A classification method and device for uterine fibroid MRI images

Abstract

The present invention relates to the technical field of medical image processing, and discloses a classification method and device for uterine fibroid MRI images, wherein the overall structure of the MRI image is divided into a contraction path and an expansion path, a jump connection is adopted in the contraction path, and the residual is used to perform an identity transformation, and the feature information extracted from the upper layer is jumped to the lower network, and the residual mapping method of Resnet is integrated with the downsampling process of U‑Net as the backbone network of the image processing contraction path; the propagation of shallow feature information across channels is directly integrated with the high-level network, and a context enhancement module and a void space convolution pooling pyramid are added between the contraction path and the expansion path; the enhanced image features are subjected to void convolution with sampling rates of different sizes; and a decoderblock is used in the expansion path of the void convolution to restore the features in the image contraction process. The present invention can effectively and comprehensively utilize the resource information of MRI scanning, and improve the imaging accuracy of MRI.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a classification method and device for uterine fibroid MRI images. Background Art

[0002] MRI is a new high-tech imaging method in recent years. It is a new medical imaging diagnosis technology that was only applied in clinical practice in the early 1980s. It has the unique advantages of no ionizing radiation (radiation) damage; no bone artifacts; multi-directional (transverse, coronal, sagittal, etc.) and multi-parameter imaging; high soft tissue resolution; and the ability to display vascular structures without the use of contrast agents. It has good resolution for soft tissues. It is better than CT in the examination of bladder, rectum, uterus, vagina, bones, joints, muscles, etc.; various parameters can be used for imaging, and multiple imaging parameters can provide rich diagnostic information, which makes medical diagnosis and the study of metabolism and function in the human body convenient and effective. For example, the T1 value of hepatitis and cirrhosis becomes larger, while the T1 value of liver cancer is even larger. T1-weighted images can distinguish benign and malignant tumors in the liver; the required section can be freely selected by adjusting the magnetic field. Images of parts that are inaccessible or difficult to access with other imaging technologies can be obtained. For the intervertebral disc and spinal cord, sagittal, coronal and cross-sectional imaging can be performed, and nerve roots, spinal cord and ganglia can be seen. Unlike CT, which can only obtain cross-sections perpendicular to the long axis of the human body; there is no ionizing radiation damage to the human body; in principle, all nuclear elements with non-zero spin can be used for imaging. Uterine fibroids are the most common benign tumors in female reproductive organs and one of the most common tumors in the human body;

[0003] The MRI characteristics of uterine fibroids are mainly manifested as equal or low signals on T1W1, low or mixed signals on T2W1, and high or equal signals around the tumor on T2W1.

[0004] In the current MRI imaging process of uterine fibroids, there are often phenomena such as image redundancy, superposition, blurring, or blurred highlights, so a classification method and device for uterine fibroid MRI images is needed. Summary of the invention

[0005] The purpose of the present invention is to provide a classification method and device for uterine fibroid MRI images. The present invention can effectively distinguish and classify MRI imaging of uterine fibroids by dividing the overall structure of the MRI image into a contraction path and an expansion path, and using the residual to perform an identity transformation. The image features are calculated using multiple dilated convolutions, and the blurred highlights are reconstructed by convolution. The resource information of MRI scans can be effectively and comprehensively utilized to improve the imaging accuracy of MRI.

[0006] The present invention is specifically performed according to the following steps:

[0007] S 1 : The overall structure of MRI images is divided into a contraction path and an expansion path. In the contraction path, jump connections are used, and the residual is used for identity transformation. The feature information extracted from the upper layer is transferred to the lower layer network by jumping. The residual mapping method of Resnet is integrated with the downsampling process of U-Net as the backbone network of the contraction path of image processing;

[0008] S 2 : The propagation of shallow feature information across channels is directly fused with the high-level network, and a context enhancement module and a dilated spatial convolutional pooling pyramid are added between the contraction path and the expansion path; the enhancement module is composed of 3×3 convolution, 3×3 dilated convolution and 5×5 depth-separable convolution. The context information is enhanced by using the nonlinearity of the Relu enhancement model, making a jump connection between the output feature information and the input of the CEblock, and using the fused feature information as the output of the CEblock to enhance the context information at multiple scales.

[0009] S 3 : Perform dilated convolution on the enhanced image features with different sampling rates;

[0010] S 4 : Decoderblock is used in the expansion path of the dilated convolution to restore the features of the image during shrinkage.

[0011] Furthermore, multiple dilated convolutions are used to operate on image features to capture image context at multiple scales. Specifically, four 3×3 dilated convolutions with sampling rates of 6, 12, 18, and 24 are used to construct new feature maps.

[0012] Furthermore, the present invention provides a method and device for classifying uterine fibroid MRI images, on which a computer program is stored, characterized in that when the program is executed by a main controller, the method as described in any one of the above is implemented.

[0013] Compared with the prior art, the present invention has the following beneficial effects:

[0014] The present invention divides the overall structure of the MRI image into a contraction path and an expansion path, adopts jump connections in the contraction path, and uses residuals for identity transformation, which can effectively distinguish and classify MRI imaging of uterine fibroids, uses multiple dilated convolutions to calculate image features, and performs convolution re-images on blurred highlights, which can effectively and comprehensively utilize the resource information of MRI scans and improve the imaging accuracy of MRI. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 It is a schematic flow chart of the method of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0018] See also Figure 1 , a classification method and device for uterine fibroid MRI images,

[0019] S 1 : The overall structure of MRI images is divided into a contraction path and an expansion path. In the contraction path, jump connections are used, and the residual is used for identity transformation. The feature information extracted from the upper layer is transferred to the lower layer network by jumping. The residual mapping method of Resnet is integrated with the downsampling process of U-Net as the backbone network of the contraction path of image processing;

[0020] S 2 : The propagation of shallow feature information across channels is directly fused with the high-level network, and a context enhancement module and a dilated spatial convolutional pooling pyramid are added between the contraction path and the expansion path; the enhancement module is composed of 3×3 convolution, 3×3 dilated convolution and 5×5 depth-separable convolution. The context information is enhanced by using the nonlinearity of the Relu enhancement model, making a jump connection between the output feature information and the input of the CEblock, and using the fused feature information as the output of the CEblock to enhance the context information at multiple scales.

[0021] S 3: Perform dilated convolution on the enhanced image features with different sampling rates;

[0022] S 4 : Decoderblock is used in the expansion path of the dilated convolution to restore the features of the image during shrinkage.

[0023] In this embodiment, multiple dilated convolutions are used to operate on image features to capture image context at multiple scales. Specifically, four 3×3 dilated convolutions with sampling rates of 6, 12, 18, and 24 are used to construct a new feature map.

[0024] In this embodiment, the present invention provides a method and device for classifying uterine fibroid MRI images, on which a computer program is stored, characterized in that when the program is executed by a main controller, a method as described in any one of the above is implemented.

[0025] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A classification method for uterine fibroid MRI images, characterized by: Please follow the steps below to execute: S1: The overall structure of the MRI image is divided into a contraction path and an expansion path. In the contraction path, jump connections are used, and the residual is used for identity transformation. The feature information extracted from the upper layer is transferred to the lower layer network by jumps. The residual mapping method of Resnet is integrated with the downsampling process of U-Net as the backbone network of the contraction path of image processing. S2: The propagation of shallow feature information across channels is directly integrated with the high-level network, and a context enhancement module and a dilated spatial convolutional pooling pyramid are added between the contraction path and the expansion path; S3: Perform dilated convolution on the enhanced image features with different sampling rates; S4: Decoderblock is used in the expansion path of the dilated convolution to restore the features of the image during the contraction process; In step S2, the enhancement module is composed of 3×3 convolution, 3×3 dilated convolution and 5×5 depthwise separable convolution; In step S2, the context information is enhanced by using the nonlinearity of the Relu enhancement model, making a jump connection between the output feature information and the input of the CEblock, and using the fused feature information as the output of the CEblock to enhance the context information at multiple scales. Multiple dilated convolutions are used to operate on image features to capture image context at multiple scales. Specifically, four 3×3 dilated convolutions with sampling rates of 6, 12, 18, and 24 are used to construct new feature maps.

2. A classification device for uterine fibroid MRI images, on which a computer program is stored, characterized in that: When the program is executed by the main controller, the method as claimed in claim 1 is implemented.

Images