CN117393043A - Thyroid papilloma BRAF gene mutation detection device - Google Patents

Thyroid papilloma BRAF gene mutation detection device Download PDF

Info

Publication number
CN117393043A
CN117393043A CN202311684302.0A CN202311684302A CN117393043A CN 117393043 A CN117393043 A CN 117393043A CN 202311684302 A CN202311684302 A CN 202311684302A CN 117393043 A CN117393043 A CN 117393043A
Authority
CN
China
Prior art keywords
braf gene
feature
image
model
module
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
CN202311684302.0A
Other languages
Chinese (zh)
Other versions
CN117393043B (en
Inventor
吴健
林彦宏
应豪超
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.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
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 Zhejiang University ZJU filed Critical Zhejiang University ZJU
Priority to CN202311684302.0A priority Critical patent/CN117393043B/en
Publication of CN117393043A publication Critical patent/CN117393043A/en
Application granted granted Critical
Publication of CN117393043B publication Critical patent/CN117393043B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B20/00—ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations
    • G16B20/50—Mutagenesis
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • 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
    • G06N3/096—Transfer learning
    • 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
    • G06N3/0985—Hyperparameter optimisation; Meta-learning; Learning-to-learn
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
    • G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
    • G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
    • G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B40/00—ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Medical Informatics (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computational Linguistics (AREA)
  • Multimedia (AREA)
  • Databases & Information Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Evolutionary Biology (AREA)
  • Biotechnology (AREA)
  • Genetics & Genomics (AREA)
  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Public Health (AREA)
  • Epidemiology (AREA)
  • Bioethics (AREA)
  • Proteomics, Peptides & Aminoacids (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a thyroid papilloma BRAF gene mutation detection device, which comprises: obtaining a dyed pathological section image of thyroid papilloma, preprocessing to obtain a preprocessed image, and dividing an image tumor area from the preprocessed image by using a medical large model to serve as a division mask channel map; constructing a learning system comprising a teacher model and a student model, wherein in the teacher model, a preprocessed image and a segmentation mask channel diagram sequentially pass through a channel attention module, a first feature extraction module and a first feature prediction module to obtain a first prediction probability, and in the student model, the preprocessed image sequentially passes through a second feature extraction module and a second feature prediction module to obtain a second prediction probability; performing distillation learning on the student model based on the pre-trained teacher model based on the learning system to optimize student model parameters; the student model with optimized parameters is used as a BRAF gene mutation detection model for detecting the BRAF gene mutation condition of the thyroid papilloma.

Description

Thyroid papilloma BRAF gene mutation detection device
Technical Field
The invention belongs to the technical field of gene mutation detection and evaluation, and particularly relates to a device for detecting a BRAF gene mutation of thyroid papilloma.
Background
Thyropapillomas (PTC) are one of the most common malignant tumor types of thyroid, and molecular markers and gene mutation detection methods in diagnosis and treatment thereof are becoming increasingly important in clinical practice. Among them, BRAF gene mutation is widely studied and focused in thyroid papilloma.
The BRAF gene codes a kinase protein, is a key member of RAS/RAF/MEK/ERK signal channels, and participates in the biological processes of regulating and controlling cell growth, proliferation, differentiation and the like. The BRAF gene mutation is closely related to the definite diagnosis of cancer, and if the BRAF gene mutation is detected in clinic, doctors consider that patients are most likely to be cancerous.
The detection method for BRAF gene mutation plays an important role in clinical diagnosis and prognosis evaluation. Currently, commonly used biological detection methods include PCR (polymerase chain reaction), sanger sequencing, next Generation Sequencing (NGS), and the like. Biological methods, while having high accuracy and reliability, inevitably require higher time and economic costs.
With the rapid development of computer technology, the deep learning method has great potential and advantages in the field of medical images. The technological innovation is changing the medical image analysis mode, and brings new possibility and development opportunity for the medical care field. Deep learning is an important branch of artificial intelligence, and becomes a key sharps in medical image processing due to its strong self-learning and feature extraction capabilities. The method can extract high-level abstract features from medical images through large-scale data training and a complex neural network structure, and further realize intelligent analysis and diagnosis of images. The method has the advantage that complex image data can be processed efficiently, rules and modes hidden behind the images can be found, and thus doctors can be assisted in making more accurate diagnosis and treatment decisions.
For BRAF gene mutation existing in thyroid papilloma, certain modes or special morphological characteristics can be displayed in thyroid pathological section images, so that a model can be trained by using a deep learning algorithm, and complex characteristics in medical images can be automatically learned and extracted, thereby realizing more accurate diagnosis and analysis. However, this field still faces some challenges: first, the amount of data is insufficient. Obtaining a large scale well-labeled medical image dataset is critical to the success of the deep learning model. However, the marking data of the pathological images of the thyroid papilloma are relatively insufficient, limiting the effectiveness of model training. Second, diversity and generalization. The diversity and complexity of medical images presents challenges to the generalization ability of deep learning models. The generalization ability of the model under different lighting conditions in different devices requires more improvement. At present, the BRAF gene mutation can be rapidly and accurately predicted by a machine learning method which is not related to the pathological image diagnosis of the papillary thyroid tumor.
Disclosure of Invention
Aiming at the technical problems, the invention provides a BRAF gene mutation detection device, which can predict the BRAF gene mutation situation by automatically analyzing and extracting the characteristics of the dyed pathological section images, provides more auxiliary information for clinic and is expected to become an important auxiliary means for diagnosing and treating future thyroid cancer.
In order to achieve the above object, the present invention provides the following technical solutions:
in a first aspect, an apparatus for detecting a mutation of a BRAF gene of a papillary thyroid tumor according to an embodiment of the present invention includes a memory and one or more processors, where executable codes are stored in the memory, and when the one or more processors execute the executable codes, the following steps of detecting a mutation of a BRAF gene are implemented:
obtaining a dyed pathological section image of thyroid papilloma, preprocessing to obtain a preprocessed image, and dividing an image tumor area from the preprocessed image by using a medical large model to serve as a division mask channel map;
constructing a learning system comprising a teacher model and a student model, wherein in the teacher model, a preprocessing image and a segmentation mask channel diagram sequentially pass through a channel attention module, a first feature extraction module and a first feature prediction module to obtain a first prediction probability of whether a BRAF gene is mutated, and in the student model, the preprocessing image sequentially passes through a second feature extraction module and a second feature prediction module to obtain a second prediction probability of whether the BRAF gene is mutated;
performing distillation learning on the student model based on the pre-trained teacher model based on the learning system to optimize the student model parameters, wherein the total loss adopted during the distillation learning comprises difference loss between the first feature extraction module outputting the first feature and the second feature extraction module outputting the second feature, KL divergence loss between the first prediction probability and the second prediction probability, and second classification task loss between the second prediction probability and the real label;
the student model with optimized parameters is used as a BRAF gene mutation detection model for detecting the BRAF gene mutation condition of the thyroid papilloma.
Preferably, pretreatment is performed on the dyed pathological section image of the thyroid papilloma, including image denoising treatment, color normalization treatment and image size normalization treatment.
Preferably, in the channel attention module, the input image is represented as an RGB three-channel map and a segmentation mask channel map of the preprocessed image, and weights of the four channel maps are calculated by an average pooling layer and a full connection layer respectively, and the weights of the four channel maps are multiplied by the input image to obtain a weighted image.
Preferably, the first feature extraction module and the first feature prediction module adopt a network structure of a Resnet34, the first feature extraction module adopts a convolution module of the Resnet34 network to perform feature extraction according to the weighted image output by the channel attention module and output a first feature, and the first feature prediction module adopts a full-connection layer of the Resnet34 network to predict and output a first prediction probability of whether the BRAF gene is mutated according to the first feature output by the first feature extraction module.
Preferably, the second feature extraction module adopts a convolutional neural network with a residual structure, and the second feature prediction module adopts the convolutional neural network to extract the second feature according to the preprocessed image;
the second feature prediction module adopts a full-connection network, and the second feature prediction module predicts and outputs a second prediction probability of whether the BRAF gene is mutated or not according to the second feature output by the second feature extraction module by adopting the full-connection network.
Preferably, the teacher model is pre-trained before participating in distillation learning, and cross entropy between the first prediction probability output by the first feature prediction module and the real tag of the BRAF gene mutation is used as first classification task loss during pre-training, and parameters of the teacher model are updated through the first classification task loss.
Preferably, the difference loss is represented by L2 loss, expressed as:
；
wherein,indicating differential loss, < >>、/>Representing the first feature->And second feature->Respectively at->First->A dimensional coordinate;
KL divergence lossExpressed as:
；
wherein,will be->The sample image is predicted as +.>Predictive probability value for class->Will be->The sample image is predicted as +.>Predictive probability value for class->The number of sample images;
the second task classification penalty employs cross entropy penaltyTotal loss of distillation learning for student model->The method comprises the following steps:
；
wherein,、/>、/>is super parameter, the value range is (0, 1) and satisfies +.>。
Preferably, when the BRAF gene mutation condition is detected by using the BRAF gene mutation detection model, preprocessing the collected dyed pathological section image of the thyroid papilloma, inputting the obtained preprocessed image into the BRAF gene mutation detection model, and carrying out feature extraction and feature prediction sequentially through the second feature extraction module and the second feature prediction module to obtain a second prediction probability, wherein the second prediction probability can represent the result of whether the BRAF gene is mutated or not.
In a third aspect, an embodiment of the present invention provides a device for detecting a mutation of a BRAF gene of a papillary thyroid tumor, including:
the image acquisition and preprocessing unit is used for acquiring a dyed pathological section image of the thyroid papilloma and preprocessing the dyed pathological section image to obtain a preprocessed image, and dividing an image tumor area from the preprocessed image by using the medical large model to serve as a division mask channel map;
the learning system construction unit is used for constructing a learning system comprising a teacher model and a student model, wherein in the teacher model, a preprocessing image and a segmentation mask channel diagram sequentially pass through a channel attention module, a first feature extraction module and a first feature prediction module to obtain a first prediction probability of whether the BRAF gene is mutated, and in the student model, the preprocessing image sequentially passes through a second feature extraction module and a second feature prediction module to obtain a second prediction probability of whether the BRAF gene is mutated;
the distillation learning unit is used for carrying out distillation learning on the student model based on the pre-trained teacher model based on the learning system so as to optimize the student model parameters, and the total loss adopted during the distillation learning comprises the difference loss between the first feature extraction module outputting the first feature and the second feature extraction module outputting the second feature, the KL divergence loss between the first prediction probability and the second prediction probability, and the second classification task loss between the second prediction probability and the real label;
the mutation detection unit is used for taking the student model with optimized parameters as a BRAF gene mutation detection model to detect the BRAF gene mutation condition of the thyroid papilloma.
In a third aspect, an embodiment of the present invention provides a computer readable storage medium, on which a program is stored, which when executed by a processor, implements a BRAF gene mutation detection step implemented in the above-mentioned thyropapilloma BRAF gene mutation detection device.
According to the device for detecting the BRAF gene mutation of the thyroid papilloma, provided by the embodiment of the invention, the RGB channel and the segmentation information of pathological sections of the thyroid papilloma are extracted by utilizing a medical large model segmentation image and a channel attention mechanism, and then a light student model is trained by utilizing knowledge distillation based on a pre-trained teacher model. The student model does not need to additionally introduce segmentation labels, and simultaneously, the timeliness and the accuracy of the mutation detection of the BRAF gene of the papillary thyroid tumor are effectively improved, the diagnosis of a pathologist is assisted, and the workload of the doctor is reduced. Compared with the traditional method, the method reduces the interference of human subjective factors and improves the objectivity and stability of gene mutation detection.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a flowchart of a method for detecting mutation of a BRAF gene of a papillary thyroid tumor according to the embodiment;
FIG. 2 is a schematic diagram of the structure and training of a learning system including a teacher model and a student model provided in an embodiment;
FIG. 3 is an expanded schematic diagram of a channel attention module in a teacher model according to an embodiment;
FIG. 4 is a schematic structural diagram of a convolution module in a second feature extraction module according to an embodiment;
FIG. 5 is a schematic structural diagram of a second feature prediction module provided in an embodiment;
fig. 6 is a schematic structural diagram of a device for detecting mutation of a BRAF gene of a papillary thyroid tumor according to the embodiment;
FIG. 7 is a schematic diagram of a computing device provided by an embodiment.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the detailed description is presented by way of example only and is not intended to limit the scope of the invention.
Fig. 1 is a flowchart of a method for detecting mutation of the BRAF gene of the papillary thyroid tumor according to the embodiment. As shown in fig. 1, the method for detecting mutation of thyroid papilloma BRAF gene provided in the embodiment includes the following steps:
s1, obtaining a dyed pathological section image of the thyroid papilloma, preprocessing the dyed pathological section image to obtain a preprocessed image, and dividing an image tumor area from the preprocessed image by using a medical large model to serve as a division mask channel map.
In an embodiment, a tissue slice of thyroid papilloma is obtained by a hospital, research institution or medical image database. These sections are usually stained with Hematoxylin and Eosin (H & E) to show tissue structure and cell morphology, which is based on the different degree of binding of tissue structure to different dyes. Hematoxylin can stain a basophilic structure, which typically includes nucleic acid-containing moieties such as ribosomes, nuclei, and ribonucleic acid (RNA) -rich regions of the cytoplasm, etc., into a bluish violet color. Eosin can stain eosinophilic structures, which are usually composed of intracellular and intercellular proteins, such as lewis bodies, alcohol bodies, a large portion of the cytoplasm, etc., into pink colors, and then convert the pathological section into a digital image using a digital pathological scanner. And then, carrying out data preprocessing on the obtained pathological section image so as to improve the image quality.
In an embodiment, in order to improve the image quality, a pretreatment image is obtained by preprocessing a dyed pathological section image of the thyroid papilloma, the pretreatment image is used as a sample image, and meanwhile, whether a corresponding pathological section has thyroid BRAF gene mutation is used as a truth label of the sample image. The image preprocessing comprises image denoising processing, color normalization processing and image size normalization processing.
The image denoising process refers to smoothing an image by using a gaussian function.
The color normalization process can avoid influencing the effect of the model by the color, and the color normalization process is as follows:
；
wherein, the image color channel is in RGB form,represents the color mean value>Representing the color standard deviation, specifically adopting the color mean and the color standard deviation of the published large image data set ImageNet as prior values, and adopting the color mean and the color standard deviation as the prior values>Representing the original image +.>Representing the color normalized image.
The image size normalization process refers to uniformly scaling the image size to 224×224 pixels.
After the pre-treatment image is obtained, the image tumor area is segmented from the pre-treatment image by using the medical large model to be used as a segmentation mask channel map. Wherein the medical large model can segment the tumor area in the picture without additional training. In an embodiment, an open source medical large model SAM-Med2D is employed to segment the image. SAM-Med2D is a large model where SAM fine-tunes over a large scale medical image segmentation dataset that is more suitable for the medical field. And the segmentation mask channel map segmented by SAM-Med2D is spliced with the preprocessing image according to the channels to serve as an input image of the teacher model.
S2, constructing a learning system comprising a teacher model and a student model.
As shown in fig. 2, the learning system includes a teacher model and a student model, where the teacher model includes a channel attention module, a first feature extraction module, and a first feature prediction module, and in the teacher model, the preprocessing image and the segmentation mask channel map sequentially pass through the channel attention module, the first feature extraction module, and the first feature prediction module to obtain a first prediction probability of whether the BRAF gene is mutated.
As shown in fig. 3, in the channel attention module, the input image is represented as an RGB three-channel map and a segmentation mask channel map of the preprocessed image, and weights of the four channel maps are calculated by an average pooling layer and a full connection layer respectively, and the weights of the four channel maps are multiplied by the input image to obtain a weighted image. The first feature extraction module and the first feature prediction module adopt a Resnet34 network structure, the first feature extraction module adopts a convolution module of the Resnet34 network to extract features according to the weighted images output by the channel attention module and output first features, and the first feature prediction module adopts a full-connection layer of the Resnet34 network to predict and output first prediction probability of whether the BRAF gene is mutated according to the first features output by the first feature extraction module.
The student module comprises a second feature extraction module and a second feature prediction module, and in the student model, the preprocessed image sequentially passes through the second feature extraction module and the second feature prediction module to obtain a second prediction probability of whether the BRAF gene is mutated.
The second feature extraction module extracts second features of the preprocessed image by using a convolutional neural network with a residual structure. The method can be formed by stacking ten convolution modules, and particularly the convolution layer step length of the first five convolution modules is set to 2, so that the effect of downsampling is doubled, and the rest convolution layer step length is set to 1. As shown in fig. 4, each convolution module in the second feature extraction module includes two convolution layers and a ReLU activation layer disposed between the two convolution layers, where each convolution layer uses 64 convolution kernels of 3*3 size to extract features, and the ReLU layer applies ReLU function activation to the convolution layer output result, so that the student model has the ability to learn nonlinearity. In each convolution module, the input of the first convolution layer is added to the output of the second convolution layer to form a residual connection to ensure the effect of model deep learning.
The second feature prediction module classifies the second features output by the second feature extraction module by adopting a fully-connected network. As shown in fig. 5, specifically, two full-connection layers may be adopted, and nonlinear features in data are learned between the two layers through a ReLU activation function, and the nonlinear features output by the two full-connection layers are processed by a Softmax activation function to output a second prediction probability.
And S3, performing distillation learning on the student model based on the pre-trained teacher model based on the learning system so as to optimize the student model parameters.
In the embodiment, the teacher model is pre-trained before participating in the distillation learning, as shown in fig. 2, and cross entropy between the first prediction probability output by the first feature prediction module and the real tag of the BRAF gene mutation is used as the first classification task loss during the pre-training, and the parameters of the teacher model are updated through the first classification task loss.
After the teacher model is pre-trained, the student model is subjected to distillation learning based on the pre-trained teacher model to optimize the student model parameters. Total loss used in distillation learningThe method comprises the following steps:
；
wherein,、/>、/>is super parameter, the value range is (0, 1) and satisfies +.>；
Outputting a difference loss between the first feature and the second feature for the first feature extraction module, specifically using +.>A loss function, expressed as:
；
wherein,、/>representing the first feature->And second feature->Respectively at->First->A dimensional coordinate;
first predictive probability representing teacher model output +.>Second prediction probability output by student model +.>KL divergence loss between, expressed as:
；
wherein,will be->The sample image is predicted as +.>Predictive probability value for class->Will be->The sample image is predicted as +.>Predictive probability value for class->The number of sample images;
representing a second predictive probability->And trueThe second classification task loss between real labels, specifically cross entropy loss.
S4, the student model with optimized parameters is used as a BRAF gene mutation detection model for detecting the BRAF gene mutation condition of the thyroid papilloma.
After parameter optimization is carried out on a student model, the student model with the optimized parameter is used as a BRAF gene mutation detection model, and the BRAF gene mutation condition is detected by utilizing the BRAF gene mutation detection model, which comprises the following steps: preprocessing the collected dyed pathological section images of the papillary thyroid tumors, inputting the obtained preprocessed images into a BRAF gene mutation detection model, and sequentially carrying out feature extraction and feature prediction through a second feature extraction module and a second feature prediction module to obtain a second prediction probability, wherein the second prediction probability can represent the result of whether the BRAF gene is mutated or not.
Based on the same inventive concept, as shown in fig. 6, the embodiment of the invention further provides a device 60 for detecting a mutation of a thyroid papilloma BRAF gene, which comprises an image acquisition and preprocessing unit 61, a learning system construction unit 62, a distillation learning unit 63 and a mutation detection unit 64, wherein the image acquisition and preprocessing unit 61 is used for acquiring a dyed pathological section image of the thyroid papilloma and preprocessing to obtain a preprocessed image, and a medical large model is used for dividing an image tumor area from the preprocessed image to be used as a division mask channel diagram; the learning system construction unit 62 is configured to construct a learning system including a teacher model in which the preprocessed image and the segmentation mask channel map sequentially pass through the channel attention module, the first feature extraction module, and the first feature prediction module to obtain a first prediction probability of whether the BRAF gene is mutated, and a student model in which the preprocessed image sequentially passes through the second feature extraction module and the second feature prediction module to obtain a second prediction probability of whether the BRAF gene is mutated; the distillation learning unit 63 is configured to perform distillation learning on the student model based on the pre-trained teacher model based on the learning system to optimize parameters of the student model, where total loss adopted during the distillation learning includes a difference loss between the first feature extraction module outputting the first feature and the second feature extraction module outputting the second feature, a KL divergence loss between the first prediction probability and the second prediction probability, and a second classification task loss between the second prediction probability and the real tag; the mutation detection unit 64 is used for using a student model with optimized parameters as a BRAF gene mutation detection model to detect the mutation situation of the BRAF gene of the thyroid papilloma.
It should be noted that, when the thyroid papilloma BRAF gene mutation detection apparatus provided in the foregoing embodiment performs thyroid papilloma BRAF gene mutation detection, the division of the foregoing functional units should be exemplified, and the foregoing functional allocation may be performed by different functional units according to needs, that is, the internal structure of the terminal or the server is divided into different functional units, so as to complete all or part of the functions described above. In addition, the thyroid papilloma BRAF gene mutation detection device provided in the above embodiment belongs to the same concept as the thyroid papilloma BRAF gene mutation detection method embodiment, and the specific implementation process is detailed in the thyroid papilloma BRAF gene mutation detection method embodiment, which is not described here again.
Based on the same inventive concept, an embodiment further provides a computing device, including a memory and one or more processors, where the memory stores executable codes, and the one or more processors implement the method for detecting a mutation of a thyroid papilloma BRAF gene in steps S1-S4 when executing the executable codes.
As shown in fig. 7, the computing device provided by the embodiment includes, at a hardware level, hardware required by other services such as internal buses, network interfaces, and memories, in addition to the processor and the memory. The memory is a nonvolatile memory, and the processor reads the corresponding computer program from the nonvolatile memory into the memory and then runs the computer program to realize the method for detecting the mutation of the BRAF gene in the thyropapilloma in the steps S1-S4. Of course, other implementations, such as logic devices or combinations of hardware and software, are not excluded from the present invention, that is, the execution subject of the following processing flows is not limited to each logic unit, but may be hardware or logic devices.
Based on the same inventive concept, the embodiments also provide a computer readable storage medium having a program stored thereon, which when executed by a processor, implements steps S1 to S4 in the above-described method for detecting a mutation of a thyroid papilloma BRAF gene.
In embodiments, computer-readable media, including both permanent and non-permanent, removable and non-removable media, may be used to implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, read only optical disk read only memory (CD-ROM), digital Versatile Disks (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. Computer-readable media, as defined herein, does not include transitory computer-readable media (transmission media), such as modulated data signals and carrier waves.
The foregoing detailed description of the preferred embodiments and advantages of the invention will be appreciated that the foregoing description is merely illustrative of the presently preferred embodiments of the invention, and that no changes, additions, substitutions and equivalents of those embodiments are intended to be included within the scope of the invention.

Claims (10)

1. A device for detecting a mutation in a BRAF gene of a papillary thyroid tumor, comprising a memory and one or more processors, wherein executable code is stored in the memory, and wherein the one or more processors, when executing the executable code, implement the following steps of detecting a mutation in a BRAF gene:
obtaining a dyed pathological section image of thyroid papilloma, preprocessing to obtain a preprocessed image, and dividing an image tumor area from the preprocessed image by using a medical large model to serve as a division mask channel map;
constructing a learning system comprising a teacher model and a student model, wherein in the teacher model, a preprocessing image and a segmentation mask channel diagram sequentially pass through a channel attention module, a first feature extraction module and a first feature prediction module to obtain a first prediction probability of whether a BRAF gene is mutated, and in the student model, the preprocessing image sequentially passes through a second feature extraction module and a second feature prediction module to obtain a second prediction probability of whether the BRAF gene is mutated;
performing distillation learning on the student model based on the pre-trained teacher model based on the learning system to optimize the student model parameters, wherein the total loss adopted during the distillation learning comprises difference loss between the first feature extraction module outputting the first feature and the second feature extraction module outputting the second feature, KL divergence loss between the first prediction probability and the second prediction probability, and second classification task loss between the second prediction probability and the real label;
the student model with optimized parameters is used as a BRAF gene mutation detection model for detecting the BRAF gene mutation condition of the thyroid papilloma.
2. The device for detecting a mutation in a BRAF gene of a thyroid papilloma according to claim 1, wherein the pretreatment of the images of the dyed pathological section of the thyroid papilloma includes an image denoising process, a color normalization process, and an image size normalization process.
3. The device for detecting the mutation of the BRAF gene of the thyroid papilloma according to claim 1, wherein in the channel attention module, an input image is represented as an RGB three-channel map and a split mask channel map of a preprocessed image, a total of four channel maps calculate weights of the four channel maps through an average pooling layer and a full connection layer respectively, and the weights of the four channel maps are multiplied by the input image to obtain a weighted image.
4. The device for detecting the mutation of the BRAF gene of the thyroid papilloma according to claim 1, wherein the first feature extraction module and the first feature prediction module adopt a network structure of Resnet34, the first feature extraction module performs feature extraction according to the weighted image output by the channel attention module by adopting a convolution module of the Resnet34 network and outputs a first feature, and the first feature prediction module performs prediction according to the first feature output by the first feature extraction module by adopting a fully-connected layer of the Resnet34 network and outputs a first prediction probability of whether the BRAF gene is mutated.
5. The device for detecting the mutation of the thyroid papilloma BRAF gene according to claim 1, wherein the second feature extraction module adopts a convolutional neural network with a residual structure, and the second feature prediction module adopts the convolutional neural network to extract a second feature according to the preprocessed image;
the second feature prediction module adopts a full-connection network, and the second feature prediction module predicts and outputs a second prediction probability of whether the BRAF gene is mutated or not according to the second feature output by the second feature extraction module by adopting the full-connection network.
6. The device for detecting the mutation of the BRAF gene of the papillary thyroid tumor according to claim 1, wherein the teacher model is pre-trained before participating in the distillation learning, and cross entropy between the first prediction probability output by the first feature prediction module and the real tag of the BRAF gene mutation is used as a first classification task loss during the pre-training, and the teacher model parameter is updated through the first classification task loss.
7. The device for detecting a mutation in the BRAF gene of thyroid papilloma according to claim 1, wherein the differential loss is represented by L2 loss:
；
wherein,indicating differential loss, < >>、/>Representing the first feature->And second feature->Respectively at->First->A dimensional coordinate;
KL divergence lossExpressed as:
；
wherein,will be->The sample image is predicted as +.>Predictive probability value for class->Model teacher withThe sample image is predicted as +.>Predictive probability value for class->The number of sample images;
the second task classification penalty employs cross entropy penaltyTotal loss of distillation learning for student model->The method comprises the following steps:
；
wherein,、/>、/>is super parameter, the value range is (0, 1) and satisfies +.>。
8. The device for detecting the mutation of the thyroid papilloma BRAF gene according to claim 1, wherein when the BRAF gene mutation detection model is used for detecting the mutation condition of the thyroid papilloma BRAF gene, the collected dyed pathological section image of the thyroid papilloma is preprocessed, the preprocessed image is input into the BRAF gene mutation detection model, and the second prediction probability is obtained by carrying out feature extraction and feature prediction through the second feature extraction module and the second feature prediction module in sequence, wherein the second prediction probability can represent the result of whether the BRAF gene is mutated.
9. A device for detecting a mutation in the BRAF gene of thyroid papilloma, comprising:
the image acquisition and preprocessing unit is used for acquiring a dyed pathological section image of the thyroid papilloma and preprocessing the dyed pathological section image to obtain a preprocessed image, and dividing an image tumor area from the preprocessed image by using the medical large model to serve as a division mask channel map;
the learning system construction unit is used for constructing a learning system comprising a teacher model and a student model, wherein in the teacher model, a preprocessing image and a segmentation mask channel diagram sequentially pass through a channel attention module, a first feature extraction module and a first feature prediction module to obtain a first prediction probability of whether the BRAF gene is mutated, and in the student model, the preprocessing image sequentially passes through a second feature extraction module and a second feature prediction module to obtain a second prediction probability of whether the BRAF gene is mutated;
the distillation learning unit is used for carrying out distillation learning on the student model based on the pre-trained teacher model based on the learning system so as to optimize the student model parameters, and the total loss adopted during the distillation learning comprises the difference loss between the first feature extraction module outputting the first feature and the second feature extraction module outputting the second feature, the KL divergence loss between the first prediction probability and the second prediction probability, and the second classification task loss between the second prediction probability and the real label;
the mutation detection unit is used for taking the student model with optimized parameters as a BRAF gene mutation detection model to detect the BRAF gene mutation condition of the thyroid papilloma.
10. A computer-readable storage medium having stored thereon a program which, when executed by a processor, implements the BRAF gene mutation detection step implemented in the thyroid papilloma BRAF gene mutation detection apparatus of any one of claims 1-8.
CN202311684302.0A 2023-12-11 2023-12-11 Thyroid papilloma BRAF gene mutation detection device Active CN117393043B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311684302.0A CN117393043B (en) 2023-12-11 2023-12-11 Thyroid papilloma BRAF gene mutation detection device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311684302.0A CN117393043B (en) 2023-12-11 2023-12-11 Thyroid papilloma BRAF gene mutation detection device

Publications (2)

Publication Number Publication Date
CN117393043A true CN117393043A (en) 2024-01-12
CN117393043B CN117393043B (en) 2024-02-13

Family

ID=89441362

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311684302.0A Active CN117393043B (en) 2023-12-11 2023-12-11 Thyroid papilloma BRAF gene mutation detection device

Country Status (1)

Country Link
CN (1) CN117393043B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110472730A (en) * 2019-08-07 2019-11-19 交叉信息核心技术研究院(西安)有限公司 A kind of distillation training method and the scalable dynamic prediction method certainly of convolutional neural networks
CN112862756A (en) * 2021-01-11 2021-05-28 中国医学科学院北京协和医院 Method for identifying pathological change type and gene mutation in thyroid tumor pathological image
CN115136245A (en) * 2020-02-20 2022-09-30 因美纳有限公司 Knowledge-based distillation and gradient pruning compression of artificial intelligence-based base detectors
US20230267721A1 (en) * 2022-02-24 2023-08-24 Vinbrain Joint Stock Company Method and system for training a machine learning model for medical image classification
WO2023221954A1 (en) * 2022-05-19 2023-11-23 浙江大学 Pancreatic tumor image segmentation method and system based on reinforcement learning and attention

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110472730A (en) * 2019-08-07 2019-11-19 交叉信息核心技术研究院(西安)有限公司 A kind of distillation training method and the scalable dynamic prediction method certainly of convolutional neural networks
WO2021023202A1 (en) * 2019-08-07 2021-02-11 交叉信息核心技术研究院(西安)有限公司 Self-distillation training method and device for convolutional neural network, and scalable dynamic prediction method
CN115136245A (en) * 2020-02-20 2022-09-30 因美纳有限公司 Knowledge-based distillation and gradient pruning compression of artificial intelligence-based base detectors
CN112862756A (en) * 2021-01-11 2021-05-28 中国医学科学院北京协和医院 Method for identifying pathological change type and gene mutation in thyroid tumor pathological image
US20230267721A1 (en) * 2022-02-24 2023-08-24 Vinbrain Joint Stock Company Method and system for training a machine learning model for medical image classification
WO2023221954A1 (en) * 2022-05-19 2023-11-23 浙江大学 Pancreatic tumor image segmentation method and system based on reinforcement learning and attention

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
胡屹杉;秦品乐;曾建潮;柴锐;王丽芳;: "结合分段频域和局部注意力的超声甲状腺分割", 中国图象图形学报, no. 10 *
高颖;沈兵;张建英;石峰;昌红;慕小京;: "甲状腺肿瘤组织中BrafV600E基因突变的检测", 诊断病理学杂志, no. 03 *

Also Published As

Publication number Publication date
CN117393043B (en) 2024-02-13

Similar Documents

Publication Publication Date Title
Yasrab et al. RootNav 2.0: Deep learning for automatic navigation of complex plant root architectures
US20220237788A1 (en) Multiple instance learner for tissue image classification
Duggento et al. Deep computational pathology in breast cancer
Xie et al. Instance-aware self-supervised learning for nuclei segmentation
Pan et al. Mitosis detection techniques in H&E stained breast cancer pathological images: A comprehensive review
Olimov et al. FU-Net: fast biomedical image segmentation model based on bottleneck convolution layers
CN112347977B (en) Automatic detection method, storage medium and device for induced pluripotent stem cells
Wang et al. A generalizable and robust deep learning algorithm for mitosis detection in multicenter breast histopathological images
Li et al. FecalNet: Automated detection of visible components in human feces using deep learning
Momeni et al. Deep recurrent attention models for histopathological image analysis
CN114445356A (en) Multi-resolution-based full-field pathological section image tumor rapid positioning method
CN114511523B (en) Gastric cancer molecular subtype classification method and device based on self-supervision learning
CN115409804A (en) Method for identifying and marking focus region of mammary gland magnetic resonance image and predicting curative effect
Liang et al. Weakly supervised deep nuclei segmentation with sparsely annotated bounding boxes for dna image cytometry
Khattar et al. Computer assisted diagnosis of skin cancer: a survey and future recommendations
Song et al. A new convolutional neural network architecture for automatic segmentation of overlapping human chromosomes
Rajput et al. A transfer learning-based brain tumor classification using magnetic resonance images
CN117393043B (en) Thyroid papilloma BRAF gene mutation detection device
Guo et al. Zero shot augmentation learning in internet of biometric things for health signal processing
Salsabili et al. Multiresolution semantic segmentation of biological structures in digital histopathology
Tian et al. Semi-supervised nuclei detection in histopathology images via location-aware adversarial image reconstruction
CN112489012A (en) Neural network architecture method for CT image recognition
Mudduluru et al. Improving medical image segmentation and classification using a novel joint deep learning model
Hasan et al. Nuclei segmentation in er-ihc stained histopathology images using mask r-cnn
Lu et al. Boundary-aware polyp segmentation network

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