Medical image anaphora segmentation method and device based on text-image diffusion model
By using a text-image diffusion model-based approach, and combining visual and language encoders with a U-Net structure to generate binary segmentation masks for medical images, the problems of target boundary differentiation and text-image alignment in medical image segmentation are solved, achieving higher-precision segmentation results and supporting medical diagnosis and treatment.
Patent Information
- Application Number
- CN202510698599.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-28
- Publication Date
- 2025-10-17
AI Technical Summary
Existing medical image segmentation methods have difficulty accurately distinguishing the boundaries of different targets and achieving efficient text-image alignment, resulting in the segmentation results being unable to accurately locate the target area.
A text-image diffusion model is adopted, which uses a visual encoder and a language encoder to convert medical images and text descriptions into latent representations. A mask generation network with a U-Net structure captures the correspondence between the two to generate a multi-channel latent map of the target region, and then converts it into a single-channel binary segmentation mask through the output network.
It improves the accuracy and reliability of medical image segmentation, accurately segments target regions in medical images based on text descriptions, provides more accurate image analysis results, and supports medical diagnosis and treatment.
Smart Images
Figure CN120807902A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of medical image processing, in particular to a medical image reference segmentation method and device based on a text-image diffusion model. BACKGROUND
[0002] Medical image segmentation is an important part of medical image processing, and its purpose is to accurately separate the region of interest in the medical image from the background. In actual medical application scenarios, doctors need to analyze medical images (such as CT, MRI images, etc.) to determine the condition, and accurate image segmentation results can help doctors more accurately observe the lesion site, measure the size of the organ, etc., which is of great significance for early diagnosis and treatment decision of diseases.
[0003] At present, traditional medical image segmentation methods mainly include threshold-based segmentation, region-based segmentation, edge-based segmentation, and machine learning-based segmentation, etc. The threshold-based segmentation method is simple and direct, but for medical images with complex gray scale distribution, the segmentation effect is often not ideal; the region-based segmentation method is sensitive to noise, and over-segmentation or under-segmentation may occur; the edge-based segmentation method has difficulty in accurately extracting the target edge when dealing with medical images with fuzzy edges; the machine learning-based method improves the segmentation accuracy to some extent, but it needs a large amount of labeled data for training, and the labeling process not only consumes time and effort, but also requires professional medical knowledge, and the generalization ability of the model still needs to be improved when facing different data sets.
[0004] In recent years, with the development of deep learning technology, medical image segmentation methods based on convolutional neural networks (CNN) and Transformers have made significant progress. However, these methods still face many challenges when dealing with complex medical images and multi-modal information fusion. For example, when dealing with medical images containing multiple tissues and lesions, it is difficult to accurately distinguish the boundaries of different targets; when combining text descriptions (such as diagnostic information in the case) for image segmentation, existing methods have difficulty in achieving efficient text-image alignment, resulting in inaccurate positioning of the target region in the segmentation results.
[0005] Therefore, it is of great practical significance to develop a method that can effectively fuse text and image information, improve the accuracy and reliability of medical image segmentation. SUMMARY
[0006] The present application provides a medical image reference segmentation method and device based on a text-image diffusion model to solve the problems that existing medical image segmentation methods are difficult to accurately distinguish the boundaries of different targets, difficult to achieve efficient text-image alignment, and further lead to inaccurate positioning of the target region in the segmentation results.
[0007] The first aspect embodiment of the present application provides a medical image reference segmentation method based on a text-image diffusion model, comprising the following steps: obtaining initial medical image data and corresponding text description to construct a training data set; training a pre-constructed text-image diffusion model based on the training data set to obtain a trained text-image diffusion model based on the training data set; inputting the medical image to be segmented and the corresponding text description into the trained text-image diffusion model to generate a binary segmentation mask of the target region.
[0008] Optionally, the trained text-image diffusion model based on the training data set comprises a visual encoder, a language encoder, a mask generation network based on a U-Net structure, and an output network, wherein,
[0009] The visual encoder converts the initial medical image data in the training data set into a latent representation form; the language encoder converts the text description in the training data set into a text representation; the mask generation network based on the U-Net structure captures the corresponding relationship between the latent representation form and the text representation to generate a multi-channel latent map of the target region; and the output network converts the multi-channel latent map of the target region into a single-channel binary segmentation mask of the target region.
[0010] Optionally, the mask generation network based on the U-Net structure comprises a plurality of U-Net layers, wherein each U-Net layer comprises a self-attention module and a cross-attention module, the self-attention module extracts semantic information in the latent representation form, and the cross-attention module generates a multi-channel latent map of the target region under the guidance of the semantic information and the text representation.
[0011] Optionally, the output network comprises a plurality of convolutional layers, which convert the multi-channel latent map of the target region into a single-channel latent map of the target region, and adjust the resolution of the single-channel latent map of the target region to be consistent with the initial medical image data by using an interpolation operation to generate a single-channel binary segmentation mask of the target region.
[0012] The second aspect embodiment of the present application provides a medical image reference segmentation device based on a text-image diffusion model, comprising: an acquisition module for acquiring initial medical image data and corresponding text description to construct a training data set; a training module for training a pre-constructed text-image diffusion model based on the training data set to obtain a trained text-image diffusion model based on the training data set; and a reference segmentation module for inputting the medical image to be segmented and the corresponding text description into the trained text-image diffusion model to generate a binary segmentation mask of the target region.
[0013] Optionally, the trained text-image diffusion model comprises a visual encoder, a language encoder, a mask generation network based on a U-Net structure, and an output network, wherein,
[0014] The visual encoder converts the initial medical image data in the training data set into a latent representation form; the language encoder converts the text description in the training data set into a text representation; the mask generation network based on the U-Net structure captures the correspondence between the latent representation form and the text representation to generate a multi-channel latent map of the target region; and the output network converts the multi-channel latent map of the target region into a single-channel binary segmentation mask of the target region.
[0015] Optionally, the mask generation network based on the U-Net structure comprises a plurality of U-Net layers, wherein each U-Net layer comprises a self-attention module and a cross-attention module, the self-attention module extracts semantic information in the latent representation form, and the cross-attention module generates a multi-channel latent map of the target region under the guidance of the semantic information and the text representation.
[0016] Optionally, the output network comprises a plurality of convolutional layers, which convert the multi-channel latent map of the target region into a single-channel latent map of the target region, and adjust the resolution of the single-channel latent map of the target region to be consistent with the initial medical image data by using an interpolation operation, to generate a single-channel binary segmentation mask of the target region.
[0017] The third aspect of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the medical image reference segmentation method based on the text-image diffusion model as described in the above embodiments.
[0018] The fourth aspect of the present application provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement the medical image reference segmentation method based on the text-image diffusion model as described above.
[0019] The medical image reference segmentation method and device based on the text-image diffusion model proposed in the embodiment of the present invention utilize the powerful multimodal understanding and generation capabilities of the text-image diffusion model to accurately segment the target area in the medical image according to the text description, thereby improving the accuracy and reliability of medical image segmentation, solving the problems existing in traditional medical image segmentation methods in processing complex medical images and text-image alignment, and providing more accurate image analysis results for medical diagnosis and treatment; it can be applied to the precise segmentation of specific target areas (such as diseased tissues, specific organs, etc.) in medical images, providing key technical support for medical diagnosis, disease analysis and treatment plan formulation.
[0020] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:
[0022] Figure 1 A flowchart of a medical image reference segmentation method based on a text-image diffusion model provided by an embodiment of the present invention;
[0023] Figure 2 A schematic block diagram of a medical image reference segmentation device based on a text-image diffusion model provided by an embodiment of the present invention;
[0024] Figure 3 The present invention provides a schematic structural diagram of an electronic device. DETAILED DESCRIPTION
[0025] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.
[0026] The following describes a method and apparatus for medical image reference segmentation based on a text-image diffusion model according to an embodiment of the present invention with reference to the accompanying drawings.
[0027] Figure 1 A flowchart of a medical image reference segmentation method based on a text-image diffusion model provided by an embodiment of the present invention.
[0028] like Figure 1 As shown, the medical image reference segmentation method based on the text-image diffusion model includes the following steps:
[0029] In step S101, initial medical image data and corresponding text description are obtained to construct a training data set.
[0030] In actual implementation, a large amount of medical image data and corresponding text description, such as case diagnosis information and image report, are collected, the pixel value of the medical image is adjusted to a unified range to eliminate the gray difference between different images, and the image is cropped according to the input requirement of the model; meanwhile, the text description is encoded to convert it into a form that can be processed by the model, and the preprocessed medical image data and the encoded text description are used to form a training data set.
[0031] In step S102, the pre-constructed text-image diffusion model is trained using the training data set to obtain a trained text-image diffusion model.
[0032] In step S103, the medical image to be segmented and the corresponding text description are input into the trained text-image diffusion model to generate a binary segmentation mask of the target region.
[0033] In some embodiments, the trained text-image diffusion model includes a visual encoder, a language encoder, a mask generation network based on a U-Net structure, and an output network, wherein,
[0034] The visual encoder converts the initial medical image data in the training data set into a latent representation form;
[0035] The language encoder converts the text description in the training data set into a text representation;
[0036] The mask generation network based on the U-Net structure captures the corresponding relationship between the latent representation form and the text representation to generate a multi-channel latent map of the target region;
[0037] The output network converts the multi-channel latent map of the target region into a single-channel binary segmentation mask of the target region.
[0038] In actual implementation, the medical image to be segmented and the corresponding text description are input into the trained text-image diffusion model, the medical image data is converted into a latent representation form that can be understood by the model using the visual encoder, and the text description is converted into a corresponding text representation using the language encoder, which facilitates subsequent model processing.
[0039] Further, a mask generation network (MGU) based on a U-Net structure including multiple U-Net layers is constructed, the MGU is initialized by using a pre-trained diffusion model, and then the MGU is adjusted and optimized by using the aforementioned processed latent representation and text representation, so that the MGU is converted from generating text-image to generating a mask under the conditions of an image and text. Specifically, each U-Net layer includes a self-attention module and a cross-attention module, the self-attention module extracts detailed semantic information in the latent representation, the cross-attention module focuses on generating a mask covering a target region under the guidance of the semantic information and the text representation, excludes regions irrelevant to the target, captures the correspondence between the latent representation and the text representation, and finally adjusts the resolution of a feature map through a convolution layer to obtain a multi-channel latent image of the target region.
[0040] Further, an output network including multiple convolution layers is constructed, the multi-channel latent image of the target region is converted into a single-channel latent image of the target region through the multiple convolution layers, and the resolution of the single-channel latent image of the target region is adjusted to be consistent with that of the initial medical image data by using an interpolation operation, so as to generate a single-channel binary segmentation mask of the target region.
[0041] The text-image diffusion model is trained by using an AdamW optimizer, a proper weight decay coefficient and an initial learning rate are set, a learning rate is adjusted by using a cosine annealing scheduler, an error between a predicted mask and a real mask is taken as a loss function, trainable parameters in the text-image diffusion model are iteratively optimized until the loss function converges, and a trained text-image diffusion model is obtained.
[0042] The medical image to be segmented and the corresponding text description are input into the trained text-image diffusion model, and a binary segmentation mask of a target region is output.
[0043] In summary, according to the medical image referential segmentation method based on the text-image diffusion model proposed in the embodiment of the present application, the powerful multi-modal understanding and generation capability of the text-image diffusion model is utilized to accurately segment a target region in a medical image according to a text description, the accuracy and reliability of medical image segmentation are improved, the problems of the traditional medical image segmentation method in processing complex medical images and text-image alignment are solved, and more accurate image analysis results are provided for medical diagnosis and treatment; the method can be applied to accurately segmenting a specific target region (such as a lesion tissue or a specific organ) in a medical image, and provides key technical support for medical diagnosis, disease analysis and treatment scheme making.
[0044] Secondly, a medical image referential segmentation device based on a text-image diffusion model proposed in an embodiment of the present application is described with reference to the accompanying drawings.
[0045] Figure 2 A block schematic diagram of a medical image reference segmentation device based on a text-image diffusion model is provided in embodiments of the present application.
[0046] As shown in Figure 2 The medical image reference segmentation device based on a text-image diffusion model 20 comprises an acquisition module 201, a training module 202 and a reference segmentation module 203.
[0047] The acquisition module 201 is configured to acquire initial medical image data and text descriptions corresponding thereto to construct a training data set. The training module 202 is configured to train a pre-constructed text-image diffusion model based on the training data set to obtain a trained text-image diffusion model based on the text-image diffusion model. The reference segmentation module 203 is configured to input a medical image to be segmented and a text description corresponding thereto into the trained text-image diffusion model based on the text-image diffusion model to generate a binary segmentation mask of a target region.
[0048] In some embodiments, the trained text-image diffusion model based on the text-image diffusion model comprises a visual encoder, a language encoder, a mask generation network based on a U-Net structure and an output network, wherein,
[0049] The visual encoder converts the initial medical image data in the training data set into a latent representation form; the language encoder converts the text description in the training data set into a text representation; the mask generation network based on the U-Net structure captures the correspondence between the latent representation form and the text representation to generate a multi-channel latent map of the target region; and the output network converts the multi-channel latent map of the target region into a single-channel binary segmentation mask of the target region.
[0050] In some embodiments, the mask generation network based on the U-Net structure comprises a plurality of U-Net layers, wherein each U-Net layer comprises a self-attention module and a cross-attention module, the self-attention module extracts semantic information in the latent representation form, and the cross-attention module generates a multi-channel latent map of the target region under the guidance of the semantic information and the text representation.
[0051] In some embodiments, the output network comprises a plurality of convolutional layers, the multi-channel latent map of the target region is converted into a single-channel latent map of the target region through the plurality of convolutional layers, and the resolution of the single-channel latent map of the target region is adjusted to be consistent with that of the initial medical image data by using an interpolation operation to generate a single-channel binary segmentation mask of the target region.
[0052] It should be noted that the foregoing explanation of the medical image reference segmentation method based on the text-image diffusion model also applies to the medical image reference segmentation device based on the text-image diffusion model, which will not be described here again.
[0053] The medical image reference segmentation device based on the text-image diffusion model according to the embodiment of the present application utilizes the powerful multi-modal understanding and generation capability of the text-image diffusion model, accurately segments the target region in the medical image according to the text description, improves the precision and reliability of the medical image segmentation, solves the problems of the traditional medical image segmentation method in processing complex medical images and text-image alignment, and provides more accurate image analysis results for medical diagnosis and treatment. The device can be applied to accurate segmentation of specific target regions (such as diseased tissues, specific organs, etc.) in medical images, and provides key technical support for medical diagnosis, disease analysis and treatment plan making.
[0054] Figure 3 The electronic device provided by the embodiment of the present application is shown in the structural schematic diagram. The electronic device can include:
[0055] The memory 301, the processor 302 and the computer program stored in the memory 301 and executable on the processor 302.
[0056] The processor 302 implements the medical image reference segmentation method based on the text-image diffusion model provided in the above embodiment when executing the program.
[0057] Further, the electronic device further includes:
[0058] The communication interface 303 is used for communication between the memory 301 and the processor 302.
[0059] The memory 301 is used to store the computer program executable on the processor 302.
[0060] The memory 301 can include a high-speed RAM memory, and can also include a non-volatile memory, for example, at least one disk memory.
[0061] If the memory 301, the processor 302 and the communication interface 303 are independently implemented, the communication interface 303, the memory 301 and the processor 302 can be connected to each other through a bus and complete the communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, Figure 3Only one bus or only one type of bus can exist, however.
[0062] Optionally, in a specific implementation, if the memory 301, the processor 302 and the communication interface 303 are integrated on a chip, the memory 301, the processor 302 and the communication interface 303 can complete the communication with each other through an internal interface.
[0063] The processor 302 can be a central processing unit (CPU) or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present application.
[0064] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement the medical image reference segmentation method based on the text-image diffusion model as described above.
[0065] In the description of the present application, the description of the terms “one embodiment”, “some embodiments”, “an example”, “a specific example” or “some examples” means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present application, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the present application and the features of the different embodiments or examples without contradiction.
[0066] In addition, the terms “first”, “second” are only used for description purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with “first”, “second” can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of “N” is at least two, for example, two, three, etc., unless otherwise specifically limited.
[0067] Any processes or methods described in the flowcharts or otherwise described herein can be understood as representing a module, segment, or portion of code that includes one or N executable instructions for implementing the specified logical function(s) or process(es), and the various embodiments of the application can include additional or fewer steps, operations, etc. as appropriate or desired for a given implementation, as will be understood by those skilled in the art.
[0068] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be embodied in non-transitory computer-readable media, which can be executed by an instruction execution system, apparatus, or device, such as a computer-based system, processor, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For purposes of this specification, a "computer-readable medium" can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-readable medium can be a computer- readable storage medium or a computer-readable communication medium. The computer-readable storage medium can be, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable storage medium include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, and a portable compact disc read-only memory (CD-ROM). In this context, a computer-readable storage medium can also be any tangible apparatus that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device.
[0069] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the above embodiments, the N steps or methods can be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented in hardware and in another embodiment, any of the following technologies, or combinations thereof, can be used: discrete logic circuitry having logic gates for implementing logic functions upon data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and so forth.
[0070] Those skilled in the art of the present technology can understand that all or part of the steps carried out by the above-mentioned embodiment method can be completed by programs instructing related hardware, and the programs can be stored in a computer readable storage medium. When the program is executed, it includes one of the steps of the method embodiment or a combination thereof.
[0071] In addition, each functional unit in each embodiment of the present application can be integrated into one processing module, or each unit can exist physically alone, or two or more units can be integrated into one module. The integrated module can be realized in the form of hardware or in the form of a software functional module. When the integrated module is realized in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer readable storage medium.
[0072] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it should be understood that the above-mentioned embodiments are exemplary and cannot be understood as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above-mentioned embodiments within the scope of the present application.
Claims
1. A method for medical image referent segmentation based on a text-image diffusion model, characterized in that: The following steps are involved: Obtain initial medical image data and its corresponding text description to construct a training dataset; Using the training data set to train a pre-built text-image diffusion model to obtain a trained text-image diffusion model; The segmented medical image to be referred to and the corresponding text description are input into the trained text-image diffusion model to generate a binary segmentation mask of the target area.
2. The medical image reference segmentation method based on the text-image diffusion model according to claim 1, characterized in that: The trained text-image diffusion model includes a visual encoder, a language encoder, a mask generation network based on a U-Net structure, and an output network, wherein: The visual encoder converts initial medical image data in the training dataset into a latent representation; The language encoder converts the text descriptions in the training dataset into text representations; The mask generation network based on the U-Net structure captures the correspondence between the potential representation and the text representation to generate a multi-channel potential map of the target area; The output network converts the multi-channel latent map of the target region into a single-channel binary segmentation mask of the target region.
3. The medical image reference segmentation method based on the text-image diffusion model according to claim 2, characterized in that: The mask generation network based on the U-Net structure includes multiple U-Net layers, wherein each U-Net layer includes a self-attention module and a cross-attention module. The self-attention module extracts semantic information in the latent representation, and the cross-attention module is guided by the semantic information and the text representation to generate a multi-channel latent map of the target area.
4. The medical image reference segmentation method based on the text-image diffusion model according to claim 2, characterized in that: The output network includes multiple convolutional layers, which convert the multi-channel potential map of the target area into a single-channel potential map of the target area through the multiple convolutional layers, and use interpolation operations to adjust the resolution of the single-channel potential map of the target area to be consistent with the initial medical image data to generate a single-channel binary segmentation mask of the target area.
5. A medical image reference segmentation device based on a text-image diffusion model, characterized in that: include: An acquisition module is used to acquire initial medical image data and its corresponding text description to construct a training dataset; A training module, configured to train a pre-built text-image diffusion model using the training data set to obtain a trained text-image diffusion model; The reference segmentation module is used to input the medical image to be reference segmented and its corresponding text description into the trained text-image diffusion model to generate a binary segmentation mask of the target area.
6. The medical image reference segmentation device based on text-image diffusion model according to claim 5, characterized in that: The trained text-image diffusion model includes a visual encoder, a language encoder, a mask generation network based on a U-Net structure, and an output network, wherein: The visual encoder converts initial medical image data in the training dataset into a latent representation; The language encoder converts the text descriptions in the training dataset into text representations; The mask generation network based on the U-Net structure captures the correspondence between the potential representation and the text representation to generate a multi-channel potential map of the target area; The output network converts the multi-channel latent map of the target region into a single-channel binary segmentation mask of the target region.
7. The medical image reference segmentation device based on text-image diffusion model according to claim 6, characterized in that: The mask generation network based on the U-Net structure includes multiple U-Net layers, wherein each U-Net layer includes a self-attention module and a cross-attention module. The self-attention module extracts semantic information in the latent representation, and the cross-attention module is guided by the semantic information and the text representation to generate a multi-channel latent map of the target area.
8. The medical image reference segmentation device based on text-image diffusion model according to claim 6, characterized in that: The output network includes multiple convolutional layers, which convert the multi-channel potential map of the target area into a single-channel potential map of the target area through the multiple convolutional layers, and use interpolation operations to adjust the resolution of the single-channel potential map of the target area to be consistent with the initial medical image data to generate a single-channel binary segmentation mask of the target area.
9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the medical image reference segmentation method based on the text-image diffusion model as described in any one of claims 1 to 4.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the medical image reference segmentation method based on the text-image diffusion model as described in any one of claims 1 to 4.