The extravasation identification module uses a multi-label classification network to independently assess the confidence levels of redness, pallor, swelling, and changes in skin temperature. The independent confidence levels of each sign are weighted and summed to obtain the confidence value of the extravasation sign.

10. The intelligent image assessment and extravasation early warning system for peripheral intravenous infusion sites in chemotherapy patients according to claim 1, characterized in that, The preset sampling period of the infusion site image acquisition module is a configurable parameter. In the initial stage of infusion, the first sampling period is used for image acquisition. When the confidence value of the extravasation sign exceeds the preset monitoring threshold, the module automatically switches to the second sampling period for image acquisition. The second sampling period is shorter than the first sampling period.

Images