CN112001895A - Thyroid calcification detection device - Google Patents

Thyroid calcification detection device Download PDF

Info

Publication number
CN112001895A
CN112001895A CN202010766984.XA CN202010766984A CN112001895A CN 112001895 A CN112001895 A CN 112001895A CN 202010766984 A CN202010766984 A CN 202010766984A CN 112001895 A CN112001895 A CN 112001895A
Authority
CN
China
Prior art keywords
calcification
thyroid
image
model
tnet
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202010766984.XA
Other languages
Chinese (zh)
Other versions
CN112001895B (en
Inventor
陈奇富
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shiwei Xinzhi Medical Technology Shanghai Co ltd
Original Assignee
Shiwei Xinzhi Medical Technology Shanghai Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shiwei Xinzhi Medical Technology Shanghai Co ltd filed Critical Shiwei Xinzhi Medical Technology Shanghai Co ltd
Priority to CN202010766984.XA priority Critical patent/CN112001895B/en
Publication of CN112001895A publication Critical patent/CN112001895A/en
Application granted granted Critical
Publication of CN112001895B publication Critical patent/CN112001895B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/40—Analysis of texture
    • G06T7/41—Analysis of texture based on statistical description of texture
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/90—Determination of colour characteristics
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10132—Ultrasound image

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • General Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Apparatus For Radiation Diagnosis (AREA)
  • Image Analysis (AREA)
  • Ultra Sonic Daignosis Equipment (AREA)

Abstract

The invention relates to a thyroid calcification detection device, which comprises: the TNet model construction and training module: constructing a TNet model based on a convolutional neural network VGG-19 model; an image of interest acquisition module: obtaining an interested image by selecting an interested coordinate point; thyroid benign and malignant detection module: detecting the image of interest by the TNet model and generating as output a "follow-up" or "puncture"; a thyroid profile generation module: the method comprises the steps of selecting preset feature maps from feature maps output by a first convolution layer in the TNet model to be combined to obtain a thyroid gland feature map; thyroid calcification detection module: and the method is used for carrying out calcification detection on the thyroid gland characteristic map. The method can effectively detect the calcification condition of the thyroid gland, has better recognition rate, and provides reliable basis for doctors to correctly judge the disease condition.

Description

Thyroid calcification detection device
Technical Field
The invention relates to the field of auxiliary medical diagnosis, in particular to a thyroid calcification detection device.
Background
Thyroid cancer is one of the most common cancers in women worldwide, with women having three times as many incidence as men. In 2018, thyroid cancer was found in 1 of every 20 female cases diagnosed with cancer. Ultrasound imaging is a non-invasive, non-radiative, and low-cost technique for cancer diagnosis. However, due to the low quality of ultrasound images, identifying the thyroid gland and detecting cancer signs by ultrasound is a difficult task.
In recent years, Convolutional Neural Networks (CNNs) have shown excellent target detection capabilities, particularly for large-scale visual recognition tasks. CNNs have been used for different computer vision tasks, including medical imaging, which exhibit a powerful function in feature learning and are capable of learning distinctive and robust object features (e.g., lines, shapes, textures, and colors) from images. There are many CNN models that have been developed for object classification problems, such as VGGNet designed in the context of the "large scale visual recognition challenge" (ILSVRC) of ImageNet datasets. The VGG model is derived from approximately 120 million labeled images trained by DCNN, which contains 1000 different classes from the ILSVRC dataset, where each individual object in the dataset is the subject and is centered in the image with a somewhat cluttered background. The VGG model takes the entire image as input and predicts the class label of the object. The architecture of the VGG model network includes a weight layer, a normalization layer, a max pooling layer, a fully connected layer, and a linear layer with softmax activation in the output layer.
Disclosure of Invention
The invention aims to provide a thyroid calcification detection device which can effectively identify the calcification condition of the thyroid.
The technical scheme adopted by the invention for solving the technical problems is as follows: provided is a thyroid calcification detection apparatus including:
the TNet model construction and training module: constructing and training a VGG-19 model based on a convolutional neural network, and adjusting the final three-layer structure of the VGG-19 model; migrating the layer trained by the VGG-19 model to a new convolutional neural network model to construct a TNet initial model, and pre-training the TNet initial model to obtain the TNet model;
an image of interest acquisition module: intercepting a thyroid gland boundary of the thyroid gland ultrasonic image in a mode of selecting an interested coordinate point to obtain an interested image;
thyroid benign and malignant detection module: for inputting the image of interest into the TNet model, which detects the image of interest and generates as output a "follow-up" or "puncture";
a thyroid profile generation module: the method comprises the steps of selecting preset feature maps from feature maps output by a first convolution layer in the TNet model to be combined to obtain a thyroid gland feature map;
thyroid calcification detection module: used for carrying out calcification detection on the thyroid gland feature map, if the average brightness of the thyroid gland is less than or equal to a preset threshold value tcystThe thyroid gland is cyst; if the average thyroid brightness is larger than the preset threshold tcystThe thyroid is a non-cystic nodule and the non-cystic nodule is detected for calcification by calcified region detection and calcification index estimation.
The detection of the calcified area in the thyroid calcification detection module comprises the following steps: thyroid ultrasound image I corresponding to the non-cystic nodulex,yMinus its corresponding thyroid profile I'x,yAnd obtaining a suspicious calcification image, wherein the formula is as follows:
Figure BDA0002615076570000021
wherein, Delta is Ix,y-I'x,yIf the suspicious calcification image I-(x, y) > Preset threshold tcaleIf so, determining that the non-cystic nodule has potential calcification and the suspicious calcification image is a potential calcification image; if the suspicious calcification image I-(x, y) is less than or equal to a preset threshold value tcaleThen the non-cystic nodules are not calcified.
The estimating of the calcification index in the thyroid calcification detecting module comprises: constructing a calcification index from the potential calcification image, the formula being:
Figure BDA0002615076570000022
wherein x ismaxAnd ymaxRespectively representing the total pixel number of an x axis and the total pixel number of a y axis in the potential calcification image;
Figure BDA0002615076570000023
the arg max function is used to express when the calcification index r iscaleWhen the fastest change is generated, the preset threshold value t iscaleIs the optimal threshold.
The detection of calcified areas in the thyroid calcification detection module further comprises: filtering the Boolean mapping generated by the latent calcification image by using a structuring element with a width of 1 pixel for morphological opening to remove non-calcified regions due to speckle noise, specifically: cutting each connecting area in the potential calcification image independently to obtain a cutting area;
extracting Haralick texture features from the gray level co-occurrence matrix of each cutting region through a pre-training multivariate Gaussian Bayes classifier, dividing each cutting region into a suspicious calcified region and a non-calcified region according to the Haralick texture features, and removing the non-calcified region from the potential calcified image.
The detection of calcified areas in the thyroid calcification detection module further comprises: detecting the median value of the intensity values around each suspicious calcification area, and if the median value of the intensity values around the suspicious calcification areas is not more than a preset threshold tcolloidAnd if so, the suspicious calcification area is colloid, and the colloid is removed from the potential calcification image to obtain a final calcification image.
The estimating of calcification index in the thyroid calcification detecting module further comprises: recalculating the calcification index r from the final calcification imagecaleIf the calcification index rcaleIf > 0, it indicates the presence of calcification in the thyroid gland, and if the calcification index r is greater thancale0 indicates absence of calcification in the thyroid gland.
The last three-layer framework of the VGG-19 model framework is adjusted in the TNet model construction and training module, and specifically comprises the following steps: and adjusting the full connection layer, the softmax layer and the output layer of the VGG-19 model.
The thyroid calcification detection module further includes: prior to thyroid calcification detection, the image of interest was scaled down 30% towards the center of the thyroid to eliminate the effect of nodule envelope.
And the thyroid characteristic map generating module is used for selecting the 2 nd, 32 nd, 40 th, 60 th and 62 th characteristic maps from the 64 characteristic maps output by the first convolution layer in the TNet model to combine to obtain the thyroid characteristic map.
The image acquisition module of interest further comprises: the image of interest is resized to 224 x 224 pixels by bicubic interpolation.
Advantageous effects
Due to the adoption of the technical scheme, compared with the prior art, the invention has the following advantages and positive effects: the method introduces the idea of transfer learning and convolutional neural network to construct the TNet model, and enables the TNet model to have better identification performance by adjusting the structure of the TNet model and setting a series of parameters of the TNet model; the calcification detection device provided by the invention can effectively detect the calcification condition of the thyroid gland, has high identification accuracy, removes non-calcification areas caused by speckle noise by filtering Boolean mapping generated by a potential calcification image, enables calcification identification to be more accurate, and is convenient for doctors to make judgment better, faster and more accurate.
Drawings
FIG. 1 is a flow chart of an embodiment of the present invention;
fig. 2 is a schematic diagram of the embodiment of the present invention for removing non-calcified regions due to speckle noise.
Detailed Description
The invention will be further illustrated with reference to the following specific examples. It should be understood that these examples are for illustrative purposes only and are not intended to limit the scope of the present invention. Further, it should be understood that various changes or modifications of the present invention may be made by those skilled in the art after reading the teaching of the present invention, and such equivalents may fall within the scope of the present invention as defined in the appended claims.
The embodiment of the invention relates to a thyroid calcification detection device, which detects benign and malignant thyroid of an ultrasonic image (US) by introducing the ideas of Transfer Learning (TL) and a Convolutional Neural Network (CNN). the embodiment adopts a VGG-19 model and trains the VGG-19 model by utilizing ImageNet data set, the ImageNet data set comprises a large number of images and a large number of image types, so that the pretrained VGG-19 model can have better generalization capability, information corresponding to the pretrained VGG-19 model is transmitted to a new neural network model for detecting the benign and malignant thyroid, and the new convolutional neural network model for detecting the benign and malignant thyroid is called as a TNet model in the embodiment.
As shown in fig. 1, which is a flowchart of an embodiment of the present invention, a thyroid calcification detection apparatus in the embodiment mainly includes: the TNet model construction and training module, the interested image acquisition module, the thyroid benign and malignant detection module, the thyroid characteristic diagram generation module and the thyroid calcification detection module are respectively used for:
the TNet model construction and training module: constructing and training a VGG-19 model based on a convolutional neural network, and adjusting the final three-layer structure of the VGG-19 model; migrating the layer trained by the VGG-19 model to a new convolutional neural network model to construct a TNet initial model, and pre-training the TNet initial model to obtain the TNet model;
the image of interest acquisition module: and intercepting the thyroid gland boundary of the thyroid gland ultrasonic image in a mode of selecting the coordinate point of interest to obtain the image of interest.
The thyroid benign and malignant detection module: for inputting the image of interest into the TNet model, which detects the image of interest and generates as output a "follow-up" or "puncture".
The thyroid characteristic map generating module: and selecting preset feature maps from the feature maps output by the first convolution layer in the TNet model to combine to obtain the thyroid feature map.
The thyroid calcification detection module: used for carrying out calcification detection on the thyroid gland feature map, if the average brightness of the thyroid gland is less than or equal to a preset threshold value tcystThe thyroid gland is cyst; if the average thyroid brightness is larger than the preset threshold tcystThe thyroid is a non-cystic nodule and the non-cystic nodule is detected for calcification by calcified region detection and calcification index estimation.
In this embodiment, the thyroid of the patient is examined by an ultrasound apparatus, and an ultrasound image of the thyroid of the patient is acquired. Typically, the acquired thyroid ultrasound images capture most of the thyroid tissue, but the thyroid ultrasound images also have noise for reasons including:
(1) thyroid ultrasound images may contain non-tumor objects;
(2) capturing thyroid ultrasound images using different brands and models of ultrasound instruments;
(3) thyroid objects have different shapes, orientations, textures, and sizes;
(4) capturing the thyroid ultrasound images at different focal points and scales results in thyroid ultrasound images having different intensities and sizes.
Further, in this embodiment, the interesting image is extracted by intercepting the thyroid gland boundary in the thyroid gland ultrasonic image, so as to perform accurate identification, specifically: the doctor can intercept an interested coordinate point (X, y) epsilon X of the thyroid in the thyroid ultrasound image, namely the doctor can extract an interested image through manual marking interception, and the extracted interested image is used as the input of the TNet model after being unified in size so as to predict the benign and malignant conditions.
In the embodiment, a convolutional neural network VGG-19 model is used as a basis, the VGG-19 model is used for identifying an object from an image, and the parameters of the VGG-19 model are pre-trained on an ImageNet data set. The VGG-19 model has a network of 16 convolutional layers for extracting general features of an image (such as lines, shapes, edges, and textures) and 3 fully-connected layers (47 layers in total) for which weights can be learned, and contains millions of parameters.
The TNet model in this embodiment has two output categories, namely, a new full connection layer (fc8'), a softmax layer (prob ') and a classification output layer (output ') are added to the two output categories, wherein the output of the last full connection layer is fed to the binary class softmax (or normalized exponential function), so as to obtain two output classification result tags: "follow-up" or "puncture", where "follow-up" indicates benign thyroid and "puncture" indicates malignant thyroid.
In the embodiment, before detection, the TNet model needs to be trained through a preset thyroid ultrasound image, so that the TNet model has better generalization capability, and after training, the thyroid image to be detected can be preprocessed and then input into the TNet model for identification. Before inputting the image of interest into the TNet model, the image of interest captured by the doctor needs to be rescaled to 224 × 224 pixels using bicubic interpolation to normalize the image, and the zoomed image of interest is used as the input of the TNet model. Further, the network parameters of the TNet model are set as follows: the number of iterations is set to 15000, the initial learning rate is set to 0.001, and the batch size (mini-batch) is set to 8, and other network parameters are set to default values. Based on the parameter configuration, the accuracy of thyroid benign and malignant identification can be ensured, and according to the identification result, the TNet model provides a follow-up suggestion or a puncture suggestion.
The TNet model in this embodiment contains 16 convolutional layers, and each convolutional layer has a set of trainable convolution filters, so that useful features can be extracted from the thyroid ultrasound image, and the useful features are the basis for the subsequent identification of benign and malignant thyroid gland. By sliding the filter over the input image of interest and performing a convolution operation at each location, the results of the convolution operation will form a signature that is subsequently used to identify calcifications.
In the TNet model, some convolutional layers have a series of filters applicable to the input image, these filters generate many feature maps in each layer, and each layer extracts different types of features, for example, shape, boundary and texture features extracted in the initial layer, and after studying the salient features of these feature maps, the present embodiment selects 5 feature maps from the 64 feature maps output by the first convolutional layer in the TNet model, and the selected 5 feature maps are located at the following positions: the 5 feature maps were chosen for the 2 nd, 32 nd, 40 th, 60 th and 62 th feature maps because they contained features that, when combined, highlighted the thyroid border for subsequent calcification detection.
In this embodiment, the thyroid calcification detecting module is configured to perform calcification detection on the thyroid feature map, and if the average brightness of the thyroid is less than or equal to a preset threshold tcystThe thyroid gland is cyst; if the average thyroid brightness is larger than the preset threshold tcystThe thyroid is a non-cystic nodule and the non-cystic nodule is detected for calcification by calcified region detection and calcification index estimation.
Experience has shown that calcification only grows on thyroid tissue and not on fluid or nodular cysts. Calcifications are most likely to grow in the thyroid but not completely on the thyroid border, a simple cyst nodule should contain only liquid or a small amount of colloid. Therefore, the embodiment reduces the image of interest by 30% toward the center of the thyroid gland before detecting thyroid calcification, so as to eliminate the influence of the nodule envelope.
The detection of the calcified area in the thyroid calcification detection module comprises the following steps: thyroid ultrasound image I corresponding to the non-cystic nodulex,yMinus its corresponding thyroid profile I'x,yAnd obtaining a suspicious calcification image, wherein the formula is as follows:
Figure BDA0002615076570000061
wherein, Delta is Ix,y-I'x,yIf the suspicious calcification image I-(x, y) > Preset threshold tcaleIf so, determining that the non-cystic nodule has potential calcification and the suspicious calcification image is a potential calcification image; if the suspicious calcification image I-(x, y) is less than or equal to a preset threshold value tcaleThen the non-cystic nodules are not calcified.
The estimating of the calcification index in the thyroid calcification detecting module comprises: constructing a calcification index from the potential calcification image, the formula being:
Figure BDA0002615076570000062
wherein x ismaxAnd ymaxRespectively representing the total pixel number of an x axis and the total pixel number of a y axis in the potential calcification image;
Figure BDA0002615076570000063
the arg max function is used to express when the calcification index r iscaleWhen the fastest change is generated, the preset threshold value t iscaleIs the optimal threshold.
As shown in fig. 2, which is a schematic diagram of removing non-calcified regions due to speckle noise in the embodiment of the present invention, the detecting of calcified regions in the thyroid calcification detecting module further includes: filtering the boolean mapping generated by the latent calcification image by using a 1-pixel wide structuring element (1-pixel disk) for morphological opening to remove non-calcified regions (fragmented regions) due to speckle noise, in particular: and cutting each connecting area in the potential calcification image to obtain a cutting area.
Haralick texture features are extracted from a gray level co-occurrence matrix of each cutting region through a pre-trained multivariate Gaussian Bayes classifier, each cutting region is divided into a suspicious calcified region and a non-calcified region according to the Haralick texture features, the non-calcified region is removed from the potential calcified image, an upper graph in FIG. 2 comprises the suspicious calcified region and the non-calcified region (fragment region), and a lower graph in FIG. 2 shows that the non-calcified region (fragment region) is removed, and only the suspicious calcified region is reserved.
The detection of calcified areas in the thyroid calcification detection module further comprises: detecting the intensity value median around each suspicious calcification area, if the intensity value median around the suspicious calcification area is not more than a preset threshold tcolloidAnd if so, the suspicious calcification area is colloid, and the colloid is removed from the potential calcification image to obtain a final calcification image.
The estimating of calcification index in the thyroid calcification detecting module further comprises: recalculating the calcification index r from the final calcification imagecaleIf the calcification index rcaleIf > 0, it indicates the presence of calcification in the thyroid gland, and if the calcification index r is greater thancale0 indicates absence of calcification in the thyroid gland.
Therefore, in order to facilitate calcification detection, the calcification detection device provided by the invention combines a thyroid gland characteristic diagram by selecting the characteristic diagram preset in the TNet model, carries out calcification detection on the thyroid gland, removes the influence of a non-calcification area, has high identification accuracy and is convenient for doctors to judge better, faster and more accurately.

Claims (10)

1. A thyroid calcification detection apparatus, comprising:
the TNet model construction and training module: constructing and training a VGG-19 model based on a convolutional neural network, and adjusting the final three-layer structure of the VGG-19 model; migrating the layer trained by the VGG-19 model to a new convolutional neural network model to construct a TNet initial model, and pre-training the TNet initial model to obtain the TNet model;
an image of interest acquisition module: intercepting a thyroid gland boundary of the thyroid gland ultrasonic image in a mode of selecting an interested coordinate point to obtain an interested image;
thyroid benign and malignant detection module: for inputting the image of interest into the TNet model, which detects the image of interest and generates as output a "follow-up" or "puncture";
a thyroid profile generation module: the method comprises the steps of selecting preset feature maps from feature maps output by a first convolution layer in the TNet model to be combined to obtain a thyroid gland feature map;
thyroid calcification detection module: used for carrying out calcification detection on the thyroid gland feature map, if the average brightness of the thyroid gland is less than or equal to a preset threshold value tcystThe thyroid gland is cyst; if the average thyroid brightness is larger than the preset threshold tcystThe thyroid is a non-cystic nodule and the non-cystic nodule is detected for calcification by calcified region detection and calcification index estimation.
2. The thyroid calcification detection apparatus according to claim 1, wherein the detecting of calcified regions in the thyroid calcification detection module includes: thyroid ultrasound image I corresponding to the non-cystic nodulex,yMinus its corresponding thyroid profile I'x,yAnd obtaining a suspicious calcification image, wherein the formula is as follows:
Figure FDA0002615076560000011
wherein, Delta is Ix,y-I'x,yIf the suspicious calcification image I-(x, y) > Preset threshold tcaleIf so, determining that the non-cystic nodule has potential calcification and the suspicious calcification image is a potential calcification image; if the suspicious calcification image I-(x, y) is less than or equal to a preset threshold value tcaleThen the non-cystic nodules are not calcified.
3. The thyroid calcification detection apparatus as claimed in claim 2, wherein the thyroid calcification detection module in-calcification index estimation includes: constructing a calcification index from the potential calcification image, the formula being:
Figure FDA0002615076560000012
wherein x ismaxAnd ymaxRespectively representing the total pixel number of an x axis and the total pixel number of a y axis in the potential calcification image;
Figure FDA0002615076560000021
the arg max function is used to express when the calcification index r iscaleWhen the fastest change is generated, the preset threshold value t iscaleIs the optimal threshold.
4. The thyroid calcification detection apparatus according to claim 3, wherein the detecting of calcified regions in the thyroid calcification detection module further comprises: filtering the Boolean mapping generated by the latent calcification image by using a structuring element with a width of 1 pixel for morphological opening to remove non-calcified regions due to speckle noise, specifically: cutting each connecting area in the potential calcification image independently to obtain a cutting area;
extracting Haralick texture features from the gray level co-occurrence matrix of each cutting region through a pre-training multivariate Gaussian Bayes classifier, dividing each cutting region into a suspicious calcified region and a non-calcified region according to the Haralick texture features, and removing the non-calcified region from the potential calcified image.
5. The thyroid calcification detection apparatus according to claim 4, wherein the detecting of calcified regions in the thyroid calcification detection module further comprises: detecting the median value of the intensity values around each suspicious calcification area, and if the median value of the intensity values around the suspicious calcification areas is not more than a preset threshold tcolloidAnd if so, the suspicious calcification area is colloid, and the colloid is removed from the potential calcification image to obtain a final calcification image.
6. The thyroid calcification detection apparatus as recited in claim 5, wherein the thyroid calcification detection module in-calcification index estimation further comprises: recalculating the calcification index r from the final calcification imagecaleIf calcification indexrcaleIf > 0, it indicates the presence of calcification in the thyroid gland, and if the calcification index r is greater thancale0 indicates absence of calcification in the thyroid gland.
7. The thyroid calcification detection apparatus according to claim 1, wherein the last three-layer architecture of the VGG-19 model architecture in the TNet model construction and training module is specifically: and adjusting the full connection layer, the softmax layer and the output layer of the VGG-19 model.
8. The thyroid calcification detection apparatus as recited in claim 1, wherein the thyroid calcification detection module further comprises: prior to thyroid calcification detection, the image of interest was scaled down 30% towards the center of the thyroid to eliminate the effect of nodule envelope.
9. The thyroid calcification detection apparatus according to claim 1, wherein the thyroid feature map generation module is configured to select and combine the 2 nd, 32 nd, 40 th, 60 th, and 62 th feature maps from 64 feature maps output by the first convolutional layer in the TNet model to obtain a thyroid feature map.
10. The thyroid calcification detection apparatus as recited in claim 1, wherein the image acquisition module of interest further comprises: the image of interest is resized to 224 x 224 pixels by bicubic interpolation.
CN202010766984.XA 2020-08-03 2020-08-03 Thyroid calcification detection device Active CN112001895B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010766984.XA CN112001895B (en) 2020-08-03 2020-08-03 Thyroid calcification detection device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010766984.XA CN112001895B (en) 2020-08-03 2020-08-03 Thyroid calcification detection device

Publications (2)

Publication Number Publication Date
CN112001895A true CN112001895A (en) 2020-11-27
CN112001895B CN112001895B (en) 2021-04-02

Family

ID=73464031

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010766984.XA Active CN112001895B (en) 2020-08-03 2020-08-03 Thyroid calcification detection device

Country Status (1)

Country Link
CN (1) CN112001895B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113450326A (en) * 2021-06-28 2021-09-28 什维新智医疗科技(上海)有限公司 Blood vessel information extraction element of thyroid gland tubercle
CN113688930A (en) * 2021-09-01 2021-11-23 什维新智医疗科技(上海)有限公司 Thyroid nodule calcification recognition device based on deep learning
CN113808100A (en) * 2021-09-16 2021-12-17 什维新智医疗科技(上海)有限公司 Device for identifying rough calcification of breast nodules
CN113822864A (en) * 2021-09-16 2021-12-21 什维新智医疗科技(上海)有限公司 Mammary nodule calcification detection device

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8366619B2 (en) * 2009-05-13 2013-02-05 University Of Washington Nodule screening using ultrasound elastography
US20150221081A1 (en) * 2012-05-09 2015-08-06 Industry-University Cooperation Foundation Sogang University Method for discriminating between background and tissue of interest, and method and apparatus for generating photoacoustic images for detecting calcified tissue
CN105427296A (en) * 2015-11-11 2016-03-23 北京航空航天大学 Ultrasonic image low-rank analysis based thyroid lesion image identification method
CN106780448A (en) * 2016-12-05 2017-05-31 清华大学 A kind of pernicious sorting technique of ultrasonic Benign Thyroid Nodules based on transfer learning Yu Fusion Features
US20180053297A1 (en) * 2016-08-18 2018-02-22 Ohio University Methods and Apparatuses for Detection of Abnormalities in Low-Contrast Images
CN109493333A (en) * 2018-11-08 2019-03-19 四川大学 Ultrasonic Calcification in Thyroid Node point extraction algorithm based on convolutional neural networks

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8366619B2 (en) * 2009-05-13 2013-02-05 University Of Washington Nodule screening using ultrasound elastography
US20150221081A1 (en) * 2012-05-09 2015-08-06 Industry-University Cooperation Foundation Sogang University Method for discriminating between background and tissue of interest, and method and apparatus for generating photoacoustic images for detecting calcified tissue
CN105427296A (en) * 2015-11-11 2016-03-23 北京航空航天大学 Ultrasonic image low-rank analysis based thyroid lesion image identification method
US20180053297A1 (en) * 2016-08-18 2018-02-22 Ohio University Methods and Apparatuses for Detection of Abnormalities in Low-Contrast Images
CN106780448A (en) * 2016-12-05 2017-05-31 清华大学 A kind of pernicious sorting technique of ultrasonic Benign Thyroid Nodules based on transfer learning Yu Fusion Features
CN109493333A (en) * 2018-11-08 2019-03-19 四川大学 Ultrasonic Calcification in Thyroid Node point extraction algorithm based on convolutional neural networks

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
YIHONG CHEN 等: "Thyroid nodule detection using attenuation value based on non-enhancement CT images", 《 2ND IET INTERNATIONAL CONFERENCE ON BIOMEDICAL IMAGE AND SIGNAL PROCESSING (ICBISP 2017）》 *
李涛 等: "甲状腺结节锦化特征的自动提取方法研究", 《医持卫生装备 》 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113450326A (en) * 2021-06-28 2021-09-28 什维新智医疗科技(上海)有限公司 Blood vessel information extraction element of thyroid gland tubercle
CN113688930A (en) * 2021-09-01 2021-11-23 什维新智医疗科技(上海)有限公司 Thyroid nodule calcification recognition device based on deep learning
CN113688930B (en) * 2021-09-01 2024-03-19 什维新智医疗科技(上海)有限公司 Thyroid nodule calcification recognition device based on deep learning
CN113808100A (en) * 2021-09-16 2021-12-17 什维新智医疗科技(上海)有限公司 Device for identifying rough calcification of breast nodules
CN113822864A (en) * 2021-09-16 2021-12-21 什维新智医疗科技(上海)有限公司 Mammary nodule calcification detection device
CN113822864B (en) * 2021-09-16 2023-10-31 什维新智医疗科技(上海)有限公司 Breast nodule calcification detection device
CN113808100B (en) * 2021-09-16 2024-03-19 什维新智医疗科技(上海)有限公司 Breast nodule coarse calcification recognition device

Also Published As

Publication number Publication date
CN112001895B (en) 2021-04-02

Similar Documents

Publication Publication Date Title
CN112001895B (en) Thyroid calcification detection device
CN109886986B (en) Dermatoscope image segmentation method based on multi-branch convolutional neural network
CN109154978B (en) System and method for detecting plant diseases
Mao et al. Supervised learning-based cell image segmentation for p53 immunohistochemistry
US20210118144A1 (en) Image processing method, electronic device, and storage medium
CN108416360B (en) Cancer diagnosis system and method based on breast molybdenum target calcification features
JP3947109B2 (en) Computer-based image analysis
CN108090906B (en) Cervical image processing method and device based on region nomination
CN111028206A (en) Prostate cancer automatic detection and classification system based on deep learning
CN112380900A (en) Deep learning-based cervical fluid-based cell digital image classification method and system
JP2017519985A (en) Digital holographic microscopy data analysis for hematology
Bhattacharjee et al. Review on histopathological slide analysis using digital microscopy
CN109389129A (en) A kind of image processing method, electronic equipment and storage medium
CN110838100A (en) Colonoscope pathological section screening and segmenting system based on sliding window
CN108830149B (en) Target bacterium detection method and terminal equipment
WO2022095258A1 (en) Image object classification method and apparatus, device, storage medium and program
CN112132166A (en) Intelligent analysis method, system and device for digital cytopathology image
Ngugi et al. A new approach to learning and recognizing leaf diseases from individual lesions using convolutional neural networks
CN114648806A (en) Multi-mechanism self-adaptive fundus image segmentation method
CN111583226B (en) Cell pathological infection evaluation method, electronic device and storage medium
Sornapudi et al. Automated cervical digitized histology whole-slide image analysis toolbox
CN116758336A (en) Medical image intelligent analysis system based on artificial intelligence
CN112001894B (en) Thyroid gland boundary smoothness detection device
CN115100494A (en) Identification method, device and equipment of focus image and readable storage medium
Mohaimin et al. Automated method for the detection and segmentation of drusen in colour fundus image for the diagnosis of age‐related macular degeneration

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant