Intelligent measuring system for muscle fiber morphology

By sharpening and equal-scale cropping of muscle fiber pictures, combined with batch reasoning and post-processing of FastSAM model, the problems of manual measurement time and high computing resources in the existing technology are solved, and fast and accurate muscle fiber morphology analysis is achieved. Improved recognition accuracy and efficiency.

CN118918335APending Publication Date: 2024-11-08CHONGQING ACAD OF ANIMAL SCI

Patent Information

Application Number
CN202411201210.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-29
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The existing muscle fiber morphology measurement methods have long time and are subjective in manual measurement. ImageJ software analysis requires manual parameters adjustment and high image quality requirements. Cellpose deep learning tools have high requirements for computing resources and require a large amount of training data, making it difficult to achieve fast and accurate Myosclerotic image analysis.

Method used

The pre-processing module is used to sharpen the images and equally scaled the images. The model inference module uses the FastSAM model to perform batch inference on the cropped images, and calculates the muscle fiber area and diameter through the post-processing module. By expanding the cropping and completing the muscle fiber information, it deletes The muscle fibers with incomplete edges remove repeated results, and the number of muscle fibers is flexibly processed. The model iteration module continues to update the FastSAM model.

Benefits of technology

It improves the accuracy and efficiency of muscle fiber recognition, improves the recognition accuracy and generalization ability of the FastSAM model, ensures the accuracy and efficiency of the output results, simplifies the operation process, and reduces the user's operation burden.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118918335A_ABST
    Figure CN118918335A_ABST
Patent Text Reader

Abstract

The invention discloses a muscle fiber morphology intelligent determination system, which comprises a preprocessing module, a model reasoning module and a post-processing module, and is characterized in that the preprocessing module carries out sharpening processing on an input picture; wherein the preprocessing module also performs equal-proportion cutting on the sharpened pictures, and the model reasoning module performs batch reasoning on the cut pictures and combines obtained reasoning results; and the post-processing module calculates the muscle fiber area and diameter according to the merged picture. According to the method, the muscle fiber picture is sharpened and then cut to amplify details, the edge integrity is ensured, batch reasoning, incomplete muscle fiber filtering, size adjusting and overlapped and invalid item removing are carried out on the cut small picture through the FastSAM model, the muscle fiber area and diameter are calculated, and the muscle fiber area and diameter are obtained. And secondary processing is flexibly carried out according to the number of muscle fibers, so that the accuracy and high efficiency of an output result are ensured.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to a measuring system, in particular to an intelligent measuring system for muscle fiber morphology. Background Art

[0002] Myofiber refers to muscle cells, which is a general term for motile and contractile cells in animals. Myofiber morphology plays a core role in muscle quality assessment and has a profound impact on modern animal husbandry breeding and medicine. In animal husbandry, muscle fiber morphology is a key parameter for measuring muscle quality and is directly related to the sensory, physical and biochemical properties of meat. The precise measurement of its fineness, density, area and diameter has become an important basis for judging the tenderness of meat and affecting market competitiveness. For example, the finer and denser the muscle fibers, the more tender the meat is, and the lower the shear force required for cutting, which satisfies consumers' pursuit of high-quality meat and provides a scientific basis for understanding the mechanism of meat formation and optimizing breeding strategies. In the pharmaceutical industry, muscle fiber morphology is also indispensable. It is a powerful tool for diagnosing neuromuscular diseases. It can reveal changes in the size, shape and type distribution of muscle fibers and provide key information for disease diagnosis. For example, in the assessment of muscle atrophy, morphology helps quantify the degree of atrophy and provides an accurate basis for the formulation of personalized treatment plans. In addition, in basic medical exploration, drug efficacy evaluation and metabolic disease research, muscle fiber morphology measurement also plays a bridging role, promoting the in-depth scientific research and the progress of clinical practice. Muscle fiber morphology measurement is not only a key technology for evaluating muscle quality, but also an important tool for promoting the development of animal husbandry and medical progress.

[0003] Conventional methods for measuring muscle fiber morphology mainly include manual measurement, ImageJ software analysis, and Cellpose deep learning tools. Each of these methods has its own advantages and disadvantages. Traditional manual measurement of segmented muscle fibers has high accuracy, but is time-consuming and highly subjective.

[0004] The automatic determination by ImageJ software analysis takes a relatively short time. For example, Chinese patent CNCN115855933A discloses a method for determining the diameter of pig muscle fibers based on Image J image analysis, which includes: S1, acquisition of images of pig skeletal muscle fibers; (1) acquisition of pig skeletal muscle samples; (2) preparation of paraffin sections of skeletal muscle; (3) HE staining of paraffin sections; (4) image acquisition; S2, quantitative analysis and processing of the images of S1 using Image J software; S3, calculation of muscle fiber diameter. By taking photos of the prepared paraffin sections of skeletal muscle after HE staining, and using ImageJ to process and analyze the photos, a better method for obtaining the diameter of pig skeletal muscle fibers is obtained, which reduces the influence of muscle fiber morphology changes and muscle fiber types on muscle fiber diameter results after HE staining, avoids the existing situation of long time consumption, low determination efficiency and human errors in muscle fiber measurement, and can realize rapid batch determination with simple operation. This method can quickly complete batch determination, but it requires manual adjustment of relevant parameters, such as manually selecting the muscle fiber area in the picture, and also has high requirements for image quality.

[0005] The Cellpose deep learning tool is the most accurate, such as a muscle fiber information extraction method, device, equipment and medium disclosed in Chinese patent CN118129763A, including: marking each type of muscle fiber in the panoramic scanning fluorescence image with different colors; segmenting the marked panoramic scanning fluorescence image according to color through the color space segmentation algorithm and recombining it to obtain a combined image; identifying the combined image of each type of muscle fiber through the cellpose model, and processing the combined image through Opencv and Numpy to obtain the total number of cells and the total area of ​​cells of each type of muscle fiber; according to the total number of cells and the total area of ​​cells of each type of muscle fiber, respectively obtain the number and area proportion of this type of muscle fiber in all types of muscle fibers; according to the ratio of the total area of ​​cells of each type of muscle fiber to the total number of cells, obtain the average area of ​​each cell, calculate the average diameter according to the average area, and complete the extraction of muscle fiber information. The present invention can avoid errors caused by manual operation. In this prior art, although the muscle fiber cell information can be accurately extracted, a large amount of training data is required and the computing resource requirements are also high. Summary of the invention

[0006] The present invention provides a muscle fiber morphology intelligent measurement system capable of quickly and accurately analyzing a large number of muscle fiber images, comprising a preprocessing module, a model reasoning module and a post-processing module, wherein the preprocessing module performs sharpening processing on an input image; wherein: the preprocessing module also performs equal-proportion cropping on the sharpened image, the model reasoning module performs batch reasoning on the cropped image, and merges the obtained reasoning results; and the post-processing module calculates the muscle fiber area and diameter according to the merged image.

[0007] Since the muscle fibers in the original muscle fiber images are usually small in size and their boundaries are not clear enough, in order to ensure accurate recognition of the original size images, the existing technology manually selects the muscle fiber area in the image, which is prone to errors. Therefore, in this solution, in order to improve the recognition accuracy of the muscle fiber image, the sharpened image is cropped in equal proportion. Compared with the existing technology, in this solution, by cropping the image into smaller areas, the details and boundaries of the muscle fibers can be effectively highlighted, thereby improving the recognition effect and accuracy of the model.

[0008] Preferably, after performing proportional cropping, the preprocessing module also performs extended cropping on the seams of the cropped images, and performs extended cropping in a circumferential direction with the intersection of the cropping lines as the center. Since proportional cropping will cause the muscle fibers on the cropping line to become incomplete, which will affect the accuracy of recognition, in this solution, the area around the cropping line is also extended and cropped. By extending the cropping area, the muscle fiber information lost during the cropping process can be supplemented, thereby ensuring that complete muscle fiber results can be obtained, and further ensuring the accuracy of the recognition results.

[0009] Preferably, the model inference module uses the FastSAM model to perform inference on all the small images obtained after cropping, and then scales the inference results to the original image size and merges them. This solution can improve the accuracy and efficiency of overall recognition.

[0010] Preferably, the model inference module also deletes muscle fibers with incomplete edges before proportionally amplifying the inference results. This solution can further ensure the accuracy and completeness of the overall recognition, and can also speed up the processing speed of subsequent steps.

[0011] Preferably, the post-processing module removes repeated and blank inference results. This solution can further improve the accuracy of overall recognition.

[0012] Preferably, the post-processing module calculates the single muscle fiber area and the total area according to the processed reasoning results; when calculating the single muscle fiber area, a single muscle fiber is composed of n points (x1, y1), (x2, y2), ..., (xn, yn), and the single muscle fiber pixel area can be calculated by the following formula: ; The total area of ​​the muscle fiber was calculated using the following formula: .

[0013] Preferably, the post-processing module also calculates the diameter of the muscle fiber based on the processed reasoning results: determine the projection contour of a single muscle fiber, then rotate the projection contour of the single muscle fiber with a certain step size, and after each rotation, measure the distance between the two boundary parallel lines of the projection contour, repeat the above rotation and measurement operations until all rotation directions are covered, and take the smallest measured value as the diameter of the muscle fiber.

[0014] Preferably, when the number of muscle fibers in the inference result exceeds a certain value, the preprocessing module re-crops the image to a smaller size, and then the model inference module re-infers the cropped image; when the number of muscle fibers in the inference result is less than a certain value, the model inference module performs refined inference on the image. This solution can further improve the accuracy of the results.

[0015] Preferably, a model iteration module is also included for training the FastSAM model, and after the training is completed, the performance is comprehensively evaluated through an independent training set. This solution can improve the recognition accuracy and generalization ability of the FastSAM model and ensure the stability of the FastSAM model.

[0016] Preferably, the model iteration module continuously updates and iterates the FastSAM model through systematic data collection, labeling, storage, preprocessing, training and verification. In this solution, the performance of the FastSAM model can be continuously improved through continuous updating and iteration of the FastSAM model, ultimately achieving accurate and efficient identification and classification of muscle fibers.

[0017] The present invention has the following beneficial effects: 1. The present invention sharpens the muscle fiber image and then crops it to enlarge the details and ensure the edge integrity. Then, the FastSAM model is used to perform batch inference on the cropped small images, filter out incomplete muscle fibers, adjust the size, remove overlapping and invalid items, calculate the muscle fiber area and diameter, and flexibly perform secondary processing according to the number of muscle fibers to ensure the accuracy and efficiency of the output results.

[0018] 2. The present invention can effectively highlight the details and boundaries of muscle fibers by cropping the image into smaller areas, thereby improving the recognition effect and accuracy of the model.

[0019] 3. The present invention can supplement the muscle fiber information lost in the cutting process by extending the cutting area, thereby ensuring that complete muscle fiber results can be obtained, and further ensuring the accuracy of the recognition results.

[0020] 4. The present invention can improve the recognition accuracy and generalization ability of the FastSAM model and ensure the stability of the FastSAM model.

[0021] 5. The present invention continuously updates and iterates the FastSAM model, so that the performance of the FastSAM model can be continuously improved, and ultimately the accurate and efficient identification and classification of muscle fibers can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a flow chart of an embodiment of the intelligent determination system of muscle fiber morphology of the present invention; Figure 2 This is a diagram showing the original image sharpening result in the embodiment of the intelligent measurement system for muscle fiber morphology of the present invention; Figure 3 This is a schematic diagram of image cropping; Figure 4 Schematic diagram of edge extension cropping; Figure 5 This is a schematic diagram of center expansion cropping; Figure 6 The resulting image is cropped for expansion; Figure 7 It is the FastSAM model architecture; Figure 8 This is the batch reasoning result graph; Fig. 9 The result map for filtering incomplete muscle fibers at the edge; Fig.10 This is a proportional enlarged result diagram; Fig.11 This is the result graph after merging all the inference results; Fig.12 This is the final processing result diagram; Fig.13 Display diagram for database data; Fig.14 Iteration flow chart for FastSAM model; Fig.15 It is the flow chart of manual measurement, intelligent measurement and ImageJ automatic measurement; Fig.16 Schematic diagram of the difference analysis between intelligent measurement, manual measurement and ImageJ automatic measurement results; Fig.17 This is a comparison chart between intelligent measurement and ImageJ measurement; Fig.18 Schematic diagram of the time consumption for manual measurement and ImageJ measurement using intelligent measurement. DETAILED DESCRIPTION

[0023] The following is further described in detail through specific implementation methods: 1. Definition FastSAM model: An image segmentation model that is an improved version of the SAM model, designed to increase the model's running speed while maintaining comparable performance to the original SAM model. The FastSAM model achieves a speed improvement of 50 times faster than the original SAM model by optimizing the algorithm and model architecture.

[0024] Step size: usually refers to the unit distance or increment value that a variable moves from one element to the next element in each iteration in a computer program, especially in a loop statement. In the present invention, it refers to the rotation angle.

[0025] 2. The embodiment is basically as follows Figure 1 Shown: The intelligent measurement system for muscle fiber morphology includes a preprocessing module, a model reasoning module and a post-processing module.

[0026] The preprocessing module sharpens and crops the input image. The sharpening process includes three steps: First, the input RGB color image is converted into a grayscale image to simplify processing and improve computational efficiency. Then, the image contrast is enhanced through histogram equalization to expand the dynamic range of grayscale values. Finally, the Laplacian filter is used for sharpening to highlight the edges and details in the image, such as Figure 2 By sharpening the image, the image can be enhanced and clarified, making it more suitable for subsequent analysis and application.

[0027] After the sharpening process is completed, the image is also cropped in equal proportion. By cropping the image into smaller areas, the details and boundaries of the muscle fibers can be effectively highlighted, thereby improving the recognition effect and accuracy. In this embodiment, the image resolution is 3840*1160, and the cropping method is 2 rows and 3 columns, resulting in a total of 6 images, with a single image resolution of 1280*1080. Figure 3 shown.

[0028] Since the above-mentioned proportional cutting makes the muscle fibers on the cutting line incomplete, it is necessary to extend the cutting of the picture at the seam of the horizontal and vertical pictures. By extending the cutting area, the muscle fiber information lost in the cutting process can be supplemented, thereby ensuring that the complete muscle fiber structure can be obtained. The horizontal cutting resolution is 1280*550, and the vertical cutting resolution is 650*1280. Figure 4 shown.

[0029] In order to complete the muscle fiber information lost in the proportional cropping process, it is also necessary to perform a circumferential expansion cropping with the intersection of the cropping lines as the center. In this embodiment, a 1280*1280 rectangle is cropped from the diagonal parts of the four adjacent pictures, such as Figure 5 shown.

[0030] Through the above extended cropping, a total of 9 cropped pictures are obtained, such as Figure 6 shown.

[0031] The model inference module then performs batch inference on the cropped images and merges the inference results. Specifically, the FastSAM model uses an architecture based on YOLOv8-seg, which can accurately and quickly complete the segmentation task. The FastSAM model architecture is as follows: Figure 7 shown.

[0032] In order to ensure that each small image is processed and analyzed separately, thereby improving the accuracy and efficiency of overall recognition, all cropped small images are batch-fed into the FastSAM model for inference. The inference parameters are set to an intersection over union (IoU) of 0.9, a confidence threshold of 0.1, and an image size (imgsz) of 1014. An NVIDIA L40 graphics card is used for single-card inference. The inference results are shown in Figure 2. Figure 8 shown.

[0033] During the image processing, in order to ensure the accuracy and completeness of the image data and to speed up the processing of subsequent steps, all muscle fibers with incomplete edges will be deleted. Fig. 9 shown.

[0034] After deleting all muscle fibers with incomplete edges, since the recognition results are obtained based on the cropped image size, in order to correctly merge all the processing results later, the inference results need to be proportionally enlarged to the original image size, such as Fig.10 Then all the enlarged inference results are merged, and the merged results are shown as Fig.11 shown.

[0035] The post-processing module calculates the area and diameter of the muscle fiber based on the merged image. Preferably, before the calculation, the repeated and blank reasoning results are removed. Since the same model may output inconsistent results when processing muscle fibers in the same part of different images, in this embodiment, the overlapping relationship between muscle fibers is used to remove duplicate muscle fibers, and the contour, area and center of mass of the muscle fiber segmentation results are extracted and compared pair by pair. If the area of ​​a muscle fiber is larger than that of another muscle fiber and contains its center of mass, it is marked as a duplicate. Finally, only the muscle fibers that are not marked as duplicates are retained.

[0036] In practical applications, the FastSAM model may mistakenly segment the impurities in the gullies. Therefore, in order to improve the accuracy of the segmentation results, a filtering mechanism based on the color ratio of pixels is adopted in this embodiment. Specifically, by traversing all segmentation results, the proportion of white pixels in each segmented area is calculated. If more than 70% of the pixels in a segmented area are white, the area is considered to be mainly composed of noise or impurities and is filtered out. This method effectively reduces the segmentation errors caused by impurities and significantly improves the accuracy of the overall segmentation. The final processing result is shown in Figure 2. Fig.12 shown.

[0037] Then, the post-processing module calculates the single muscle fiber area and the total area based on the processed inference results. When calculating the single muscle fiber area, a single muscle fiber is composed of n points (x1, y1), (x2, y2), ..., (xn, yn), and the pixel area of ​​a single muscle fiber can be calculated by the following formula: ; The total area of ​​the muscle fiber was calculated using the following formula: The post-processing module also calculates the diameter of the muscle fiber based on the processed reasoning results: determine the projection contour of a single muscle fiber, and then rotate the projection contour of a single muscle fiber with a certain step length. In this embodiment, 0.5 degrees is taken as an example. After each rotation, the distance between the parallel lines of the two boundaries of the projection contour is measured. Repeat the above rotation and measurement operations until all rotation directions are covered, and take the smallest measurement value as the diameter of the muscle fiber. When the diameter and area of ​​each muscle fiber are obtained, the area and diameter of each muscle fiber are automatically marked.

[0038] When the number of muscle fibers in the processing reasoning result exceeds a certain value, the preprocessing module re-crops the image to a smaller size, and then the model reasoning module re-reasons the cropped image; when the muscle fibers in the reasoning result are less than a certain value, the model reasoning module performs refined reasoning on the image. In this embodiment, if the number of detected muscle fibers exceeds 100, the image is re-cropped into small blocks of 2 rows and 3 columns to further magnify the details of the muscle fibers, and reasoning is performed again to improve accuracy. On the contrary, if the number of detected muscle fibers is less than 100, it indicates that the average area of ​​the muscle fibers in the image is large. At this time, there is no need to perform image cutting, and the muscle fiber image is directly refined and reasoned and the result is output.

[0039] Preferably, in order to standardize the output and facilitate subsequent processing, all segmentation results will be converted to COCO format. COCO format is a widely used image dataset format with the advantages of standardization and structure. The converted results can not only be conveniently returned to the front-end page of the system for processing, but also the data can be stored in the database for subsequent model secondary training and performance optimization.

[0040] The intelligent muscle fiber morphology measurement system also includes a model iteration module, which is used to train the FastSAM model. After the training is completed, the performance is comprehensively evaluated through an independent training set. Through systematic data collection, labeling, storage, preprocessing, training and verification, the FastSAM model is continuously updated and iterated.

[0041] Specifically, the system collects and classifies the original muscle fiber images, such as classifying the collected back muscles into the same category. Then, useless data is eliminated to ensure data quality, such as data that is segmented incorrectly during the segmentation process. The FastSAM model is then used to automatically annotate the muscle fiber area and diameter of the images. After manually reviewing the automatic annotation information of each muscle fiber area and diameter, the muscle fiber images that have passed the review are stored in the database to form a complete muscle fiber annotation information system. The database data is shown as follows: Fig.13 shown.

[0042] The labeled data is extracted from the database for preprocessing, including data cleaning, normalization and data enhancement. Data cleaning is used to remove noise and outliers, normalization is used to ensure data consistency, and data enhancement is used to increase the diversity during model training, thereby effectively improving the quality and robustness of model training. The preprocessed data is then used to train the FastSAM model, continuously optimizing the loss function and adjusting the hyperparameters to improve the recognition accuracy and generalization ability of the model. The hyperparameters are parameters whose values ​​are set before the learning process begins in the context of machine learning. After the model training is completed, its performance is comprehensively evaluated through an independent validation set to ensure the stability of the model. In addition, a cross-validation strategy is used to further verify the stability and reliability of the model. With the continuous influx of new data, the model is constantly updated and iterated, and the performance is continuously improved, ultimately achieving accurate and efficient recognition of muscle fibers. The model iteration flow chart is shown below. Fig.14 shown.

[0043] Comparative determination ① Simplicity of process: Based on 101 muscle fiber morphology images, the automatic measurement using conventional ImageJ1.53 software was compared with the intelligent measurement system in this embodiment. First, the processes of the two automatic measurement methods were compared. Fig.15As shown in the figure, the automatic measurement function process of ImageJ still requires users to manually adjust the image parameters to optimize the recognition effect, and this pre-step is relatively complex and time-consuming. In contrast, the intelligent measurement system in this embodiment shows significant ease of operation, and its overall process is more intuitive and efficient, greatly simplifying the entire process from image import to data output, effectively reducing the user's operating burden, and improving measurement efficiency and accuracy.

[0044] ② Reliability of measurement results: In the detailed measurement of 101 muscle fiber morphology images, key parameters such as the number of muscle fibers, average area, and measurement time of each image were examined. In order to evaluate the data consistency between different measurement methods, the average area results and muscle fiber number obtained by the ImageJ automatic measurement and the intelligent measurement system in this embodiment were analyzed with the results of manual measurement. Fig.16 and Fig.17 As shown. The results show that the results of the intelligent determination in the present embodiment are indistinguishable from the results of the manual determination, and the results of the automatic determination by ImageJ are larger. This may be because when processing cell images in ImageJ, it is first necessary to convert the image into a grayscale image, and segment the cells by accurately adjusting the threshold. However, if the threshold is set too low, not only the cell area will be identified, but also the background or other non-cellular areas may be mistakenly included therein, directly causing the calculated cell area to be larger. In addition, the noise in the image, such as halo, granular impurities, etc., also can interfere with the accuracy of the threshold segmentation, making the segmented cell area larger. Furthermore, when there is overlap between cells, if the segmentation algorithm cannot effectively distinguish them, the calculated single cell area will also be larger.

[0045] The results of ImageJ automatic determination are significantly higher than those of the intelligent determination system and manual determination in this embodiment (P<0.01). The results of Spearman correlation analysis are shown in Table 1. The analysis results show that the intelligent determination system in this embodiment and the ImageJ manual determination method show a high correlation (r=0.93, P<0.01), indicating that the two have good consistency in measuring the average area of ​​muscle fibers, verifying the accuracy and reliability of the intelligent determination system. In contrast, although the measurement results of ImageJ automatic determination and the intelligent determination system in this embodiment are also significantly correlated (r=0.80, P<0.01), the correlation is slightly lower than the former, which may be attributed to the limited ability of ImageJ automatic determination to recognize complex image details without too much manual intervention. Further, the results of ImageJ automatic measurement and manual measurement also show a significant correlation (r=0.82, P<0.01), but this correlation level is still lower than the correlation between the intelligent measurement system and the ImageJ manual measurement, proving that the intelligent measurement system in this embodiment has advantages in both measurement accuracy and efficiency.

[0046] Table 1 Correlation between the results of intelligent determination, ImageJ automatic determination and manual determination Intelligent measurement Manual determination ImageJ automatic determination Intelligent measurement 1 Manual determination <![CDATA[0.93 ** ]]> 1 ImageJ automatic determination <![CDATA[0.80 ** ]]> <![CDATA[0.82 ** ]]> 1 Note: ** represents P <0.01 ③ Time comparison: When comparing the average time required for measuring a single muscle fiber slice by different methods, the test results showed significant efficiency differences, such as Fig.18 As shown, the results show that the time taken to process a single slice using the intelligent measurement system in this embodiment is significantly lower than the time taken by professionals using conventional manual measurement. The time taken to measure a single slice is reduced by 11 minutes, which is 2 minutes less than the average time taken to automatically measure a single slice using ImageJ.

[0047] The above is only an embodiment of the present invention. The common sense such as the known specific structure and characteristics in the scheme is not described in detail here. The ordinary technicians in the relevant field know all the common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all the existing technologies in the field, and have the ability to apply the conventional experimental means before that date. The ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the protection scope of the present invention, which will not affect the effect of the implementation of the present invention and the practicality of the patent. The protection scope required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A muscle fiber morphology intelligent measurement system, comprising a preprocessing module, a model reasoning module and a post-processing module, wherein the preprocessing module performs sharpening processing on an input image; the system is characterized in that: The pre-processing module also performs proportional cropping on the sharpened images, the model inference module performs batch inference on the cropped images, and merges the obtained inference results; the post-processing module calculates the muscle fiber area and diameter based on the merged images.

2. The muscle fiber morphology intelligent measurement system according to claim 1, characterized in that: After performing the proportional cropping, the preprocessing module further performs expansion cropping on the images at the seams of the cropped images, and performs expansion cropping in a circumferential direction with the intersection of the cropping lines as the center.

3. The muscle fiber morphology intelligent measurement system according to claim 2, characterized in that: The model inference module uses the FastSAM model to perform inference on all the small images obtained after cropping, and then scales the inference results to the original image size and merges them.

4. The muscle fiber morphology intelligent measurement system according to claim 3, characterized in that: The model reasoning module also deletes muscle fibers with incomplete edges before proportionally enlarging the reasoning results.

5. The muscle fiber morphology intelligent measurement system according to claim 4, characterized in that: The post-processing module removes duplicate and blank inference results.

6. The muscle fiber morphology intelligent measurement system according to claim 5, characterized in that: The post-processing module calculates the single muscle fiber area and the total area according to the processed reasoning results; when calculating the single muscle fiber area, a single muscle fiber is composed of n points (x1, y1), (x2, y2), ..., (xn, yn), and the single muscle fiber pixel area can be calculated by the following formula: ; The total area of ​​the muscle fiber was calculated using the following formula: 。 7. The intelligent muscle fiber morphology measurement system according to claim 6, characterized in that: The post-processing module also calculates the diameter of the muscle fiber based on the processed reasoning results: determine the projection contour of a single muscle fiber, then rotate the projection contour of the single muscle fiber with a certain step length, measure the distance between the two boundary parallel lines of the projection contour after each rotation, repeat the above rotation and measurement operations until all rotation directions are covered, and take the smallest measured value as the diameter of the muscle fiber.

8. The intelligent muscle fiber morphology measurement system according to claim 7, characterized in that: When the number of muscle fibers in the inference result exceeds a certain value, the preprocessing module re-crops the image to a smaller size, and then the model inference module re-infers the cropped image; when the number of muscle fibers in the inference result is less than a certain value, the model inference module performs refined inference on the image.

9. The intelligent muscle fiber morphology measurement system according to any one of claims 1 to 8, characterized in that: It also includes a model iteration module for training the FastSAM model and, after training, comprehensively evaluating the performance through an independent training set.

10. The muscle fiber morphology intelligent measurement system according to claim 9, characterized in that: The model iteration module continuously updates and iterates the FastMAM model through systematic data collection, labeling, storage, preprocessing, training and verification.

Citation Information

Patent Citations

  • Method and system for classifying and identifying muscle fiber cell images based on ATPase dyeing

    CN114419359A

  • Pig muscle fiber diameter measuring method based on Image J image analysis

    CN115855933A

  • Muscle fiber information extraction method, device, equipment and medium

    CN118229763A

Cited By

  • Medical cell detection method

    CN120125582A