Molding quality identification method of medicine balloon

Abstract

The invention discloses a forming quality identification method for a drug balloon. The method comprises the following steps: step 1, preprocessing a picture; step 2, performing image reconstruction by using a variational auto-encoder VAE; step 3, performing picture subtraction; 4, judging whether a quality defect exists or not, determining whether the drug balloon has the defect or not through analysis of the difference image, and clearly positioning the position of the drug balloon forming quality defect; the method has the beneficial effects that the VAE is used for image reconstruction, fine features in the image can be accurately extracted and analyzed, the recognition accuracy of complex and tiny defects is remarkably improved, the risks of missing detection and false detection are reduced, a difference image can be directly generated, any existing defect can be visually displayed in a highlight manner, and the defect recognition accuracy is improved. The abnormal area on the surface of the drug balloon can be quickly and visually identified, and an operator can quickly judge and make a decision conveniently, so that the requirement of modern large-scale production is met.

Description

Technical Field

[0001] This invention relates to the field of drug-eluting balloon technology, and more specifically to a method for identifying the molding quality of a drug-eluting balloon. Background Technology

[0002] Drug-eluting balloons play a crucial role in interventional cardiovascular treatment, and their manufacturing quality is critical to treatment outcomes. The primary function of a drug-eluting balloon is to directly deliver medication to the lesion site to promote angiogenesis. However, various defects can occur during the manufacturing process (such as air bubbles, uneven coating, and material damage), which not only affect balloon performance but may also pose potential risks to patients.

[0003] Traditional quality inspection methods mainly rely on manual visual quality inspection and simple algorithm-based automatic machine inspection.

[0004] Manual inspection involves visually examining and measuring drug-eluting balloons by experienced technicians. However, manual inspection is limited by the operator's experience, resulting in strong subjectivity, large errors, and inconsistencies in the test results. Furthermore, the process is time-consuming and inefficient, failing to meet the high-efficiency requirements of large-scale production lines. It also requires a large amount of human resources, increasing production costs.

[0005] Traditional algorithm-based automatic detection uses simple image processing techniques (such as edge detection and threshold segmentation) to analyze drug balloon images. When dealing with complex defects, the accuracy is insufficient, and it is easy to miss or falsely detect. In addition, the processing speed and efficiency are low, and real-time detection cannot be achieved. Furthermore, existing automatic detection equipment is often expensive and difficult to maintain, which increases the operational burden on enterprises.

[0006] This makes it an urgent technical problem to be solved in large-scale production how to efficiently and accurately detect the molding quality of drug-eluting balloons. Summary of the Invention

[0007] To overcome the problems of low efficiency, large error, high cost, and inability to achieve real-time detection in traditional drug-eluting balloon quality testing methods, this invention provides a method for identifying the molding quality of drug-eluting balloons.

[0008] The technical solution adopted by the present invention to achieve the above objectives includes the following steps:

[0009] Step 1: Image preprocessing;

[0010] Step 2: Apply Variational Autoencoder (VAE) for image reconstruction;

[0011] Step 3: Subtract the images;

[0012] Step 4: Determine if there are quality defects. Through the analysis of difference images, confirm whether there are defects in the drug-eluting balloon and pinpoint the location of the quality defects in the drug-eluting balloon molding.

[0013] Preferably, step 1 specifically includes the following steps:

[0014] Step 1.1: Use a high-resolution camera to photograph the drug-eluting balloon to obtain clear, raw images;

[0015] Step 1.2: Adjust the image size to match the input layer size;

[0016] Step 1.3: Perform noise reduction, grayscale conversion, grayscale inversion, and enhancement preprocessing on the acquired image.

[0017] Preferably, the variational autoencoder (VAE) described in step 2 includes an encoder and a decoder.

[0018] Preferably, step 2 specifically includes the following steps:

[0019] Step 2.1: Compress the input image and convert it into a latent space representation;

[0020] Step 2.2: The decoder reconstructs the original image from the variables in the latent space.

[0021] Preferably, step 3 involves subtracting the image reconstructed by the VAE from the original image input to the VAE to generate a new difference image.

[0022] Preferably, step 4 involves generating and analyzing difference images to pinpoint the location of defects in the drug-eluting balloon molding quality; the image processing algorithm performs quantitative analysis on the highlighted areas in the difference images to identify and classify possible defect types.

[0023] Preferably, in step 2, the preprocessed drug balloon image is input into the trained variational autoencoder (VAE) for image reconstruction. The encoder automatically extracts deep features of the image and inputs these features into the decoder. The decoder reconstructs the image to generate a reconstructed image that is as similar as possible to the original image.

[0024] Preferably, the encoder includes a graphics input layer, a convolutional layer and an activation function, a fully connected layer, and a sampling layer; the decoder includes a feature input layer, a projection and reshaping layer, a transposed convolutional layer and an activation function, and an output layer; the convolutional layers and activation functions are multiple convolutional layers and activation functions arranged alternately; the activation function is the Modified Linear Unit (ReLU) activation function.

[0025] Preferably, in step 2.1, the encoder has multiple convolutional layers and ReLU activation functions arranged alternately; these convolutional layers progressively downsample the image to extract deep features using a 3×3 convolutional kernel and a stride of 2, gradually increasing the number of channels from 32 to 256, and using "same" padding to ensure that the output size is consistent with the input; subsequently, the image features are connected to a fully connected layer with 256 latent channels, which generates the mean and variance parameters of the latent space; the last layer of the encoder is a sampling layer, which converts the mean and variance into latent space variables through random sampling, supporting the randomness of the generative model; step 2.2: through an input layer... The input image is received from the latent space of the encoder. Next, the latent variables are projected into a high-dimensional feature space and reshaped into tensors suitable for further processing. The decoder consists of multiple transposed convolutional layers and ReLU activation functions arranged alternately to progressively enlarge and reconstruct the image. The number of channels in these deconvolutional layers gradually decreases from 256 to 32, the kernel size is reduced from 7×7 to 3×3, and enlargement is achieved with a stride of 2. Finally, the decoder outputs the reconstructed image through deconvolution. The output layer applies a sigmoid activation function to restrict the output pixel values ​​to the range [0,1]. The VAE generates a high-quality reconstructed image by learning the latent distribution of the input image.

[0026] Preferably, the specific mechanism of step 3 includes the following steps:

[0027] Step 3.1: When the input VAE image is normal and without defects, the VAE reconstructed image will also appear normal. At this time, the two images are almost identical; the difference image generated after subtracting the two images will be an image that is close to black.

[0028] Step 3.2: When there are defects in the input VAE image, but the VAE-reconstructed image still appears as a normal image; the difference image generated by subtracting the two images will highlight the parts with quality problems.

[0029] The beneficial effects of this invention are:

[0030] 1. Improved detection accuracy: This invention utilizes a variational autoencoder (VAE) for image reconstruction, which can accurately extract and analyze subtle features in the image, significantly improving the accuracy of identifying complex and minute defects and reducing the risk of missed and false detections.

[0031] 2. Defect Highlighting: By subtracting the original image from the VAE-reconstructed image, this invention directly generates a difference image, visually highlighting any existing defects. This method can quickly and intuitively identify abnormal areas on the surface of drug-eluting balloons without complex manual analysis or additional inspection steps. This intuitive results display interface facilitates rapid judgment and decision-making by operators.

[0032] 3. Improved detection efficiency: VAE can quickly process and reconstruct images. This invention can complete the quality inspection of drug-eluting balloons in a short time and has the potential for real-time analysis, which helps to carry out rapid screening on the production line to meet the needs of modern large-scale production.

[0033] 4. High adaptability: The variational autoencoder model has good generalization ability for images of different types and sizes, making it suitable for drug balloon detection of various specifications, and has strong flexibility and adaptability.

[0034] 5. Reduce total costs: By reducing reliance on manual inspections, labor costs are reduced, overall production efficiency is improved, business operating costs are lowered, and economic benefits are increased.

[0035] 6. It achieves efficient encoding of input images and accurate reconstruction of the latent space, making it widely applicable to various image generation and feature learning applications. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in this invention 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 invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0037] Figure 1 This is a flowchart of a method for identifying the molding quality of a drug-eluting balloon.

[0038] Figure 2 This is a schematic diagram of the variational autoencoder (VAE). Detailed Implementation

[0039] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0040] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the present invention or its application or use. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0042] like Figure 1 As shown, this embodiment provides a method for identifying the molding quality of drug-eluting balloons. First, a high-resolution camera is used to photograph the drug-eluting balloon to obtain a clear original image, ensuring that the photograph is taken under good lighting conditions to improve image quality. Then, the image size is adjusted to match the input layer size. Next, the acquired image undergoes preprocessing such as denoising, grayscale conversion, grayscale inversion, and enhancement to improve the clarity and accuracy of subsequent analysis.

[0043] Then, a variational autoencoder (VAE) is applied for image reconstruction. This invention relates to a variational autoencoder (VAE) architecture for image processing, comprising two main components: an encoder and a decoder.

[0044] like Figure 2As shown, the encoder section aims to compress the input image and convert it into a latent space representation. The encoder contains multiple convolutional layers and alternating ReLU (Modified Linear Unit) activation functions. These convolutional layers progressively downsample the image using 3×3 kernels and a stride of 2 to extract deep features, gradually increasing the number of channels from 32 to 256, and employing "same" padding to ensure the output size matches the input. Subsequently, the image features are concatenated to a fully connected layer with 256 latent channels, which generates the mean and variance parameters of the latent space. The final layer of the encoder is a sampling layer that converts the mean and variance into latent space variables through random sampling, supporting the randomness of the generative model. The decoder section aims to reconstruct the original image from the variables in the latent space. It receives the latent space input from the encoder through an input layer. Next, the latent variables are projected into a high-dimensional feature space and reshaped into tensors suitable for further processing. The decoder contains multiple deconvolutional (transposed convolutional) layers and alternating ReLU activation functions for progressively upscaling and reconstructing the image. The number of channels in these deconvolutional layers gradually decreases from 256 to 32, the kernel size is reduced from 7×7 to 3×3, and a stride of 2 is used for scaling. Finally, the decoder outputs the reconstructed image through deconvolution operations. To suit the image representation, a sigmoid activation function is applied to the output layer, restricting the output pixel values ​​to the range [0,1]. The VAE can generate high-quality reconstructed images by learning the latent distribution of the input image. Through this architecture, this invention achieves efficient encoding of the input image and accurate reconstruction of the latent space, making it widely applicable to various image generation and feature learning applications.

[0045] Next, the preprocessed drug-eluting balloon image is input into the trained variational autoencoder (VAE) for image reconstruction. The encoder automatically extracts deep features of the image and inputs these features into the decoder, which then reconstructs the image.

[0046] Next, the image reconstructed by the VAE is compared with the original image input to the VAE. A new difference image is generated by subtracting the two images. The difference image effectively highlights the areas of difference between the original and reconstructed images. During image reconstruction and comparative analysis, comparing the reconstructed image with the original image input to the VAE can effectively determine the forming quality of the drug-eluting balloon. The specific mechanism is as follows: when the input VAE image is normal and without defects, the reconstructed image will also appear normal, at which point the two images are almost identical. Through comparative analysis, the difference image generated after subtracting the two images will be a near-black image, indicating that there is almost no difference between the original and reconstructed images, and the drug-eluting balloon has no quality problems.

[0047] However, if the input VAE image contains defects, while the reconstructed image from the VAE still appears as a normal image, the subtraction operation will highlight the defective parts. In this case, distinct highlighted areas will appear in the difference image, representing anomalies in the input image relative to the reconstructed image. These highlighted areas directly indicate potential quality defects during the drug delivery system formation process.

[0048] Finally, by generating and analyzing the difference images, the location of defects in the drug-eluting balloon molding quality can be clearly identified. Further image processing algorithms can perform quantitative analysis on salient regions in the difference images to identify and classify possible defect types.

[0049] Through the above steps, the present invention significantly improves the detection accuracy of drug-eluting balloons, and can quickly distinguish whether the forming quality of drug-eluting balloons meets the requirements, thereby ensuring the forming quality of drug-eluting balloons.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. 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. Such 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 identifying the molding quality of a drug-eluting balloon, characterized in that: Includes the following steps: Step 1: Image preprocessing; Step 2: Apply Variational Autoencoder (VAE) for image reconstruction; Step 3: Subtract the images; Step 4: Determine if there are quality defects. Through the analysis of difference images, confirm whether there are defects in the drug-eluting balloon and pinpoint the location of the quality defects in the drug-eluting balloon molding.

2. The method for identifying the molding quality of a drug-eluting balloon according to claim 1, characterized in that: Step 1 specifically includes the following steps: Step 1.1: Use a high-resolution camera to photograph the drug-eluting balloon to obtain clear, raw images; Step 1.2: Adjust the image size to match the input layer size; Step 1.3: Perform noise reduction, grayscale conversion, grayscale inversion, and enhancement preprocessing on the acquired image.

3. The method for identifying the molding quality of a drug-eluting balloon according to claim 1, characterized in that: The variational autoencoder (VAE) described in step 2 includes an encoder and a decoder.

4. The method for identifying the molding quality of a drug-eluting balloon according to claim 1, characterized in that: Step 2 specifically includes the following steps: Step 2.1: Compress the input image and convert it into a latent space representation; Step 2.2: The decoder reconstructs the original image from the variables in the latent space.

5. The method for identifying the molding quality of a drug-eluting balloon according to claim 1, characterized in that: Step 3 involves subtracting the image reconstructed by the VAE from the original image input to the VAE to generate a new difference image.

6. The method for identifying the molding quality of a drug-eluting balloon according to claim 1, characterized in that: Step 4 involves generating and analyzing difference images to pinpoint the location of defects in the drug-eluting balloon molding process; the image processing algorithm performs quantitative analysis on the highlighted areas in the difference images to identify and classify possible defect types.

7. The method for identifying the molding quality of a drug-eluting balloon according to claim 3, characterized in that: In step 2, the preprocessed drug balloon image is input into the trained variational autoencoder (VAE) for image reconstruction. The encoder automatically extracts the deep features of the image and inputs these features into the decoder. The decoder reconstructs the image and generates a reconstructed image that is as similar as possible to the original image.

8. The method for identifying the molding quality of a drug-eluting balloon according to claim 3, characterized in that: The encoder includes a graphics input layer, convolutional layers and activation functions, a fully connected layer, and a sampling layer; the decoder includes a feature input layer, a projection and reshaping layer, a transposed convolutional layer and activation function, and an output layer; the convolutional layers and activation functions consist of multiple convolutional layers and activation functions arranged alternately. The activation function is the modified linear unit ReLU activation function.

9. The method for identifying the molding quality of a drug-eluting balloon according to claim 3, characterized in that: The encoder in step 2.1 has multiple convolutional layers and ReLU activation functions arranged alternately; these convolutional layers use a strategy of 3×3 convolutional kernels and stride of 2 to progressively downsample the image to extract deep features, the number of channels gradually increases from 32 to 256, and "same" padding is used to ensure that the output size is consistent with the input. Subsequently, the image features are connected to a fully connected layer with 256 latent channels, which generates the mean and variance parameters of the latent space; the last layer of the encoder is a sampling layer, which converts the mean and variance into latent space variables through random sampling, supporting the randomness of the generative model; step 2.2: receiving latent space input from the encoder through an input layer; Next, the latent variables are projected into a high-dimensional feature space and reshaped into tensors suitable for further processing. The decoder consists of multiple transposed convolutional layers and ReLU activation functions arranged alternately to gradually enlarge and reconstruct the image. The number of channels in these deconvolutional layers gradually decreases from 256 to 32, the kernel size is reduced from 7×7 to 3×3, and the enlargement is achieved with a stride of 2. Finally, the decoder outputs the reconstructed image through deconvolution. The output layer applies the sigmoid activation function to restrict the output pixel values ​​to the range [0,1]. The VAE generates a high-quality reconstructed image by learning the latent distribution of the input image.

10. The method for identifying the molding quality of a drug-eluting balloon according to claim 5, characterized in that: The specific mechanism of step 3 includes the following steps: Step 3.1: When the input VAE image is normal and without defects, the VAE reconstructed image will also appear normal. At this time, the two images are almost identical; the difference image generated after subtracting the two images will be an image that is close to black. Step 3.2: When there are defects in the input VAE image, but the VAE-reconstructed image still appears as a normal image; the difference image generated by subtracting the two images will highlight the parts with quality problems.

Images