Methods and Systems for Image Assessment of Skin Injuries Caused by Chemotherapy Drug Extravasation

Abstract

This application belongs to the field of skin injury image assessment technology, specifically providing a method and system for assessing skin injury caused by chemotherapy drug extravasation. The method mainly includes: acquiring a multispectral image of the affected area; acquiring a damage reference spectral vector representing typical drug-induced damage characteristics; based on the damage reference spectral vector, performing orthogonal projection decomposition on the spectrum of each pixel in the multispectral image to generate a projection component image and a spectral residual image; extracting the local texture pattern of the spectral residual image to generate a residual texture map for quantifying the heterogeneity of tissue microstructure. This application effectively solves the problem in existing technologies where early drug extravasation injury and subcutaneous bruising are difficult to accurately distinguish under conventional visual conditions, easily leading to clinical misjudgment and delayed intervention. It enables early warning of the risk of early, occult tissue necrosis caused by chemotherapy drug extravasation, thereby potentially reducing the incidence of severe tissue damage.

Description

Technical Field

[0001] This application belongs to the field of skin injury image assessment technology, and specifically relates to a method and system for assessing skin injury caused by chemotherapy drug extravasation. Background Technology

[0002] In the clinical monitoring and early complication identification of intravenous chemotherapy, medical staff generally make a preliminary assessment of whether drug extravasation has occurred and its severity by visually observing the color, swelling, and skin temperature changes of the skin around the puncture site and making judgments based on their personal clinical experience.

[0003] However, after the extravasation of certain chemotherapy drugs, the early physical signs are visually highly similar to ordinary subcutaneous bruising caused by intravenous puncture. They generally appear similar red or purplish-red under visible light. In conventional visible light images captured from the skin surface, the key bio-optical information characterizing early occult tissue necrosis caused by drug toxicity and the interfering optical information characterizing benign subcutaneous hemorrhage are highly overlapping in color and morphology, making them difficult to distinguish. This may lead to the misjudgment of drug extravasation with a high risk of tissue necrosis as ordinary bruising that can heal on its own, thus delaying the critical intervention time. Summary of the Invention

[0004] This application provides a method and system for evaluating skin damage caused by chemotherapy drug extravasation. It effectively solves the problem in the prior art where early drug extravasation damage and subcutaneous bruising are difficult to distinguish accurately under conventional visual conditions, which can easily lead to clinical misjudgment and delayed intervention. It enables early warning of the risk of early occult tissue necrosis caused by chemotherapy drug extravasation, and is expected to reduce the incidence of severe tissue damage.

[0005] To achieve the above objectives, this application adopts the following technical solution:

[0006] In a first aspect, this application provides a method for image assessment of skin damage caused by chemotherapy drug extravasation, including:

[0007] Obtain multispectral images of the affected area; obtain damage reference spectral vectors for typical drug-induced damage characteristics.

[0008] Based on the damage reference spectral vector, the spectrum of each pixel in the multispectral image is orthogonally projected and decomposed to generate the projected component image and the spectral residual image.

[0009] Local texture patterns are extracted from spectral residual images to generate residual texture maps for quantifying the heterogeneity of tissue microstructure.

[0010] The mutual information between the projected component image and the residual texture map is calculated to generate an early concealment damage score.

[0011] Risk assessment results are generated based on early occult injury scores and preset scoring thresholds.

[0012] Secondly, this application provides an image assessment system for skin damage caused by chemotherapy drug extravasation, comprising:

[0013] Data acquisition module: used to acquire multispectral images of the affected area; acquire damage reference spectral vectors of typical drug-induced damage characteristics.

[0014] Feature separation module: Based on the damage reference spectral vector, it performs orthogonal projection decomposition on the spectrum of each pixel in the multispectral image to generate projection component images and spectral residual images.

[0015] Texture Analysis Module: Used to extract local texture patterns from spectral residual images and generate residual texture maps for quantifying the heterogeneity of tissue microstructure.

[0016] Fusion evaluation module: used to calculate the mutual information between the projected component image and the residual texture map, and generate an early hidden damage score.

[0017] Results output module: Used to generate risk assessment results based on early occult injury scores and preset scoring thresholds.

[0018] Thirdly, this application provides a readable storage medium, comprising: computer program instructions stored in the readable storage medium, wherein the computer program instructions are read and executed by a processor to perform the steps of a method for evaluating skin damage caused by extravasation of chemotherapy drugs.

[0019] The beneficial effects of this application are:

[0020] This application acquires multispectral images of the affected area, uses a drug spectral template for orthogonal projection to separate the damage from the background components, extracts background texture features, and fuses the mutual information of the spectrum and texture to quantify the risk. This effectively solves the problem in existing technologies where early drug extravasation damage and subcutaneous bruising are difficult to distinguish accurately under conventional visual conditions, which can easily lead to clinical misjudgment and delayed intervention. It enables early warning of the risk of early occult tissue necrosis caused by chemotherapy drug extravasation, providing a key decision-making basis for timely and targeted intervention in clinical practice, thereby potentially reducing the incidence of severe tissue damage.

[0021] Other features and advantages of this application will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description and the accompanying drawings. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A schematic flowchart of the image assessment method for skin damage caused by chemotherapy drug extravasation according to this application is shown;

[0024] Figure 2 A schematic diagram of the process for generating residual texture maps for quantifying the heterogeneity of tissue microstructure is shown in this application;

[0025] Figure 3 A schematic diagram of the process for obtaining the optimal morphological enhancement control parameter pair in this application is shown;

[0026] Figure 4 A schematic diagram of the process for obtaining the fusion coefficient in this application is shown. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0028] Early clinical assessment of chemotherapy drug extravasation is mainly based on clinical observation by medical staff, judging characteristics such as skin color and swelling. In addition, the use of visible light imaging technology to assist in recording and comparative analysis is also a common method. In the early stage of extravasation of certain chemotherapy drugs such as anthracyclines, the skin color and shape of the damaged area are similar to those of ordinary subcutaneous bruising caused by venipuncture under conventional imaging.

[0029] To address the aforementioned issues, the inventors discovered that early tissue damage caused by drug extravasation exhibits separable differences in reflectance characteristics at specific wavelengths in multispectral imaging compared to subcutaneous bruising. Furthermore, the resulting changes in the deep tissue microstructure produce specific representations in image texture patterns. Based on this, a method was proposed to target and separate components matching a pre-defined drug-induced damage spectral template from multispectral images using orthogonal projection technology. This method further quantifies tissue texture heterogeneity from the residual information after separation. By fusing the separated spectral feature components with the extracted texture features and evaluating their spatial consistency, a model capable of quantifying the risk of early, occult tissue damage was constructed.

[0030] In some embodiments, such as Figure 1 As shown, this application provides a method for image evaluation of skin damage caused by chemotherapy drug extravasation, including:

[0031] S1. Obtain multispectral images of the affected area; obtain damage reference spectral vectors for typical drug-induced damage characteristics.

[0032] The affected area refers to a specific area of ​​skin on the patient's body where chemotherapy drugs have occurred or are suspected to have extravasated. Multispectral imaging equipment is used to photograph the affected area, thereby obtaining multispectral images.

[0033] The damage reference spectral vector is a normalized mathematical vector pre-learned from historical confirmed cases, with the same dimension as the number of imaging bands. Each dimension of the damage reference spectral vector reflects the typical reflectance characteristics of extravasated skin damage caused by a specific chemotherapy drug in the corresponding imaging band. During implementation, the system automatically retrieves and calls the corresponding damage reference spectral vector from a pre-built drug feature database based on the input chemotherapy drug name.

[0034] For example, to evaluate a patient who has experienced suspected extravasation on the back of their hand after receiving chemotherapy with anthracycline drugs, a multispectral skin imaging system is used to capture multispectral images of the suspected extravasation area on the back of the patient's right hand. Based on the drug name, the damage reference spectral vector for that drug is retrieved from a pre-stored database.

[0035] S2. Based on the damage reference spectral vector, orthogonal projection decomposition is performed on the spectrum of each pixel in the multispectral image to separate components related to the spectral characteristics of drug damage from the multispectral image. The remaining spectral components are then separated, generating projection component images and spectral residual images respectively. The projection component images reflect the degree of matching between each location in the affected area image and the typical spectral characteristics of drug damage. The spectral residual images contain the remaining spectral information after deducting the aforementioned matching components.

[0036] S3. Extract local texture patterns from the spectral residual image and optimize them to generate residual texture maps for quantifying tissue microstructural heterogeneity. The residual texture maps reflect changes in tissue microstructure that may be caused by early damage.

[0037] S4. Calculate the mutual information between the projected component image and the residual texture map to generate an early concealed damage score. The early concealed damage score is used to quantify the risk level of early concealed damage.

[0038] S5. Generate risk assessment results based on the early occult injury score and the preset scoring threshold.

[0039] In some embodiments, determining a damage reference spectral vector representing typical drug-induced damage characteristics includes:

[0040] S11. Obtain multispectral images of the affected area of ​​patients diagnosed with extravasation injury from specific chemotherapy drugs, and obtain image samples.

[0041] The raw data used to learn the spectral characteristics of the injury are obtained, specifically historical cases of skin damage caused by specific chemotherapy drugs that have been clinically diagnosed. High-quality multispectral images taken in the early stages of the injury are selected to form the image samples to be learned.

[0042] S12. Extract the average spectral reflectance curve from the damaged core region of the image sample.

[0043] The core damage region specifically refers to the area in an image where changes in color, texture, etc., are most typical and clearly defined; it is usually the part clinically judged as the most severely damaged. The average reflectance of all pixels within this core damage region is calculated for each imaging band. This process is repeated for all bands to obtain an average spectral reflectance curve. The average spectral reflectance curve is a vector, where the value of the j-th element represents the average reflectance of the core damage region of the image sample in the j-th imaging band.

[0044] S13. Calculate the mean vector of all average spectral reflectance curves, and normalize the mean vector to obtain the damage reference spectral vector.

[0045] The average spectral reflectance curves of all image samples are arithmetically averaged along the same spectral band dimension to obtain a mean vector, which characterizes the average spectral reflectance characteristics of damage caused by a specific chemotherapy drug at the population level. This mean vector is then normalized to eliminate the influence of absolute reflectance intensity, and the resulting unit vector is the damage reference spectral vector for that specific chemotherapy drug. This process is repeated for other chemotherapy drugs. to The process of establishing a corresponding drug damage spectral feature library can be achieved.

[0046] For example, to establish a damage reference spectral vector for an anthracycline drug, 100 early images of cases diagnosed with extravasation of the drug were found in the database as image samples. For each image, the physician delineated the core area of ​​damage and calculated the average spectral reflectance curve of the area. The 100 average spectral reflectance curves were averaged and normalized point by point, and the resulting vector was used as the damage reference spectral vector for the anthracycline drug.

[0047] In some embodiments, based on the damage reference spectral vector, orthogonal projection decomposition is performed on the spectrum of each pixel in the multispectral image to generate a projected component image and a spectral residual image, including:

[0048] S21. For each pixel of the multispectral image, calculate the dot product of the spectral vector and the damage reference spectral vector to obtain the spectral projection component.

[0049] For each pixel in a multispectral image, its reflectance values ​​in each band constitute a spectral vector. If the normalized damage reference spectrum vector is Calculate the spectral vector In reference vector The projection scalar value in the direction, i.e., calculating the dot product of the two. , The sign and magnitude of the value reflect the spectrum of that pixel. Compared with reference spectrum The similarity and intensity, if and If the directions are completely consistent, then The absolute value is the largest.

[0050] S22. Arrange the spectral projection components of all pixels according to their original image spatial positions to generate a projection component image; subtract the spectral projection components from the spectral vector to obtain the residual vector, and calculate the magnitude of the residual vector.

[0051] Iterate through all pixels of the multispectral image and project the spectral components for each pixel. Create a single-channel blank image with the same spatial dimensions as the multispectral image, and calculate the position of each pixel. The value is used as the pixel gray value at the corresponding position in the blank image to fill in the blank image. The resulting single-channel image is the projection component image, which represents the degree of matching between each position in the whole image and the spectral characteristics of drug damage.

[0052] For the same pixel, when obtaining the spectral projection component Then, calculate the original spectral vector. With it The difference between the projection vectors in the direction yields the residual vector. , The residual vector contains the reference spectral vector from the original spectrum that cannot be damaged. The explanation section.

[0053] Calculate this residual vector Length of the module , , module length It represents the total intensity of the remaining spectral components.

[0054] S23. Arrange the modulus values ​​of all pixels according to their spatial positions in the original image to generate a spectral residual image.

[0055] For each pixel, calculate the magnitude m of its residual vector. Similar to generating the projected component image, create a single-channel blank image of the same size as the multispectral image, and calculate the magnitude m for each pixel position. Fill in the corresponding positions, and the resulting single-channel image is the spectral residual image. The spectral residual image mainly reflects spectral variations that do not conform to the preset drug damage spectral pattern.

[0056] For example, in the case of evaluating extravasation on the back of the hand, for each pixel of the multispectral image of the back of the hand, its spectral vector This includes the reflectance of this point in multiple wavelength bands, including red, green, blue, and near-infrared. It is compared with the damage reference spectral vector of an anthracycline drug. Perform a dot product operation to obtain the spectral projection component of that point. All points The values ​​form the projected component image, and the brighter areas in the image indicate that the spectrum at that location is highly similar to the damage characteristics of doxorubicin. Simultaneously, the residual vector is calculated. The modulus m is taken, and the m values ​​of all points constitute a spectral residual image, which may highlight structures such as subcutaneous bruising or blood vessels that are not related to drug damage.

[0057] In some embodiments, such as Figure 2 As shown, local texture patterns are extracted from the spectral residual image to generate a residual texture map for quantifying the heterogeneity of tissue microstructure, including:

[0058] S31. Extract texture features that characterize the heterogeneity of tissue microstructure from the spectral residual image to generate an initial texture image.

[0059] For each pixel in the spectral residual image, a small neighborhood, such as a 3x3 rectangular neighborhood, is defined. The gray value of the center pixel is used as a threshold and compared with the gray values ​​of all other pixels in the neighborhood. If the value of a neighboring pixel is greater than or equal to the value of the center pixel, the position is marked as 1; otherwise, it is marked as 0. This results in a binary pattern string consisting of 0s and 1s. This binary string is converted into a decimal number and used as the new gray value of the center pixel. The above operation is performed on each pixel of the spectral residual image. The resulting new image is the initial texture image. The initial texture image encodes the local spatial structure information of the original spectral residual image. High gray value regions usually correspond to regions with complex textures.

[0060] S32. Based on a preset set of structuring element sizes and a set of morphological operation types, the morphological features of the initial texture image are optimized and evaluated to obtain the optimal morphological enhancement control parameters. , This represents the optimal structuring element size in the set of structuring element sizes. It represents the optimal morphological operation type in the set of morphological operation types.

[0061] The initial texture image may contain noise or fine textures. To enhance texture patterns associated with potential tissue damage, morphological image processing techniques can be employed. The effectiveness of morphological processing heavily depends on the size of the structuring elements and the type of operation performed. An optimal combination of parameters can be selected from a set of pre-defined candidate parameters to obtain the optimal morphological enhancement control parameter pair. .

[0062] Candidate parameters are the set of structuring element sizes and the set of morphological operation types. The set of structuring element sizes includes multiple sizes, and the set of morphological operation types includes multiple operations. For example, the size has a radius of 1, 3, or 5 pixels, and the operation includes erosion, dilation, opening, and closing.

[0063] S33. The size used is... The structuring element performs an operation of type on the initial texture image. Morphological operations are performed to obtain residual texture maps.

[0064] Build a size of Structural elements, such as those with a radius of A circular structuring element, on which the operation type is performed on the initial texture image. Morphological operations. For example, if For the closing operation, the image is first dilated, and then the dilated result is eroded. This can smooth the contours, connect adjacent bright areas, and fill small holes, thereby enhancing the significant texture structure and suppressing noise. The output image after this optimized morphological processing is the final residual texture map. The residual texture map more clearly quantifies the heterogeneous changes in the microstructure of the tissue.

[0065] For example, for the spectral residual image of a case of extravasation on the back of the hand, firstly through... The algorithm generates an initial texture image, which may display some scattered texture. Then, through the optimization process in Example 5, the optimal parameters may be determined to be using a radius of... The circular structuring element of the pixel is closed. After processing the initial texture image with this parameter, a residual texture map is obtained. The map may show connected, coarse texture areas, indicating that the subcutaneous tissue may have edema or structural damage.

[0066] In some embodiments, such as Figure 3 As shown, based on a preset set of structuring element sizes and a set of morphological operation types, the morphological features of the initial texture image are optimized and evaluated to obtain the optimal morphological enhancement control parameters. ,include:

[0067] S321. Perform multi-scale decomposition on the initial texture image to obtain a set of feature images.

[0068] A Gaussian pyramid downsampling operation is performed on the initial texture image to perform multi-scale decomposition. Specifically, the initial texture image is used as the bottom layer of the pyramid, and Gaussian filtering and downsampling are applied to it to obtain a first-layer image with lower resolution; this process is repeated to obtain the second and third-layer images, until the preset number of layers is reached. Finally, a set of images was obtained. That is, a set of feature images, which, through multi-scale analysis, helps to capture texture features of different granularities, from fine to coarse.

[0069] S322. [It is provided] Size of each candidate structuring element and Candidate morphological operation types For each pair of parameter combinations in the set of structuring element sizes and the set of morphological operation types Computational morphological enhancement quality , , The first in the set of representative feature images A feature image, Represented by size The structuring element performs the operation type on the feature image. The resulting image obtained from morphological operations, Representative to The arithmetic mean of all pixels is calculated, reflecting the average shift in the overall grayscale level of the image after this morphological operation. This can be considered as an enhanced response at this scale. The number of feature images at different scales in the feature image set is represented by the number of feature images at different scales, and all parameter combinations are iterated over. Calculate their respective Arrange them into a OK The matrix of columns yields the morphological enhancement quality matrix. .

[0070] S323. To evaluate the stability of the enhancement effect, i.e., whether the performance is consistent across different scales, calculate each pair of parameter combinations based on the feature image set and the morphological enhancement quality matrix. At each scale The response below and overall response Calculate the response at a single scale relative to its average value. The fluctuations yielded morphologically enhanced robustness. , , The standard deviation of each scale response value relative to its overall mean. The value equals the total reinforcement mass. Subtracting that standard deviation, therefore, It requires not only a large total strength of reinforcement, but also that such reinforcement is stable at different scales.

[0071] Iterate through all parameter combinations The morphologically enhanced robust mass matrix is ​​obtained. .

[0072] S324. Obtain the morphologically enhanced robust quality matrix The maximum value of morphological enhancement robustness is used to point the row index and column index of this maximum value to the structuring element size and morphological operation type, respectively. and Assume that the maximum value is located in the matrix. The line, number column, i.e. [ , If ] is the maximum value, then the row index is... It corresponds to the size in the set of structural element sizes that results in the highest robust quality. Column index This corresponds to the operation type in the set of morphological operation types that results in the highest robustness quality. Therefore, the optimal morphological enhancement control parameter pair is... .

[0073] For example, the set of structuring element sizes is The set of morphological operation types is The initial texture image is decomposed into a 3-layer pyramid to obtain a set of feature images. Each layer of the image is processed using each parameter combination, the average gray-level response is calculated, and the results are summed to obtain the final value. Matrix, then calculate robust mass Matrix, if a parameter combination is found corresponding The maximum value indicates the optimal morphological enhancement control parameter. That is .

[0074] In some embodiments, the mutual information between the projected component image and the residual texture map is calculated to generate an early concealed damage score, including:

[0075] S41. Calculate the local entropy map of the projected component image and the local entropy map of the residual texture image respectively to obtain the first entropy map and the second entropy map; perform binarization segmentation on the projected component image to generate the mask region.

[0076] Use a fixed-size sliding window for the projected component image, for example... For each pixel, the information entropy of all the gray values ​​of all pixels in the window is calculated for each pixel in the center of the window. The calculated entropy value is assigned to the corresponding position of the pixel in the center of the window in the new image. After the traversal is completed, the new image is the first entropy map. The first entropy map describes the complexity and uncertainty of the local region of the projected component image.

[0077] Similarly, the exact same operation is performed on the residual texture map to calculate its local entropy, resulting in a second entropy map, which characterizes the complexity and uncertainty of local regions of the residual texture map.

[0078] Automatic thresholding segmentation of the projected component image is performed using the Otsu method. By traversing all possible grayscale thresholds, the inter-class variance between the foreground and background is calculated, and the threshold that maximizes the inter-class variance is selected as the segmentation threshold. The projected component image is then binarized using this threshold, where pixel regions with grayscale values ​​higher than the threshold are defined as the foreground, and the rest are defined as the background. The connected regions of the foreground in this binary image are the mask regions, which roughly locate the suspected regions that may be related to damage in terms of spectral features.

[0079] S42. Evaluate the consistency of the first entropy map and the second entropy map within the mask region to obtain the fusion coefficient. The fusion coefficient is a value between... arrive The values ​​between these values ​​are used to quantify the consistency of pixel value change patterns in the first and second entropy maps within the mask region. The higher the fusion coefficient value, the more synchronous the changes in tissue texture complexity and spectral features are within the spectral anomaly region.

[0080] Calculate the global mutual information between the projected component image and the residual texture map. This indicates a stronger overall statistical correlation between the projected component image and the residual texture map. Represents the projected component image. Represents the residual texture map.

[0081] S43. Calculate the early concealment damage score based on the fusion coefficient and global mutual information. , This represents the score for early, occult lesions. Represents the fusion coefficient. This represents the global mutual information between the projected component image and the residual texture map.

[0082] Early occult injury score To quantify the risk level of early, occult lesions, two conditions must be met simultaneously: "within the suspected area, the variation patterns of spectral complexity and texture complexity are highly consistent" and "the spectral lesion feature map and the tissue structure texture map are statistically closely correlated globally." High value and A high value indicates that the two indicators are combined through multiplication, and a low value in either indicator will lower the overall score.

[0083] For example, in the case of extravasation on the back of the hand, the masked area of ​​the projected component image may cover the red and swollen area on the back of the hand. The consistency of the first entropy map and the second entropy map in this area is calculated to obtain the fusion coefficient. Simultaneously calculate the global mutual information of the two original images. The calculated early hidden injury score .

[0084] In some embodiments, such as Figure 4 As shown, the consistency of the first entropy map and the second entropy map within the mask region is evaluated to obtain the fusion coefficient, including:

[0085] S421. Extract all pixel values ​​within the mask region from the first entropy map and the second entropy map respectively to obtain the first entropy value sequence and the second entropy value sequence.

[0086] Iterate through each value in the mask region The pixel position is recorded, and the corresponding gray value at that position is recorded in the first entropy image. Simultaneously, record the corresponding grayscale value in the second entropy map. All extracted The values ​​are arranged in the order of extraction to form the first entropy value sequence P. Similarly, all extracted values ​​are... The values ​​are arranged to form the second entropy value sequence G.

[0087] S422. Calculate the mean of the first entropy value sequence. Mean of the second entropy sequence , and These represent the average level of complexity characterized by the first entropy map and the second entropy map within the mask region, respectively.

[0088] S423. Based on the mean and mean Calculate core difference measure It is used to quantify the difference in the relative changes of corresponding pixel values ​​in two entropy maps within the mask region. , This represents the total number of pixels within the mask area. and These represent the i-th element of the first entropy value sequence and the second entropy value sequence, respectively. and These represent the centering deviations of each element in the sequence from its mean. This represents preventing the denominator from being The smallest positive number, and Both represent normalization factors, making and It becomes a standardized relative deviation relative to its own mean. This represents the difference in the standardized relative deviation of the corresponding pixels. If the two sequence change patterns are completely synchronized, this difference should be close to 0.

[0089] S424. Based on the mean mean and core difference measurement Calculate the fusion coefficient , , Represents the relative difference between the means of two sequences, and its value is in arrive Between them, the smaller the difference, the closer it is to 0. Representatives will reflect differences in internal change patterns. Multiplying the relative difference by the mean, which reflects the overall level of difference, yields a comprehensive measure of difference. The exponential function maps the input to... The interval is monotonically decreasing, and when the overall difference is 0... As the overall differences increase, Towards Attenuation. Fusion coefficient. The study comprehensively measures the consistency of change patterns and the similarity of average levels between two entropy map sequences within the mask region; a higher value indicates better consistency.

[0090] For example, if the first entropy sequence extracted from the mask region The mean is , the second sequence The mean is 85, calculated as follows Then the fusion coefficient This indicates a high degree of consistency.

[0091] In some embodiments, a risk assessment result is generated based on an early concealed injury score and a preset scoring threshold, including: comparing the early concealed injury score with the scoring threshold; if the early concealed injury score is greater than the scoring threshold, a high-risk conclusion is generated as the risk assessment result; otherwise, a normal observation conclusion is generated as the risk assessment result.

[0092] Determining the scoring threshold involves: obtaining a historical case dataset containing early occult lesion scores from clinically diagnosed cases of occult necrosis; calculating the mean and standard deviation of all early occult lesion scores; and setting the sum of the mean and twice the standard deviation as the scoring threshold.

[0093] In some embodiments, this application provides an image assessment system for skin damage caused by chemotherapy drug extravasation, comprising:

[0094] Data acquisition module: used to acquire multispectral images of the affected area; acquire damage reference spectral vectors of typical drug-induced damage characteristics;

[0095] Feature separation module: used to perform orthogonal projection decomposition of the spectrum of each pixel in the multispectral image based on the damage reference spectral vector, and generate projection component images and spectral residual images;

[0096] Texture analysis module: used to extract local texture patterns from spectral residual images and generate residual texture maps for quantifying the heterogeneity of tissue microstructure;

[0097] Fusion evaluation module: used to calculate the mutual information between the projected component image and the residual texture map, and generate an early concealment damage score;

[0098] Results output module: Used to generate risk assessment results based on early occult injury scores and preset scoring thresholds.

[0099] In some embodiments, this application provides a readable storage medium, comprising: computer program instructions stored in the readable storage medium, wherein the computer program instructions are read and executed by a processor to perform the steps of a method for evaluating skin damage images caused by extravasation of chemotherapy drugs.

[0100] It should be noted that, in this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0101] Any references to memory, storage, database, or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory.

[0102] Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for image assessment of skin damage caused by chemotherapy drug extravasation, characterized in that, include: Acquire multispectral images of the affected area; Obtain the damage reference spectral vector of typical drug-induced damage characteristics; Based on the damage reference spectral vector, the spectrum of each pixel in the multispectral image is orthogonally projected and decomposed to generate a projection component image and a spectral residual image. Local texture patterns are extracted from the spectral residual image to generate a residual texture map for quantifying the heterogeneity of tissue microstructure; Calculate the mutual information between the projected component image and the residual texture map to generate an early concealed damage score; Based on the early hidden injury score and the preset scoring threshold, a risk assessment result is generated.

2. The method based on claim 1, characterized in that, The damage reference spectral vector for determining typical drug-induced damage characteristics includes: Obtain multispectral images of the affected area in patients diagnosed with extravasation injury from specific chemotherapy drugs to obtain image samples; The average spectral reflectance curve is extracted from the damaged core region of the image sample; Calculate the mean vector of all average spectral reflectance curves, and normalize the mean vector to obtain the damage reference spectral vector.

3. The method based on claim 1, characterized in that, Based on the damage reference spectral vector, orthogonal projection decomposition is performed on the spectrum of each pixel in the multispectral image to generate a projection component image and a spectral residual image, including: For each pixel of the multispectral image, the dot product of the spectral vector and the damage reference spectral vector is calculated to obtain the spectral projection component; Arrange the spectral projection components of all pixels according to their original image spatial positions to generate a projection component image; subtract the spectral projection components from the spectral vector to obtain the residual vector, and calculate the magnitude of the residual vector. Arrange the modulus values ​​of all pixels according to their spatial positions in the original image to generate a spectral residual image.

4. The method based on claim 1, characterized in that, Extracting local texture patterns from the spectral residual image to generate a residual texture map for quantifying the heterogeneity of tissue microstructure includes: Texture features characterizing the heterogeneity of tissue microstructure are extracted from the spectral residual image to generate an initial texture image; Based on a preset set of structuring element sizes and a set of morphological operation types, the morphological features of the initial texture image are optimized and evaluated to obtain the optimal morphological enhancement control parameters. , This represents the optimal structuring element size in the set of structuring element sizes. The optimal morphological operation type in the set of morphological operation types; The size is The structuring element performs an operation of type on the initial texture image. Morphological operations are performed to obtain residual texture maps.

5. The method based on claim 4, characterized in that, Based on a preset set of structuring element sizes and a set of morphological operation types, the morphological features of the initial texture image are optimized and evaluated to obtain the optimal morphological enhancement control parameters. ,include: The initial texture image is decomposed into a multi-scale model to obtain a set of feature images. For each pair of parameter combinations in the set of structural element dimensions and the set of morphological operation types Computational morphological enhancement quality , , The first in the set of representative feature images A feature image, Represented by size The structuring element performs the operation type on the feature image. The resulting image obtained from morphological operations, Representative to Calculate the arithmetic mean of all pixels. The number of feature images at different scales in the feature image set is represented by the number of feature images at different scales, and all parameter combinations are iterated over. The morphological enhancement quality matrix is ​​obtained. ; Based on the feature image set and the morphological enhancement quality matrix, calculate each pair of parameter combinations. Morphological enhancement of robustness , Iterate through all parameter combinations The morphologically enhanced robust mass matrix is ​​obtained. ; Obtaining the morphologically enhanced robust quality matrix The maximum value of morphological enhancement robustness is used to point the row index and column index of this maximum value to the structuring element size and morphological operation type, respectively. and .

6. The method based on claim 1, characterized in that, The mutual information between the projected component image and the residual texture map is calculated to generate an early concealed damage score, including: The local entropy map of the projected component image and the local entropy map of the residual texture map are calculated respectively to obtain the first entropy map and the second entropy map; the projected component image is subjected to binarization segmentation to generate the mask region; The consistency of the first entropy map and the second entropy map within the mask region is evaluated to obtain the fusion coefficient; the global mutual information between the projected component image and the residual texture map is calculated. Based on the fusion coefficient and global mutual information, an early concealment damage score is calculated. , This represents the score for early, occult lesions. Represents the fusion coefficient. This represents the global mutual information between the projected component image and the residual texture map.

7. The method based on claim 6, characterized in that, The consistency of the first entropy map and the second entropy map within the mask region is evaluated to obtain the fusion coefficient, including: Extract all pixel values ​​within the mask region from the first entropy map and the second entropy map respectively to obtain the first entropy value sequence and the second entropy value sequence; Calculate the mean of the first entropy value sequence. Mean of the second entropy sequence ; According to the mean and mean Calculate core difference measure , , This represents the total number of pixels within the mask area. and These represent the i-th element of the first entropy value sequence and the second entropy value sequence, respectively. This represents preventing the denominator from being The smallest positive number; According to the mean mean and core difference measurement Calculate the fusion coefficient , .

8. The method based on claim 1, characterized in that, Determine the scoring threshold, including: Obtain a historical case dataset containing early occult lesion scores of clinically diagnosed cases of occult necrosis; Calculate the mean and standard deviation of all early occult injury scores, and set the sum of the mean and twice the standard deviation as the scoring threshold.

9. The method based on claim 1, characterized in that, Based on the early occult injury score and a preset scoring threshold, a risk assessment result is generated, including: The early hidden injury score is compared with the scoring threshold. If the early hidden injury score is greater than the scoring threshold, a high-risk conclusion is generated as the risk assessment result; otherwise, a normal observation conclusion is generated as the risk assessment result.

10. A skin injury image assessment system for chemotherapy drug extravasation, characterized in that, It includes: Data acquisition module: used to acquire multispectral images of the affected area; Obtain the damage reference spectral vector of typical drug-induced damage characteristics; Feature separation module: used to perform orthogonal projection decomposition on the spectrum of each pixel in the multispectral image based on the damage reference spectral vector, and generate projection component image and spectral residual image; Texture analysis module: used to extract local texture patterns from the spectral residual image and generate residual texture maps for quantifying the heterogeneity of tissue microstructure; Fusion evaluation module: used to calculate the mutual information between the projected component image and the residual texture map, and generate an early concealment damage score; Results output module: used to generate risk assessment results based on the early hidden injury score and the preset scoring threshold.

Images