Microscopic image adaptive deconvolution method and device, storage medium and imaging system

Abstract

The invention discloses a microscopic image adaptive deconvolution method and device, a storage medium and an imaging system, and the method comprises the steps: obtaining two image pairs of each sampling region in a plurality of different sampling regions of a target sample as samples, and forming a training set through all the samples; training an adaptive deconvolution model containing primary and secondary branches through the training set; inputting an image to be deconvolved into a main branch in the trained adaptive deconvolution model to obtain a corresponding deconvolution image; the device and the storage medium are used for implementing the method process. The system comprises an optical imaging system for image acquisition, and a control and data processing system for controlling the optical imaging system and implementing the process of the method. According to the method, the deconvolution network, the optical point spread function network and the loss function are meticulously designed, deep feature representation of training data and microscopic imaging optical degradation representation contained in the data are automatically mined, and the purpose of self-adaptive deconvolution is achieved.

Description

Technical Field

[0001] This invention relates to the field of microscopic image processing methods, specifically to an adaptive deconvolution method, device, storage medium, and imaging system for microscopic images. Background Technology

[0002] Optical microscopy, with its gentle, rapid, simple sample preparation, and convenient operation, is widely used in research in life sciences, materials science, and industrial inspection. However, due to the diffraction limit, the imaging resolution of ordinary optical microscopes is limited to above 200 nm, making it impossible to observe more detailed and critical material structures. In recent years, with the continuous improvement and advancement of computer hardware, computational optical imaging has made great strides. Through algorithmic post-processing, the resolution of microscopic images can be improved by 1.5-2 times, which is of great significance for the study of microscopic subcellular interactions and microscopic defects in materials.

[0003] Existing deconvolution algorithms for improving the resolution of microscopic images are mainly divided into two categories: (1) deconvolution methods based on analytical models and (2) deconvolution methods based on deep learning.

[0004] Deconvolution methods based on analytical models, such as Richardson-Lucy (RL) deconvolution, Fast Iterative Soft Thresholding Deconvolution (FISTA), and Sparse deconvolution, first artificially impose noise simplification and regularization constraints (such as continuity, sparsity, and symmetry constraints) on the optical imaging model. Then, they improve image resolution through continuous parameter tuning and inversion iterations. These algorithms are easily affected by parameter settings, artificial model constraints, image noise levels, and the number of iterations, and their generalization and robustness need improvement.

[0005] Deep learning-based deconvolution algorithms, such as DFCAN and Meta-rLLS-VSIM, learn direct mappings between data end-to-end, avoiding the introduction of constraints from artificial physical models and empirical parameter settings. This results in superior generalization and robustness compared to analytical model algorithms. However, deep learning algorithms require the prior collection of large amounts of paired training data to build the network, which is typically very expensive and time-consuming.

[0006] The ZS-DeconvNet algorithm employs a self-supervised approach to construct and train the deconvolutional network, which alleviates the aforementioned problems to some extent. However, ZS-DeconvNet requires prior information about the point spread function (PSF) of the optical imaging system to be provided to the neural network in order to incorporate the physical priors of the optical imaging process into the network training. Obtaining the PSF is often accompanied by random noise interference and system aberrations, which in turn affects the final performance of ZS-DeconvNet, producing a series of deconvolution artifacts such as ringing effects. Summary of the Invention This invention provides an adaptive deconvolution method, device, storage medium, and imaging system for microscopic images, to solve the problems of deconvolution artifacts and deconvolution instability caused by mismatched point spread function priors in the existing ZS-DeconvNet algorithm for microscopic image deconvolution.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] The adaptive deconvolution method for microscopic images is as follows:

[0009] Images of multiple different sampling regions of the target sample are acquired. Multiple image pairs are constructed based on the acquired images. Each image pair includes two images of the same sampling region. Different image pairs have different sampling regions. Each image pair is used as a sample, and the training set is composed of these samples.

[0010] An adaptive deconvolution model is trained using a training set. The adaptive deconvolution model includes a deconvolution network and an optical point spread function network. The training process is as follows:

[0011] The original image of each sample in the training set is input into the deconvolution network. The deconvolution network obtains the deconvolution image of the original image of each sample. At the same time, an optical point spread function network generates a prior blur kernel. The deconvolution image of the original image of each sample and the corresponding prior blur kernel are convolved to obtain the optically degraded image. Based on the target image, the corresponding deconvolution image, and the optically degraded image of each sample, the loss error value is calculated using a loss function, and the gradient of the loss error value is obtained. The gradient is backpropagated to update the parameters of the deconvolution network and the point spread function network. Thus, through repeated iterative training with multiple samples in the training set, a trained deconvolution network and a point spread function network are obtained, forming a trained adaptive deconvolution model.

[0012] The unconvolution image of the target region of the target sample is obtained, and the unconvolution image is input into the deconvolution network in the trained adaptive deconvolution model. The deconvolution network outputs the corresponding unconvolution image.

[0013] Furthermore, images of each sampling region of the target sample are acquired under high-power illumination and low-power illumination. The image under high-power illumination is the image with a high signal-to-noise ratio for each sampling region, and the image under low-power illumination is the image with a low signal-to-noise ratio for each sampling region. The image with a high signal-to-noise ratio for each sampling region is used as the target image, and the image with a low signal-to-noise ratio is used as the original image. The target image and the original image of each sampling region are used to form an image pair. Each image pair is used as a sample, and each sample is used to form a supervised modality training set.

[0014] Then, the adaptive deconvolution model is trained in a supervised manner using a supervised modal training set.

[0015] Furthermore, images of multiple different sampling regions of the target sample under low-power illumination are acquired. Data amplification is performed on the image of each sampling region to obtain two images with different signal-to-noise ratios for each sampling region. One image is randomly selected from the two images of each sampling region as the original image and the other image as the target image. The target image and the original image in each sampling region constitute an image pair. Each image pair is used as a sample, and each sample constitutes a self-supervised modality training set.

[0016] Then, the adaptive deconvolution model is trained using a self-supervised modal training set.

[0017] Furthermore, the deconvolution network in the adaptive deconvolution model can be any convolutional neural network or any Vision Transformer.

[0018] Furthermore, the optical point spread function network in the adaptive deconvolution model can be any convolutional neural network or any VisionTransformer.

[0019] Furthermore, the loss function during training of the adaptive deconvolution model includes a fidelity loss term, a Hessian loss term, and a perceptual loss term.

[0020] Furthermore, when the training reaches the convergence of the loss error value calculated by the loss function, the training ends, and the trained adaptive deconvolution model is obtained.

[0021] An electronic device includes a processor and a memory, wherein program instructions in the memory are read and executed by the processor to perform the aforementioned adaptive deconvolution method for microscopic images.

[0022] A storage medium storing program instructions, which, when read and executed, perform the aforementioned adaptive deconvolution method for microscopic images.

[0023] An adaptive deconvolution microscopy imaging system, comprising:

[0024] An optical imaging system (100) performs optical imaging on a target sample and acquires images of various regions on the target sample;

[0025] The control and data processing system (200) controls the optical imaging system (100), acquires the image obtained by the optical imaging system (100), and implements the above-mentioned adaptive deconvolution method for microscopic images.

[0026] Compared with the prior art, the advantages of the present invention are:

[0027] This invention, through the careful design of deconvolution networks, optical point spread function networks, and loss functions, automatically mines the deep feature representations of training data and the microscopic imaging optical degradation representations contained within the data, achieving the purpose of adaptive deconvolution. This avoids problems such as deconvolution artifacts and deconvolution instability caused by mismatched point spread function priors. Attached Figure Description

[0028] Figure 1 This is a basic framework diagram of the adaptive deconvolution microscopic imaging system in this embodiment of the invention.

[0029] Figure 2 This is a diagram of the adaptive deconvolution model in an embodiment of the present invention.

[0030] Figure 3 This is a basic principle diagram of adaptive deconvolution network training under supervised training mode in this embodiment of the invention.

[0031] Figure 4 This is a basic principle diagram of adaptive deconvolution network training under the self-supervised training mode in this embodiment of the invention.

[0032] Figure 5 This is a diagram illustrating the inference process of the adaptive deconvolution microscopic imaging system in the inference mode of this invention. Detailed Implementation

[0033] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0034] Example 1

[0035] This embodiment discloses an adaptive deconvolution method for microscopic images, the process of which is as follows:

[0036] Step 1: Acquire images of multiple different sampling areas of the target sample.

[0037] In this embodiment, each sampling region of the target sample is sequentially illuminated by two light sources of different power levels, and images of each sampling region under the two different power illuminations are acquired, forming an image pair for each sampling region. In each image pair, the image under the higher power light source has a higher signal-to-noise ratio and is used as the target image, while the image under the lower power light source has a lower signal-to-noise ratio and is used as the original image. Each sampling region's target image and original image constitute an image pair. Each image pair of the sampling region is used as a sample. For any sample of the i-th sampling region, let the target image be... The original image is , =1, 2, 3, …, , It is an integer greater than or equal to 2. This represents the total number of sampling regions. The supervised modality training set, composed of these samples, is used for subsequent supervised modality training of the adaptive deconvolution model.

[0038] Alternatively, in this embodiment, each sampling region of the target sample is placed under low-power light source illumination to acquire an image of each sampling region under low-power illumination. Then, the image of each sampling region is augmented using Neighbor2Neighbor technology (see the published paper Neighbor2Neighbor: Self-Supervised Denoising from Single Noisy Images by Tao Huang, Songjiang Li, Xu Jia, Huchuan Lu, and Jianzhuang Liu) or Recorded2Recorrupted technology (see the published paper Recorded-to-Recorrupted: Unsupervised Deep Learning for Image Denoising by Tongyao Pang, Huan Zheng, Yuhui Quan, and Hui Ji) to generate two images with different signal-to-noise ratios. One of these images is randomly selected as the target image, and the other as the original image. Each sampling region's target image and original image constitute an image pair. Each sampling region's image pair is used as a sample. Let the image of any acquired i-th sampling region be... The target image generated after data amplification is The original image is Then the corresponding sample of any i-th sampling region contains the original image. Target image The self-supervised modality training set is composed of individual samples, and this self-supervised modality training set is used for subsequent self-supervised modality training of the adaptive deconvolution model.

[0039] Step 2: Generate an adaptive deconvolution model, such as... Figure 2 As shown, the adaptive deconvolution model includes a deconvolution network and a point spread function network. The adaptive deconvolution model is trained using the training set obtained in step 1 to obtain a trained adaptive deconvolution model.

[0040] In this embodiment, the adaptive deconvolution model consists of a main branch and a secondary branch. The main branch is a deconvolution network, whose main function is to deconvolve the microscopic image; the secondary branch is a point spread function network, whose main function is to calculate the prior blur kernel PSF.

[0041] As an example, the deconvolutional network of the main branch adopts a typical U-shaped convolutional network architecture (U-net), while the point spread function network of the secondary branch adopts a residual convolutional neural network architecture (Res-net). The main parameter reference values ​​for the adaptive deconvolution model are shown in Table 1.

[0042] Table 1. Parameter Table of Adaptive Deconvolution Model

[0043] It should be noted that although this embodiment provides an example of an adaptive deconvolution model, those skilled in the art should understand that the main branch and secondary branch of the adaptive deconvolution model in this embodiment can be constructed using any suitable neural network. Adaptive deconvolution models that select other suitable neural networks to form the main and secondary branches should also be considered to fall within the protection scope of this invention.

[0044] In this embodiment, as shown Figure 3 As shown, the process of supervising the adaptive deconvolution model using the supervised modality training set obtained in step 1 is as follows:

[0045] First, the weights of each branch of the adaptive deconvolution model are randomly initialized. Within one training cycle, b groups of samples are randomly selected from the supervised modality training set, and the original image of each sample in the b groups is... The input is fed into the deconvolutional network of the main branch. =1, 2, …, b), and obtain the original image of each sample through a convolutional network. Deconvolution image Simultaneously, the current prior fuzzy kernel is generated through a point spread function network. Take the original image of each sample. Deconvolution image and the corresponding prior fuzzy kernel Convolution yields an optically degraded image. Target image based on each sample , corresponding deconvolution image and optically degraded images The loss error value is calculated using a loss function, and its gradient is obtained. This gradient is then backpropagated to update the parameters of the deconvolution network and the point spread function network. Following this process, b groups of samples are randomly selected from the training set for training in each training cycle. This process is repeated iteratively until the loss error value converges. Training then ends, resulting in an adaptive deconvolution model composed of the trained deconvolution network and the point spread function network. The parameters of the deconvolution network and the point spread function network in the adaptive deconvolution model are stored for subsequent deployment of the inference modality.

[0046] In this embodiment, as shown Figure 4 As shown, the process of using the self-supervised modality training set obtained in step 1 to perform self-supervised modality training on the adaptive deconvolution model is as follows:

[0047] First, the weights of each branch of the adaptive deconvolution model are randomly initialized. Within one training cycle, b groups of samples are randomly selected from the self-supervised modality training set. The original image of each sample in these b groups is then... The input is fed into the deconvolutional network of the main branch. =1, 2, …, b), and the original image of each sample is obtained by deconvolutional network. Deconvolution image Simultaneously, the current prior fuzzy kernel is generated through a point spread function network. Take the original image of each sample. Deconvolution image and the corresponding prior fuzzy kernel Convolution yields an optically degraded image. Target image based on each sample , corresponding deconvolution image and optically degraded images The loss error value is calculated using a loss function, and its gradient is obtained. This gradient is then backpropagated to update the parameters of the deconvolution network and the point spread function network. Following this process, b groups of samples are randomly selected from the training set for training in each training cycle. This process is repeated iteratively until the loss error value converges. Training then ends, resulting in an adaptive deconvolution model composed of the trained deconvolution network and the point spread function network. The parameters of the deconvolution network and the point spread function network in the adaptive deconvolution model are stored for subsequent deployment of the inference modality.

[0048] In this embodiment, the loss function used in the two training methods described above is... The two training methods are identical, both including fidelity loss, Hessian loss, and perceptual loss. The target image for both training methods is set to... Both deconvolution images are set to Optical degradation images are all set to The fidelity loss term is the target image. With optically degraded images The first norm of the difference; the Hessian loss term, i.e., the deconvolution of the image. The first norm of the second gradient; the perceptual loss term is used to calculate the deconvolutioned image. With target image The first-order norm of the differences between feature maps at different depths in the pre-trained feature extraction model VGG-19. Loss function. The specific calculation formula is as follows:

[0049] in: Loss items to ensure authenticity; For Heisen's loss item, The weights of the Hessian loss term; For the perceived loss term, Weights for the perceived loss term; , To deconvolution image The coordinates of the middle pixel; VGG is a pre-trained feature extraction model. These represent different depths in the VGG model.

[0050] Using the above loss function The adaptive deconvolution model is designed to automatically mine deep feature space representations from the training data through continuous training, adaptively adjust the PSF, and perform deconvolution operations. The main training hyperparameters used in the above training process depend on the specific training dataset (dataset signal-to-noise ratio, dataset size). A set of parameters for reference is shown in Table 2.

[0051] Table 2. Hyperparameters for Model Training

[0052] Step 3, as follows Figure 5 As shown, the deconvolution image of the target region of the target sample is obtained, and the deconvolution image is input into the deconvolution network of the main branch of the adaptive deconvolution model trained in step 2. The deconvolution network outputs the corresponding deconvolution image.

[0053] Example 2

[0054] This embodiment discloses an electronic device, including a processor and a memory. The program instructions stored in the memory include at least a data acquisition module, a model generation and training module, and a deconvolution module. The data acquisition module acquires images for constructing a training set and images to be deconvolved. The model generation and training module generates an adaptive deconvolution model and trains it based on the training set. The deconvolution module processes the images to be deconvolved using the trained adaptive deconvolution model to obtain corresponding deconvolutioned images. The program instructions in the memory are read and executed by the processor to perform the adaptive deconvolution method for microscopic images described in Embodiment 1.

[0055] This embodiment also discloses a storage medium storing program instructions, which at least include a data acquisition module, a model generation and training module, and a deconvolution module. When the storage medium is read and the program instructions are executed by any processor capable of reading and executing the program instructions from the storage medium, the adaptive deconvolution method for microscopic images described in Embodiment 1 is executed.

[0056] Example 3

[0057] like Figure 1 As shown, this embodiment discloses an adaptive deconvolution microscopy imaging system, including an optical imaging system 100 and a control and data processing system 200.

[0058] The optical imaging system 100 is used to perform optical imaging on a target sample and acquire images of various regions on the target sample. Specifically, the optical imaging system 100 includes an excitation optical path and a probe optical path, wherein the excitation optical path includes an excitation objective lens and other optical components for generating excitation light; the probe optical path includes a probe objective lens and other optical components for imaging.

[0059] Those skilled in the art will understand that the excitation objective and the probe objective can be the same objective or different objectives. The optical imaging system 100 may include, but is not limited to, a wide-field microscopy imaging system, a point scanning confocal imaging system, and a rotating disk confocal imaging system.

[0060] The control and data processing system 200 is used to control the optical imaging system 100, acquire images obtained by the optical imaging system 100, and implement the adaptive deconvolution method for microscopic images described in Embodiment 1. Specifically, the control and data processing system 200 includes a computer and related components (such as a processor and data storage), which can control the operation of the optical imaging system 100 and acquire image data from the optical imaging system 100 and perform corresponding post-processing. Therefore, the control and data processing system 200 typically includes an optical microscope control module 210 and a microscopic image neural network processing module 220. The microscopic image neural network processing module 220 mainly consists of an adaptive deconvolution module.

[0061] It should be noted that, within the scope of this application, the modules described herein can be understood to include data storage, such as computer-readable storage media, in which programs and deep neural network models can be stored and executed by a computer, particularly the computer of the control and data processing system 200. These programs and neural network models, when executed by a computer, can implement the methods / steps described below. Specific programming methods for the programs are not discussed in this application; those skilled in the art can implement the relevant functions using any well-known programming software and / or commercial software. Therefore, the following description of the operation of related systems or modules or methods should be understood as meaning that they can also be programmed for computer invocation and execution.

[0062] like Figure 3 and Figure 4 As shown, in step 1 of the adaptive deconvolution method for microscopic images described in Example 1, the control module 210 in the control and data processing system 200 first controls the optical imaging system 100 to scan the sample with a laser, while collecting the fluorescence emitted by the sample and storing the fluorescence signal in the image sequence through photoelectric conversion to form image pairs for constructing the training set.

[0063] For the supervised modality training set, the optical imaging system 100 illuminates each sampling region of the sample with both high and low laser power, and acquires images of each sampling region under high and low laser power illumination, respectively. This results in image pairs for each sampling region constituting a sample. The image under high laser power illumination in each sample has a higher signal-to-noise ratio and is used as the target image. The image under low laser power illumination in each sample has a low signal-to-noise ratio and is used as the original image. This process yields various samples and forms a supervised modality training set.

[0064] For the self-supervised modal training set, the optical imaging system 100 illuminates each sampling region of the sample using only low laser power and acquires an image of each sampling region under low laser power. Then, for each sampled region of the image... Data augmentation is performed to obtain two images with different signal-to-noise ratios for each sampling region, and one of these images is randomly selected as the target image. Another image as the original image This process yields individual samples and forms a self-supervised modality training set.

[0065] like Figure 3 and Figure 4As shown, in step 2 of the adaptive deconvolution method for microscopic images described in Embodiment 1, the microscopic image neural network processing module 220 in the control and data processing system 200 generates the main branch and secondary branch of the adaptive deconvolution model, and calls the supervised modality training set or the supervised modality training set to train the adaptive deconvolution model to obtain the trained adaptive deconvolution model.

[0066] like Figure 5 As shown, in step 3 of the adaptive deconvolution method for microscopic images described in Example 1, the optical microscope control module 210 in the control and data processing system 200 first controls the optical imaging system 100 to acquire the microscopic image to be deconvolved of the target region of the target sample. Next, the microscopic image to be deconvolved... The input is to the microscopic image neural network processing module 220, which calls the deconvolution network of the main branch of the trained adaptive deconvolution module to receive the microscopic image to be deconvolved. The image is then processed, and finally the main branch of the trained adaptive deconvolution module outputs the final deconvolutioned image.

[0067] The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings. These embodiments are merely descriptions of preferred embodiments and are not intended to limit the scope or concept of the invention. The specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. Such combinations, as long as they do not violate the spirit of the present invention, should also be considered as part of this disclosure. To avoid unnecessary repetition, the present invention will not further describe the various possible combinations.

[0068] This invention is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this invention and without departing from the design idea of ​​this invention, all modifications and improvements made by those skilled in the art to the technical solutions of this invention should fall within the protection scope of this invention. The technical content for which protection is sought in this invention has been fully described in the claims.

Claims

1. An adaptive deconvolution method for microscopic images, characterized in that, The process is as follows: Images of multiple different sampling regions of the target sample are acquired. Multiple image pairs are constructed based on the acquired images. Each image pair includes two images of the same sampling region. Different image pairs have different sampling regions. Each image pair is used as a sample, and the training set is composed of these samples. An adaptive deconvolution model is trained using a training set. The adaptive deconvolution model includes a deconvolution network and an optical point spread function network. The training process is as follows: The original image of each sample in the training set is input into the deconvolution network. The deconvolution network obtains the deconvolution image of the original image of each sample. At the same time, an optical point spread function network generates a prior blur kernel. The deconvolution image of the original image of each sample and the corresponding prior blur kernel are convolved to obtain the optically degraded image. Based on the target image, the corresponding deconvolution image, and the optically degraded image of each sample, the loss error value is calculated using a loss function, and the gradient of the loss error value is obtained. The gradient is backpropagated to update the parameters of the deconvolution network and the point spread function network. Thus, through repeated iterative training with multiple samples in the training set, a trained deconvolution network and a point spread function network are obtained, forming a trained adaptive deconvolution model. The unconvolution image of the target region of the target sample is obtained, and the unconvolution image is input into the deconvolution network in the trained adaptive deconvolution model. The deconvolution network outputs the corresponding unconvolution image.

2. The adaptive deconvolution method for microscopic images according to claim 1, characterized in that, Images of each sampling region of the target sample under high-power illumination and low-power illumination are acquired. The image under high-power illumination is the image with a high signal-to-noise ratio for each sampling region, and the image under low-power illumination is the image with a low signal-to-noise ratio for each sampling region. The image with a high signal-to-noise ratio for each sampling region is used as the target image, and the image with a low signal-to-noise ratio is used as the original image. The target image and the original image of each sampling region are used to form an image pair. Each image pair is used as a sample, and the samples are used to form a supervised modality training set. Then, the adaptive deconvolution model is trained in a supervised manner using a supervised modal training set.

3. The adaptive deconvolution method for microscopic images according to claim 1, for the training set in self-supervised training mode, is characterized in that, Images of multiple different sampling regions of the target sample under low-power illumination are acquired. Data amplification is performed on the image of each sampling region to obtain two images with different signal-to-noise ratios for each sampling region. One image is randomly selected from the two images of each sampling region as the original image and the other image as the target image. The target image and the original image in each sampling region form an image pair. Each image pair is used as a sample, and each sample forms a self-supervised modality training set. Then, the adaptive deconvolution model is trained using a self-supervised modal training set.

4. The adaptive deconvolution method for microscopic images according to claim 1, characterized in that, The deconvolution network in the adaptive deconvolution model can be any convolutional neural network or any Vision Transformer.

5. The adaptive deconvolution method for microscopic images according to claim 1, characterized in that, The optical point spread function network in the adaptive deconvolution model can be any convolutional neural network or any VisionTransformer.

6. The adaptive deconvolution method for microscopic images according to claim 1, characterized in that, The loss function during training of the adaptive deconvolution model includes a fidelity loss term, a Hessian loss term, and a perceptual loss term.

7. The adaptive deconvolution method for microscopic images according to claim 1, characterized in that, Training ends when the loss error value calculated by the loss function converges, resulting in a well-trained adaptive deconvolution model.

8. An electronic device comprising a processor and a memory, characterized in that, The program instructions in the memory are read and executed by the processor to perform the adaptive deconvolution method for microscopic images as described in any one of claims 1-7.

9. A storage medium storing program instructions, characterized in that, When the program instructions are read and executed, the adaptive deconvolution method for microscopic images as described in any one of claims 1-7 is performed.

10. An adaptive deconvolution microscopic imaging system, characterized in that, include: An optical imaging system (100) performs optical imaging on a target sample and acquires images of various regions on the target sample; A control and data processing system (200) controls the optical imaging system (100), acquires the image obtained by the optical imaging system (100), and implements the adaptive deconvolution method for microscopic images as described in any one of claims 1-7.

Images