Medical image hysteromyoma segmentation method

Abstract

The invention discloses a medical image hysteromyoma segmentation method, which comprises the following steps of: combining a U-shaped network of cavity convolution and an attention mechanism, inhibiting unimportant features by applying a channel attention mechanism between layers of an encoder, and combining a cavity space convolution pooling pyramid and a convolution attention module at the bottom of the encoder, so as to improve the segmentation accuracy of hysteromyoma. The recognition capability of the model for small-size tumors can be improved, the segmentation accuracy and reliability are further improved, in addition, by constructing a semantic segmentation model, dividing a pelvic medical image into a plurality of areas, constructing an instance segmentation model, segmenting the uterine fibroid, obtaining the volume of the uterine fibroid and extracting the contour of the uterine fibroid, the segmentation accuracy and reliability are improved, and the segmentation accuracy and reliability are improved. And determining an area adjacent to each pixel point on the contour, thereby realizing automatic typing of the hysteromyoma.

Description

Technical Field

[0001] The present invention relates to the technical field of uterine fibroid processing, and more particularly to a uterine fibroid segmentation method of a medical image. Background Art

[0002] Nowadays, high-intensity focused ultrasound technology is used to treat uterine fibroids. HI FU ablation technology uses a certain focusing technology to focus low-energy ultrasound in vitro to a specific area in the body, and uses the biological effect of ultrasound to cause irreversible coagulative necrosis of the tissue, but does not damage the surrounding normal tissue, thereby achieving the purpose of ablating the tumor. More and more deep learning models are being applied to medical image research, among which the most widely used is the technology based on convolutional neural networks. However, in practice, it is found that due to the indirect transmission of feature information, the features after transmission increase inaccuracy, resulting in inaccurate position information and inaccurate segmentation of the lesion edge. There are often problems such as difficulty in identifying small-volume tumors and poor boundary segmentation, resulting in low reliability of segmentation prediction results. In addition, the current classification mainly relies on the subjective judgment of doctors, which requires a lot of time for analysis, and there is no automated classification algorithm and device. The doctor's diagnosis depends on the doctor's seniority and diagnostic experience, which is highly subjective and has poor consistency. To this end, the present invention proposes an automated classification method based on a deep learning segmentation model. The areas of the serosal membrane and mucosa are determined based on semantic segmentation, and then the area of ​​the uterine fibroids is obtained based on instance segmentation on medical images. After extracting relevant features, the classification is automatically obtained. When clinically diagnosing uterine fibroids, it is usually necessary to segment different tissues and organs in the ultrasound image so that doctors can diagnose and treat the disease more accurately. The diagnosis of traditional ultrasound images usually requires doctors to manually segment organs, which takes a lot of time and experience from doctors. In related technologies, deep learning is used to segment ultrasound images, but due to insufficient data, some important features that are not obvious are easily missed, causing doctors to misdiagnose or miss the diagnosis. Summary of the invention

[0003] 1. Technical issues to be resolved

[0004] In view of the problems existing in the prior art, the present invention provides a uterine fibroid segmentation method of a medical image to solve the technical problems mentioned in the background technology.

[0005] (II) Technical solution

[0006] To achieve the above object, the present invention provides the following technical solution: a method for segmenting uterine fibroids in medical images, comprising the following steps:

[0007] Step 1: Collect several medical images for preprocessing, obtain several sample images to form a sample data set, build a semantic segmentation model, input the pelvic medical image into the trained model to divide the image into different areas: intramucosal; muscularis, that is, the area between the mucosa and the serosa; extraserosa; build an instance segmentation model, input the pelvic medical image into the trained model, segment the uterine fibroids, and obtain the volume of the uterine fibroids;

[0008] Step 2: When the noise point of the ultrasound image of the patient with uterine fibroids is similar to the preset important feature point, the noise point is confirmed as the target noise point, the contour of the uterine fibroid is extracted, the area adjacent to each pixel point on the contour is determined, and the volume VO of the extraserous area and the volume VN of the intramucosal area of ​​the uterine fibroid are calculated; the uterine fibroid is subjected to FIGO classification based on VO, VN and VT;

[0009] Step 3: segmenting the ultrasound image retaining the target noise point, determining the segmented image corresponding to each organ in the ultrasound image, and extracting label information corresponding to each of the organs; confirming the target organ among the multiple organs through the label information, analyzing the image features in the segmented image corresponding to the target organ, and constructing a U-shaped network model;

[0010] Step 4: training a uterine fibroid state model based on the image features, wherein the uterine fibroid state model is used to assist in analyzing the uterine fibroid state;

[0011] Step 5: Use the sample data set to train the U-shaped network model to obtain a uterine fibroid image segmentation model with optimal model parameters, input the medical image to be tested into the uterine fibroid image segmentation model, and obtain the tumor area corresponding to the uterine fibroid.

[0012] The present invention is further configured such that, before confirming that the noise point is a target noise point in step four, it also includes: obtaining a first ultrasound image data set for image restoration training; the first ultrasound image data set includes a first original ultrasound image sample output by an ultrasound device, and a first image segmentation sample obtained by segmenting the first original ultrasound image sample; reorganizing the first image segmentation sample according to a segmentation route to obtain a first restored image sample; removing the first noise point in the first original image sample to obtain a first preprocessed image sample; and determining a first similarity loss between the first preprocessed image sample and the first restored image sample according to a preset cross entropy loss function.

[0013] The present invention is further configured such that, when the first similarity loss satisfies a preset similarity threshold, the first noise point is determined to be the important feature point; after determining the first similarity loss between the first preprocessed image sample and the first restored image according to a preset cross entropy loss function, it also includes: when the first similarity loss is lower than a preset similarity threshold, retaining a second noise point in the first original image sample in sequence and obtaining a second preprocessed image sample; determining the second similarity loss between the second preprocessed image sample and the first restored image sample according to a preset cross entropy loss function; and when the second similarity loss satisfies a preset similarity threshold, determining the second noise point to be the important feature point.

[0014] The present invention is further configured such that each layer of the decoder is provided with a dual convolution module; the decoding of the adaptive feature map layer by layer by the decoder includes: fusing the input adaptive feature map with the features of the corresponding layer of the jump-connected encoder through the dual convolution module of each layer of the encoder, and then performing an upsampling operation to restore the image size.

[0015] The present invention is further configured such that the output of the semantic segmentation model is two types: serous membrane and mucous membrane, and the activation function of the last layer is si gmoid; when the model predicts, the area with si gmoid output greater than 0.5 is the predicted type; if a pixel is predicted to be both serous membrane and mucous membrane, then the final type of the pixel is mucous membrane.

[0016] The present invention is further configured such that the segmentation retains the target noise point in the ultrasound image, and determines the segmented image corresponding to each organ in the ultrasound image, specifically including: performing a preprocessing operation on the ultrasound image, and inputting the ultrasound image into a preset image segmentation model to obtain the segmented image corresponding to each organ.

[0017] The present invention is further configured such that the preprocessing operation on the ultrasonic image specifically includes: obtaining unprocessed ultrasonic image samples, and performing resolution normalization and grayscale normalization on the unprocessed ultrasonic image samples to obtain multiple normalized image samples; performing edge detection on the normalized image samples to obtain corresponding edge image samples; and performing image fusion on each of the normalized image samples and the corresponding edge image samples to obtain the ultrasonic image.

[0018] The present invention is further configured such that a method for determining an area adjacent to each pixel point on a contour comprises: selecting a neighborhood centered on any pixel point A on the contour; if all pixel points other than uterine fibroids in the neighborhood belong to one of three areas, then the one area is the area adjacent to pixel point A; if all pixel points other than uterine fibroids in the neighborhood belong to multiple areas among the three areas, then selecting any one of the multiple areas as the area adjacent to pixel point A.

[0019] (III) Beneficial effects

[0020] Compared with the prior art, the present invention provides a uterine fibroid segmentation method for medical images, which has the following beneficial effects:

[0021] The present invention proposes a U-shaped network combining a dilated convolution and an attention mechanism. By applying a channel attention mechanism between each layer of the encoder to suppress unimportant features, and combining a dilated spatial convolution pooling pyramid with a convolution attention module at the bottom of the encoder, the model's recognition ability for small-volume tumors can be improved, thereby improving segmentation accuracy and reliability. In addition, by constructing a semantic segmentation model, the pelvic medical image is divided into multiple regions, an instance segmentation model is constructed, uterine fibroids are segmented, the volume of the uterine fibroids is obtained, the contour of the uterine fibroids is extracted, and the area adjacent to each pixel on the contour is determined, thereby realizing automatic classification of uterine fibroids. The present invention does not require manual participation in the entire process of uterine fibroid classification, greatly reduces the analysis pressure of radiologists, and helps to improve the accuracy and consistency of classification judgment. The present invention is based on the uterine fibroid state monitoring and analysis method and ultrasound system based on ultrasound images. By retaining the target noise point information with important information in the ultrasound image, when the patient's ultrasound image is segmented, the image features carried by the target noise point can also be reflected in the segmented image, and the uterine fibroid state model is used to assist doctors in analyzing the growth state of uterine fibroids, reducing the doctor's diagnosis time. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is an overall flow chart of the uterine fibroid segmentation method of the medical image in the present invention; DETAILED DESCRIPTION

[0023] Embodiment 1:

[0024] like Figure 1 As shown, the present invention discloses a method for segmenting uterine fibroids in a medical image, comprising the following steps:

[0025] Step 1: Collect several medical images for preprocessing, obtain several sample images to form a sample data set, build a semantic segmentation model, input the pelvic medical image into the trained model to divide the image into different areas: intramucosal; muscularis, that is, the area between the mucosa and the serosa; extraserosa; build an instance segmentation model, input the pelvic medical image into the trained model, segment the uterine fibroids, and obtain the volume of the uterine fibroids;

[0026] Step 2: When the noise point of the ultrasound image of the patient with uterine fibroids is similar to the preset important feature point, the noise point is confirmed as the target noise point, the contour of the uterine fibroid is extracted, the area adjacent to each pixel point on the contour is determined, and the volume VO of the extraserous area and the volume VN of the intramucosal area of ​​the uterine fibroid are calculated; the uterine fibroid is subjected to FIGO classification based on VO, VN and VT;

[0027] Step 3: segmenting the ultrasound image retaining the target noise point, determining the segmented image corresponding to each organ in the ultrasound image, and extracting label information corresponding to each of the organs; confirming the target organ among the multiple organs through the label information, analyzing the image features in the segmented image corresponding to the target organ, and constructing a U-shaped network model;

[0028] Step 4: training a uterine fibroid state model based on the image features, wherein the uterine fibroid state model is used to assist in analyzing the uterine fibroid state;

[0029] Step 5: Use the sample data set to train the U-shaped network model to obtain a uterine fibroid image segmentation model with optimal model parameters, input the medical image to be tested into the uterine fibroid image segmentation model, and obtain the tumor area corresponding to the uterine fibroid.

[0030] In a further embodiment of the present invention, before confirming that the noise point is a target noise point in step four, it also includes: obtaining a first ultrasound image data set for image restoration training; the first ultrasound image data set includes a first original ultrasound image sample output by an ultrasound device, and a first image segmentation sample obtained by segmenting the first original ultrasound image sample; reorganizing the first image segmentation sample according to the segmentation route to obtain a first restored image sample; removing the first noise point in the first original image sample to obtain a first preprocessed image sample; determining a first similarity loss between the first preprocessed image sample and the first restored image sample according to a preset cross entropy loss function.

[0031] In a further embodiment of the present invention, when the first similarity loss satisfies a preset similarity threshold, the first noise point is determined to be the important feature point; after determining the first similarity loss between the first preprocessed image sample and the first restored image according to a preset cross entropy loss function, it also includes: when the first similarity loss is lower than the preset similarity threshold, retaining a second noise point in the first original image sample in sequence and obtaining a second preprocessed image sample; determining the second similarity loss between the second preprocessed image sample and the first restored image sample according to a preset cross entropy loss function; when the second similarity loss satisfies the preset similarity threshold, determining the second noise point as the important feature point.

[0032] Embodiment 2:

[0033] The uterine fibroid segmentation method of medical images includes the following steps:

[0034] Step 1: Collect several medical images for preprocessing, obtain several sample images to form a sample data set, build a semantic segmentation model, input the pelvic medical image into the trained model to divide the image into different areas: intramucosal; muscularis, that is, the area between the mucosa and the serosa; extraserosa; build an instance segmentation model, input the pelvic medical image into the trained model, segment the uterine fibroids, and obtain the volume of the uterine fibroids;

[0035] Step 2: When the noise point of the ultrasound image of the patient with uterine fibroids is similar to the preset important feature point, the noise point is confirmed as the target noise point, the contour of the uterine fibroid is extracted, the area adjacent to each pixel point on the contour is determined, and the volume VO of the extraserous area and the volume VN of the intramucosal area of ​​the uterine fibroid are calculated; the uterine fibroid is subjected to FIGO classification based on VO, VN and VT;

[0036] Step 3: segmenting the ultrasound image retaining the target noise point, determining the segmented image corresponding to each organ in the ultrasound image, and extracting label information corresponding to each of the organs; confirming the target organ among the multiple organs through the label information, analyzing the image features in the segmented image corresponding to the target organ, and constructing a U-shaped network model;

[0037] Step 4: training a uterine fibroid state model based on the image features, wherein the uterine fibroid state model is used to assist in analyzing the uterine fibroid state;

[0038] Step 5: Use the sample data set to train the U-shaped network model to obtain a uterine fibroid image segmentation model with optimal model parameters, input the medical image to be tested into the uterine fibroid image segmentation model, and obtain the tumor area corresponding to the uterine fibroid.

[0039] In a further embodiment of the present invention, each layer of the decoder is provided with a dual convolution module; the decoding of the adaptive feature map layer by layer by the decoder comprises: fusing the input adaptive feature map with the features of the corresponding layer of the jump-connected encoder through the dual convolution module of each layer of the encoder, and then performing an upsampling operation to restore the image size; the output of the semantic segmentation model is two types of serosa and mucosa, and the activation function of the last layer is sigmoid; when the model predicts, the area with sigmoid output greater than 0.5 is the predicted type; if a pixel is predicted to be both serosa and mucosa, then the final type of the pixel is mucosa; the segmentation of the ultrasound image retaining the target noise point and determining the segmented image corresponding to each organ in the ultrasound image specifically comprises: performing a preprocessing operation on the ultrasound image and inputting the ultrasound image into a preset image segmentation model to obtain the segmented image corresponding to each organ; the preprocessing operation on the ultrasound image specifically comprises: obtaining an unprocessed ultrasound image sample, and performing resolution normalization and grayscale normalization on the unprocessed ultrasound image sample to obtain a plurality of normalized image samples This method comprises the following steps: performing edge detection on the normalized image samples to obtain corresponding edge image samples; performing image fusion on each of the normalized image samples and the corresponding edge image samples to obtain the ultrasonic image; and determining the area adjacent to each pixel point on the contour comprises: selecting a neighborhood centered on any pixel point A on the contour; if all the pixels other than the uterine fibroids in the neighborhood belong to one of three areas, then the one area is the area adjacent to the pixel point A; if the pixels other than the uterine fibroids in the neighborhood belong to multiple areas among the three areas, then selecting any one of the multiple areas as the area adjacent to the pixel point A.

[0040] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be listed as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for segmenting uterine fibroids in medical images, characterized in that: The following steps are involved: Step 1: Collect several medical images for preprocessing, obtain several sample images to form a sample data set, build a semantic segmentation model, input the pelvic medical image into the trained model to divide the image into different areas: intramucosal; muscularis, that is, the area between the mucosa and the serosa; Extraserosa; constructing an instance segmentation model, inputting the pelvic medical image into the trained model, segmenting the uterine fibroids, and obtaining the volume of the uterine fibroids; Step 2: When the noise point of the ultrasound image of the patient with uterine fibroids is similar to the preset important feature point, the noise point is confirmed as the target noise point, the contour of the uterine fibroid is extracted, the area adjacent to each pixel point on the contour is determined, and the volume VO of the extraserous area and the volume VN of the intramucosal area of ​​the uterine fibroid are calculated; the uterine fibroid is classified by F IGO based on VO, VN and VT; Step 3: segmenting the ultrasound image retaining the target noise point, determining the segmented image corresponding to each organ in the ultrasound image, and extracting label information corresponding to each of the organs; confirming the target organ among the multiple organs through the label information, analyzing the image features in the segmented image corresponding to the target organ, and constructing a U-shaped network model; Step 4: training a uterine fibroid state model based on the image features, wherein the uterine fibroid state model is used to assist in analyzing the uterine fibroid state; Step 5: Use the sample data set to train the U-shaped network model to obtain a uterine fibroid image segmentation model with optimal model parameters, input the medical image to be tested into the uterine fibroid image segmentation model, and obtain the tumor area corresponding to the uterine fibroid.

2. The method for segmenting uterine fibroids of a medical image according to claim 1, wherein: Before confirming that the noise point is a target noise point in step 4, the method further includes: obtaining a first ultrasound image data set for image restoration training; the first ultrasound image data set includes a first original ultrasound image sample output by an ultrasound device, and a first image segmentation sample obtained by segmenting the first original ultrasound image sample; reorganizing the first image segmentation sample according to the segmentation route to obtain a first restored image sample; removing the first noise point in the first original image sample to obtain a first preprocessed image sample; and determining a first similarity loss between the first preprocessed image sample and the first restored image sample according to a preset cross entropy loss function.

3. The method for segmenting uterine fibroids of a medical image according to claim 1, wherein: When the first similarity loss meets a preset similarity threshold, determining the first noise point as the important feature point; After determining the first similarity loss between the first preprocessed image sample and the first restored image according to a preset cross entropy loss function, the method further includes: when the first similarity loss is lower than a preset similarity threshold, retaining a second noise point in the first original image sample in sequence and obtaining a second preprocessed image sample; determining the second similarity loss between the second preprocessed image sample and the first restored image sample according to a preset cross entropy loss function; and when the second similarity loss meets the preset similarity threshold, determining the second noise point as the important feature point.

4. The method for segmenting uterine fibroids of a medical image according to claim 1, wherein: Each layer of the decoder is provided with a dual convolution module; the adaptive feature map is decoded layer by layer by the decoder, including: fusing the input adaptive feature map with the features of the corresponding layer of the jump-connected encoder through the dual convolution module of each layer of the encoder, and then performing an upsampling operation to restore the image size.

5. The method for segmenting uterine fibroids of medical images according to claim 1, characterized in that: The output of the semantic segmentation model is two types: serous membrane and mucous membrane. The activation function of the last layer is sigmoid. When the model predicts, the area with sigmoid output greater than 0.5 is the predicted type. If a pixel is predicted to be both serous membrane and mucous membrane, the final type of the pixel is mucous membrane.

6. The method for segmenting uterine fibroids of a medical image according to claim 5, characterized in that: The segmentation of the ultrasound image retaining the target noise point and determining the segmented image corresponding to each organ in the ultrasound image specifically includes: performing a preprocessing operation on the ultrasound image and inputting the ultrasound image into a preset image segmentation model to obtain the segmented image corresponding to each organ.

7. The method for segmenting uterine fibroids of a medical image according to claim 6, characterized in that: The preprocessing operation on the ultrasonic image specifically includes: obtaining unprocessed ultrasonic image samples, and performing resolution normalization and grayscale normalization on the unprocessed ultrasonic image samples to obtain multiple normalized image samples; performing edge detection on the normalized image samples to obtain corresponding edge image samples; and performing image fusion on each of the normalized image samples and the corresponding edge image samples to obtain the ultrasonic image.

8. The method for segmenting uterine fibroids of medical images according to claim 3, characterized in that: The method for determining the area adjacent to each pixel point on the contour includes: selecting a neighborhood centered on any pixel point A on the contour; if all pixel points other than uterine fibroids in the neighborhood belong to one of three areas, then the one area is the area adjacent to pixel point A; if all pixel points other than uterine fibroids in the neighborhood belong to multiple areas among the three areas, then any one of the multiple areas is selected as the area adjacent to pixel point A.

Images