CN116186574A - Thyroid sampling data identification method based on artificial intelligence - Google Patents
Thyroid sampling data identification method based on artificial intelligence Download PDFInfo
- Publication number
- CN116186574A CN116186574A CN202211101689.8A CN202211101689A CN116186574A CN 116186574 A CN116186574 A CN 116186574A CN 202211101689 A CN202211101689 A CN 202211101689A CN 116186574 A CN116186574 A CN 116186574A
- Authority
- CN
- China
- Prior art keywords
- tensor
- output
- thyroid
- dimension
- block
- 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
Links
- 210000001685 thyroid gland Anatomy 0.000 title claims abstract description 76
- 238000005070 sampling Methods 0.000 title claims abstract description 48
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000013473 artificial intelligence Methods 0.000 title claims abstract description 23
- 230000004927 fusion Effects 0.000 claims abstract description 25
- 230000005284 excitation Effects 0.000 claims abstract description 21
- 238000000605 extraction Methods 0.000 claims abstract description 19
- 230000005856 abnormality Effects 0.000 claims abstract description 8
- 210000002569 neuron Anatomy 0.000 claims description 27
- 230000002779 inactivation Effects 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 11
- 238000011176 pooling Methods 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 8
- 238000012549 training Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 7
- 238000013507 mapping Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 210000004205 output neuron Anatomy 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 230000008859 change Effects 0.000 claims description 3
- 238000009826 distribution Methods 0.000 claims description 3
- 210000002364 input neuron Anatomy 0.000 claims description 3
- 238000003325 tomography Methods 0.000 claims description 3
- 238000001514 detection method Methods 0.000 abstract description 4
- 238000002059 diagnostic imaging Methods 0.000 abstract description 2
- 238000003384 imaging method Methods 0.000 description 11
- 238000010586 diagram Methods 0.000 description 9
- 238000005516 engineering process Methods 0.000 description 9
- 208000024770 Thyroid neoplasm Diseases 0.000 description 7
- 238000004590 computer program Methods 0.000 description 7
- 230000006378 damage Effects 0.000 description 5
- 238000002593 electrical impedance tomography Methods 0.000 description 5
- 238000005259 measurement Methods 0.000 description 5
- 208000013076 thyroid tumor Diseases 0.000 description 5
- 210000001519 tissue Anatomy 0.000 description 5
- 238000003745 diagnosis Methods 0.000 description 4
- 238000003860 storage Methods 0.000 description 4
- 208000009453 Thyroid Nodule Diseases 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 206010028980 Neoplasm Diseases 0.000 description 1
- 208000027418 Wounds and injury Diseases 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000002591 computed tomography Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000002847 impedance measurement Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 208000014674 injury Diseases 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000005865 ionizing radiation Effects 0.000 description 1
- 210000000867 larynx Anatomy 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000013421 nuclear magnetic resonance imaging Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000001575 pathological effect Effects 0.000 description 1
- 230000007170 pathology Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 210000000664 rectum Anatomy 0.000 description 1
- 238000007493 shaping process Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 210000000534 thyroid cartilage Anatomy 0.000 description 1
- 210000003437 trachea Anatomy 0.000 description 1
- 210000005062 tracheal ring Anatomy 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/05—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves
- A61B5/053—Measuring electrical impedance or conductance of a portion of the body
- A61B5/0536—Impedance imaging, e.g. by tomography
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/42—Detecting, measuring or recording for evaluating the gastrointestinal, the endocrine or the exocrine systems
- A61B5/4222—Evaluating particular parts, e.g. particular organs
- A61B5/4227—Evaluating particular parts, e.g. particular organs endocrine glands, i.e. thyroid, adrenals, hypothalamic, pituitary
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/30—Assessment of water resources
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Pathology (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- Veterinary Medicine (AREA)
- Artificial Intelligence (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Physiology (AREA)
- Fuzzy Systems (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Radiology & Medical Imaging (AREA)
- Psychiatry (AREA)
- Signal Processing (AREA)
- Urology & Nephrology (AREA)
- Endocrinology (AREA)
- Gastroenterology & Hepatology (AREA)
- Image Processing (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention discloses a thyroid sampling data identification method based on artificial intelligence, which comprises the following steps: the bioelectrical impedance information collection platform is used for collecting resistivity values of thyroid positions of patients to be detected under different frequency excitation source frequencies, the sampled data are preprocessed and then input into the thyroid identification model, and the network structure of the thyroid identification model comprises three parts: a feature extraction part, a feature fusion part and a result output part. Inputting the sampling data of the object to be tested into a thyroid identification model, and finally outputting the model to predict whether thyroid abnormality exists in the object to be tested; the identification method can identify the thyroid condition of the object to be detected by analyzing the intrinsic characteristics of the data without medical imaging, and can greatly simplify the detection flow.
Description
Technical Field
The invention relates to the technical field of medical intelligent identification, in particular to a thyroid sampling data identification method based on artificial intelligence.
Background
Thyroid nodule refers to a tumor in thyroid gland, and in view of the high incidence rate of thyroid nodule, if as many thyroid tumors as possible can be identified by a non-surgical mode, the number of unnecessary diagnostic operations can be greatly reduced, the damage to patients caused by the operations can be reduced, and limited medical resources can be more reasonably applied.
In the aspect of medical diagnosis, pathological tissues in a human body can be found by carrying out high-definition imaging on a detected object to replace a mode of surgical operation diagnosis, so that unnecessary injury to the human body is greatly reduced, pain caused by operation is relieved, but the current clinical common medical imaging technology such as electronic computer tomography has higher imaging precision and stable performance, and ionizing radiation hazard can cause damage to the human tissues to a certain extent; although nuclear magnetic resonance imaging and ultrasonic technology do not damage human tissues with radioactivity, the two technologies neglect the interaction between ultrasonic waves and human tissues and filter a lot of useful information, so that certain problems exist in aspects of definition, resolution, accuracy and the like.
The electrical impedance tomography technology has small damage to human body and low cost, and can realize real-time imaging. Through the development of the last forty years, the excitation signal of the electrical impedance tomography technology is developed from single-frequency excitation to multi-frequency excitation, and the excitation signal adopts a current signal; the front-end acquisition platform is mainly converted from analog technology to digital technology, and in particular in the detection stage of signals, digital signal processing is often carried out after acquisition by a high-speed analog-digital conversion circuit. In the aspect of image reconstruction, artificial intelligence processing methods are added in recent years, and the aim is to improve the imaging resolution. However, the current electrical impedance tomography system has technical problems of nonlinearity, morbidity, discomfort and the like in the imaging process, so that the imaging quality effect is not ideal enough, the imaging resolution is low, and the electrical impedance tomography system has not been applied to clinic.
In view of the advantages of electrical impedance tomography and imaging problems, it is desirable to provide a method for simply and accurately identifying thyroid sampling data without the need for imaging procedures.
Disclosure of Invention
Aiming at the defects or improvement demands of the prior art, the invention provides an artificial intelligence-based thyroid sampling data identification method, which uses a bioelectrical impedance measurement technology to obtain sampling data, and effectively extracts and fuses the data, thereby effectively avoiding the defect that the imaging effect of the impedance tomography technology is not ideal, realizing identification of the sampling data without imaging, greatly simplifying the diagnosis process and having high accuracy.
According to an embodiment of the invention, there is provided an artificial intelligence based thyroid sampling data identification method, comprising the steps of:
step 1, applying an excitation source to the front part of a neck of an object to be detected by using a bioelectrical impedance tomography front end acquisition platform, and acquiring sampling data of the object to be detected under different excitation conditions;
step 2, preprocessing the sampling data to obtain a data set, and sending the data set into a thyroid identification model, wherein the thyroid identification model comprises a feature extraction part, a feature fusion part and a result output part;
and step 3, identifying whether the sampling data exceeds a preset range or not according to the output of the thyroid identification model result output part.
Further, the pretreatment step in the step 2 is as follows: the sampling data comprises the frequency of an excitation source and the resistivity corresponding to the frequency of the excitation source, the sampling data is subjected to type conversion, the sampling data is converted into two-dimensional data in a word embedding mode, and finally the two-dimensional data is converted into three-dimensional data in a dimension mapping mode, so that the data set is formed.
Further, the feature extraction part in step 2 includes 4 transform blocks, the feature fusion part includes 3 convolution blocks, and the structure output part includes 2 fully connected layers and 1 softmax layer.
Further, each transducer block in the feature extraction part comprises two transducer layers, the first transducer layer is a global maximum pooling layer added after the first normalized sublayer of the original transducer layer, and no feature addition operation is performed in the transducer layer; the second transducer layer maintains the original transducer layer structure unchanged.
Further, the specific operation of the transducer block is as follows: after tensor is input into a transitor block, entering a first transitor layer, firstly passing through a multi-head attention machine sublayer and a normalization sublayer, outputting an attention tensor, then carrying out global maximization pooling to obtain a first dimension-unchanged weight tensor, and multiplying the attention tensor and the weight tensor in a first dimension mode with the two dimensions of 1 to obtain an attention sample dimension tensor; then performing random inactivation operation on the neurons to obtain output tensors of the first part of the first transducer layer;
combining the last two dimensions of the output tensor of the first part of the first transducer layer into 1 dimension, and performing two full join operations: the first full-connection operation is to change the number of neurons in the last dimension into 1/16 of the original number, then to perform the second full-connection operation, to restore the number of neurons in the last dimension back to the original number, and to restore the tensor dimension back to the dimension before the first full-connection operation;
then performing random inactivation operation on the neurons to obtain output tensor of the first transducer layer;
the input of the second transducer layer is the output tensor of the first transducer layer, the input of the second transducer layer is firstly passed through a multi-head attention machine sublayer and a normalization sublayer, then the random inactivation of neurons is carried out, and the obtained tensor is added with the output tensor of the first transducer layer to obtain the output tensor of the first part of the second transducer layer;
combining the last two dimensions of the output tensor of the first part of the second transducer layer into 1 dimension, and performing two full join operations: the first full-connection operation changes the number of neurons in the last dimension into 1/16 of the original number, then the second full-connection operation is carried out, the number of neurons in the last dimension is restored to the original number, and the tensor dimension is restored to the dimension before the first full-connection operation;
then outputting a tensor during the random inactivation operation of the neuron; this tensor is added to the output tensor of the first part of the second transducer layer to obtain the final output tensor of the transducer block.
Further, the specific processing procedure of the convolution block in the feature fusion part is as follows:
after tensor is input into a convolution block, 3×3 convolution is performed to output a first tensor, the dimension of the first tensor is unchanged, and then 1×1 max pooling operation is performed to obtain a second tensor with the first dimension of 1 and the other two dimensions unchanged; multiplying the second tensor by the first tensor output by the 3×3 convolution to obtain a feature tensor with more focused attributes and word embedding dimensions, splicing the feature tensor with the corresponding tensor output by the transform block, performing 1×1 convolution on the spliced tensor, and restoring the first dimension of the spliced tensor to the dimension of the input convolution block without changing the other two dimensions, thereby obtaining the output result of the convolution block.
Further, the specific operation of the thyroid recognition model is as follows:
inputting the data set subjected to dimension mapping into a thyroid identification model, and sequentially passing through a feature extraction part, a feature fusion part and a result output part;
firstly, the data set is sent to a first transducer block in a feature extraction part to obtain an output tensor, then the input of a second transducer block is the output tensor of the first transducer block, the output tensor of the second transducer block is obtained through the second transducer block, and then the 3 rd and 4 th transducer blocks perform the same operation to obtain corresponding output tensors, wherein the dimension of the output tensor of each transducer block is always the same as that of the input tensor;
then, entering a feature fusion part, wherein the input of the 1 st convolution block is the output tensor of the 4 th convolution block, when the 1 st convolution block feature tensor is obtained, splicing the feature tensor with the output tensor of the 3 rd convolution block, and then restoring the first dimension of the spliced tensor to the dimension size when the 1 st convolution block is input through the 1X 1 convolution of the last layer in the 1 st convolution block to obtain the output tensor of the 1 st convolution block; the input of the 2 nd convolution block is the output tensor of the 1 st convolution block, when the characteristic tensor of the 2 nd convolution block is obtained, the characteristic tensor is spliced with the output tensor of the 2 nd convolution block, and then the first dimension of the spliced tensor is restored to the dimension size when the 2 nd convolution block is input through the 1X 1 convolution of the last layer in the 2 nd convolution block, so that the output tensor of the 2 nd convolution block is obtained; the input of the 3 rd convolution block is the output tensor of the 2 nd convolution block, when the characteristic tensor of the 3 rd convolution block is obtained, the characteristic tensor is spliced with the output tensor of the 1 st convolution block, then the first dimension of the spliced tensor is restored to the dimension size when the 3 rd convolution block is input through the 1X 1 convolution of the last layer in the 3 rd convolution block, and finally the output tensor of the characteristic fusion part is obtained;
finally, the input of the result output part is the output tensor of the feature fusion part, wherein the last two-dimensional data of the input tensor of the result output part is combined firstly, and then the number of the last one-dimensional neurons is adjusted by using the first full-connection layer, namely, the number of the output neurons is 1/4 of the number of the input neurons; and then using a second full connection layer to adjust the output neurons to the final classification number, and finally calling a softmax function to adjust the output to the probability distribution.
Further, the thyroid recognition model uses a loss function in the training process as follows:
wherein y represents sample tag data, and the tag data y is obtained in the step 1 and indicates whether thyroid abnormality exists in the object to be detected corresponding to the group of index data; y is pred Representing the prediction probability.
Another object of the present invention is to provide an electronic device comprising a memory and a processor for executing the steps of
The thyroid gland identification model stored in the memory is used for realizing the thyroid gland sampling data identification method based on artificial intelligence.
In general, the above technical solutions conceived by the present invention, compared with the prior art, enable the following beneficial effects to be obtained: compared with the prior art, the method and the device have the advantages that the bioelectrical impedance information collection platform is used for collecting the resistivity data of the thyroid gland part, the resistivity data are preprocessed, the improved transformer block structure is used for carrying out feature extraction and feature fusion operation based on the combination of the convolutional neural network and the transformer block, more data are obtained, the correlation information among the data is obtained, the difference among the data is increased, and therefore the follow-up classification result is more accurate. In addition, the thyroid identification model employs an exponential loss function to calculate and adjust training so that the network converges more quickly.
According to the method, whether the detected object is abnormal or not can be identified based on the resistivity data of the thyroid gland of the detected object, shooting or image generation is not needed, the detection flow is simplified, and the accuracy is high.
Drawings
FIG. 1 is a schematic flow chart of an artificial intelligence based thyroid sampling data identification method in an embodiment of the invention;
fig. 2 is a schematic diagram of a feature extraction network and a feature fusion network of an artificial intelligence based thyroid identification model in an embodiment of the invention.
Detailed Description
In order to better understand the technical solutions in the present application, the following description will clearly and completely describe the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only some embodiments of the present application, not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present disclosure.
It should be noted that the terms "first," "second," and "second" are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implying a number of technical features being indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. In the description of the present application, the meaning of "a plurality" or "a number" is two or more, unless explicitly defined otherwise.
The term "and/or" in the present invention is merely an association relation describing the association object, and indicates that three kinds of relations may exist, for example, a and/or B may indicate: the three cases of A alone, B alone, and both A and B exist. It will be apparent that the described embodiments are some, but not all, embodiments of the invention.
In one implementation of the present invention, fig. 1 is a schematic diagram of steps of an artificial intelligence based thyroid sampling data identification method according to the present invention, as shown in fig. 1, where the method includes the following steps:
step 1: the method comprises the steps of applying an excitation source to the front part of the neck of a to-be-detected object, applying an external excitation source to the sampling area, acquiring resistivity data in the sampling data and recording the resistivity data as thyroid part resistivity of the to-be-detected object under the excitation source, wherein the data to be sampled are thyroid measurement data, the left and right thyroid leaves are cone-shaped and are attached to the sides of the larynx and the trachea, the upper end of the thyroid measurement data reach the middle part of the thyroid cartilage, the lower end of the thyroid measurement data are against a fourth tracheal ring, the size of the thyroid measurement data is about 5cm, and the width of the thyroid measurement data is about 2.4 cm.
The invention adopts the biological impedance acquisition platform to acquire the resistivity of thyroid parts of the object to be detected under different excitation sources, wherein the biological impedance acquisition platform is not related to the actual invention point of the scheme, and is not repeated here.
Wherein the frequency of the excitation source is sequentially selected from low to high as follows: 10000 11874.14, 14592.33, 17932.276, 22037.87, 27082.71, 33282.39, 42330.9, 53839.43, 61770.64, 75910.98, 93288.28, 114643.54, 136129.31, 167291.61, 205587.48, 232649.92, 270291.61, 300000 in hertz (Hz). And the unit of resistivity at each excitation source measured at the object to be measured is ohm-cm (Ω -cm), 19 data indexes can be detected for each object to be measured.
Step 2: preprocessing the acquired sampling data and sorting the data into a data set. The step of preprocessing the collected sampling data comprises the following steps:
the sampling data comprises index data, namely the frequency of an excitation source and the resistivity corresponding to the frequency of the excitation source, the index data is subjected to type conversion, the index data is converted into two-dimensional data in a word embedding mode, and finally the two-dimensional data is converted into three-dimensional data in a dimension mapping mode, so that a data set is formed.
Inputting sample data in the data set into a thyroid identification model, wherein the thyroid identification model comprises a feature extraction part, a feature fusion part and a result output part;
step 3: and identifying whether the sampling data exceeds a preset range or not according to the output of the thyroid identification model result output part.
As an implementation manner of the present invention, fig. 2 is a schematic diagram of a feature extraction network and a feature fusion network based on an artificial intelligence thyroid identification model in an embodiment of the present invention, and as shown in fig. 2, a feature extraction part 101 includes 4 transform blocks, a feature fusion part 102 includes 3 convolution blocks, and a structure output part includes 2 full connection layers and 1 softmax layer.
Each transducer block in the feature extraction part comprises two transducer layers, wherein the first transducer layer is a global maximum pooling layer added after the first normalized sublayer of the original transducer layer, and no feature addition operation exists in the transducer layers; the second transducer layer maintains the original transducer layer structure unchanged.
Further, the specific operation of the transducer block is as follows:
s1: after tensor is input into a transitor block, entering a first transitor layer, firstly passing through a multi-head attention machine sublayer and a normalization sublayer, outputting an attention tensor, then carrying out global maximization pooling to obtain a first dimension-unchanged weight tensor, and multiplying the attention tensor and the weight tensor in a first dimension mode with the two dimensions of 1 to obtain an attention sample dimension tensor; then performing random inactivation operation on the neurons to obtain output tensors of the first part of the first transducer layer; the global maximum pooling layer is added after the first normalized sublayer of the original transducer layer, the maximum value weight tensor of the word embedding dimension can be obtained, the attention tensor and the maximum value weight tensor are multiplied, the value characteristics can be further enhanced, and the difference of the characteristics is enlarged;
combining the last two dimensions of the output tensor of the first part of the first transducer layer into 1 dimension, and performing two full join operations: the first full-connection operation is to change the number of neurons in the last dimension into 1/16 of the original number, then to perform the second full-connection operation, to restore the number of neurons in the last dimension back to the original number, and to restore the tensor dimension back to the dimension before the first full-connection operation;
then performing random inactivation operation on the neurons to obtain output tensor of the first transducer layer;
s2: the input of the second transducer layer is the output tensor of the first transducer layer, the input of the second transducer layer is firstly passed through a multi-head attention machine sublayer and a normalization sublayer, then the random inactivation of neurons is carried out, and the obtained tensor is added with the output tensor of the first transducer layer to obtain the output tensor of the first part of the second transducer layer;
combining the last two dimensions of the output tensor of the first part of the second transducer layer into 1 dimension, and performing two full join operations: the first full-connection operation changes the number of neurons in the last dimension into 1/16 of the original number, then the second full-connection operation is carried out, the number of neurons in the last dimension is restored to the original number, and the tensor dimension is restored to the dimension before the first full-connection operation;
then outputting a tensor during the random inactivation operation of the neuron; this tensor is added to the output tensor of the first part of the second transducer layer to obtain the final output tensor of the transducer block.
Further, the specific processing procedure of the convolution block in the feature fusion portion is as follows:
after tensor is input into a convolution block, 3×3 convolution is performed to output a first tensor, the dimension of the first tensor is unchanged, and then 1×1 max pooling operation is performed to obtain a second tensor with the first dimension of 1 and the other two dimensions unchanged; multiplying the second tensor by the first tensor output by the 3×3 convolution to obtain a feature tensor with more focused attributes and word embedding dimensions, splicing the feature tensor with the corresponding tensor output by the transform block, performing 1×1 convolution on the spliced tensor, and restoring the first dimension of the spliced tensor to the dimension of the input convolution block without changing the other two dimensions, thereby obtaining the output result of the convolution block.
As one embodiment of the present invention, the network structure of the thyroid recognition model specifically operates as follows:
inputting the data set subjected to dimension mapping into a thyroid identification model, and sequentially passing through a feature extraction part, a feature fusion part and a result output part;
firstly, the data set is sent to a first transducer block in a feature extraction part to obtain an output tensor, then the input of a second transducer block is the output tensor of the first transducer block, the output tensor of the second transducer block is obtained through the second transducer block, and then the 3 rd and 4 th transducer blocks perform the same operation to obtain corresponding output tensors, wherein the dimension of the output tensor of each transducer block is always the same as that of the input tensor;
then, entering a feature fusion part, wherein the input of the 1 st convolution block is the output tensor of the 4 th convolution block, when the 1 st convolution block feature tensor is obtained, splicing the feature tensor with the output tensor of the 3 rd convolution block, and then restoring the first dimension of the spliced tensor to the dimension size when the 1 st convolution block is input through the 1X 1 convolution of the last layer in the 1 st convolution block to obtain the output tensor of the 1 st convolution block; the input of the 2 nd convolution block is the output tensor of the 1 st convolution block, when the characteristic tensor of the 2 nd convolution block is obtained, the characteristic tensor is spliced with the output tensor of the 2 nd convolution block, and then the first dimension of the spliced tensor is restored to the dimension size when the 2 nd convolution block is input through the 1X 1 convolution of the last layer in the 2 nd convolution block, so that the output tensor of the 2 nd convolution block is obtained; the input of the 3 rd convolution block is the output tensor of the 2 nd convolution block, when the characteristic tensor of the 3 rd convolution block is obtained, the characteristic tensor is spliced with the output tensor of the 1 st convolution block, then the first dimension of the spliced tensor is restored to the dimension size when the 3 rd convolution block is input through the 1X 1 convolution of the last layer in the 3 rd convolution block, and finally the output tensor of the characteristic fusion part is obtained;
finally, the input of the result output part is the output tensor of the feature fusion part, wherein the last two-dimensional data of the input tensor of the result output part is combined firstly, and then the number of the last one-dimensional neurons is adjusted by using the first full-connection layer, namely, the number of the output neurons is 1/4 of the number of the input neurons; and then using a second full connection layer to adjust the output neurons to the final classification number, and finally calling a softmax function to adjust the output to the probability distribution.
As one embodiment of the invention, when training the thyroid identification model, the sampling data also comprises label data, namely whether thyroid abnormality exists in the object to be detected corresponding to the group of index data. When training the model, the tag data is converted into a shaping tensor. The thyroid recognition model carries out model training by using the difference of resistivity values of the object to be tested with thyroid tumor in a deep learning method, and the sample data of the object to be tested is recognized by using the model after training is completed to judge whether thyroid abnormality exists in the object to be tested. The thyroid condition can be accurately judged by only inputting sampling data in the process, and the thyroid condition detection method is simple to operate and high in accuracy.
Further, to make the network converge faster, calculations and tests are often repeated to determine the loss function used in the training process
Where y represents sample tag data, y pred Representing the prediction probability.
In step 3, according to the output of the thyroid identification model result output part, whether the sampling data exceeds a preset range is identified.
Specifically, the output of the thyroid recognition model result output unit is 0 or 1,0 represents that the sampling data does not exceed the preset range, that is, there is no thyroid abnormality, and 1 represents that the sampling data exceeds the preset range, that is, there is a thyroid abnormality.
In addition, the invention also provides electronic equipment, which comprises a memory and a processor, wherein the processor is used for executing the thyroid identification model stored in the memory so as to realize the identification method based on the artificial intelligence thyroid sampling data.
The invention provides an artificial intelligence based thyroid sampling data identification method, which can identify whether a patient has thyroid abnormality or not through data acquired by a bioelectrical impedance acquisition platform and can also improve the accuracy of a prediction result. The method has the specific application that the team cooperates with related medical institutions, data provided by the medical institutions are used, a data set is manufactured through the method of the embodiment and a network is trained to obtain a thyroid gland identification model, and in verification under the actual condition, the thyroid tumor identification accuracy of the model can reach more than 92%, so that whether a patient suffers from thyroid tumor or not can be effectively identified, and the thyroid tumor identification model can be used as an effective data support for doctor diagnosis. The identification method not only can be used for identifying thyroid sampling data, but also can be derived to identifying other tissue pathologies such as skin, rectum, cervical and the like.
Any reference to memory, storage, database, or other medium used herein may include non-volatile and/or volatile memory. Suitable nonvolatile memory can include Read Only Memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM), which acts as external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), memory bus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (9)
1. The thyroid sampling data identification method based on artificial intelligence is characterized by comprising the following steps of:
step 1, applying an excitation source to the front part of a neck of an object to be detected by using a bioelectrical impedance tomography front end acquisition platform, and acquiring sampling data of the object to be detected under different excitation conditions;
step 2, preprocessing the sampling data to obtain a data set, and sending the data set into a thyroid identification model, wherein the thyroid identification model comprises a feature extraction part, a feature fusion part and a result output part;
and step 3, identifying whether the sampling data exceeds a preset range or not according to the output of the thyroid identification model result output part.
2. The artificial intelligence based thyroid sampling data identification method of claim 1, wherein: the pretreatment step in the step 2 is as follows: the sampling data comprises the frequency of an excitation source and the resistivity corresponding to the frequency of the excitation source, the sampling data is subjected to type conversion, the sampling data is converted into two-dimensional data in a word embedding mode, and finally the two-dimensional data is converted into three-dimensional data in a dimension mapping mode, so that the data set is formed.
3. The artificial intelligence based thyroid sampling data identification method of claim 1, wherein: the feature extraction part in step 2 comprises 4 convertor blocks, the feature fusion part comprises 3 convolution blocks, and the structure output part comprises 2 full connection layers and 1 softmax layer.
4. The artificial intelligence based thyroid sampling data identification method of claim 3, wherein: each transducer block in the feature extraction part comprises two transducer layers, the first transducer layer is a global maximum pooling layer added after the first normalized sublayer of the original transducer layer, and no feature addition operation exists in the transducer layer; the second transducer layer maintains the original transducer layer structure unchanged.
5. The artificial intelligence based thyroid sampling data identification method of claim 4, wherein: the specific operation of the transducer block is as follows:
after tensor is input into a transitor block, entering a first transitor layer, firstly passing through a multi-head attention machine sublayer and a normalization sublayer, outputting an attention tensor, then carrying out global maximization pooling to obtain a first dimension-unchanged weight tensor, and multiplying the attention tensor and the weight tensor in a first dimension mode with the two dimensions of 1 to obtain an attention sample dimension tensor; then performing random inactivation operation on the neurons to obtain output tensors of the first part of the first transducer layer;
combining the last two dimensions of the output tensor of the first part of the first transducer layer into 1 dimension, and performing two full join operations: the first full-connection operation is to change the number of neurons in the last dimension into 1/16 of the original number, then to perform the second full-connection operation, to restore the number of neurons in the last dimension back to the original number, and to restore the tensor dimension back to the dimension before the first full-connection operation;
then performing random inactivation operation on the neurons to obtain output tensor of the first transducer layer;
the input of the second transducer layer is the output tensor of the first transducer layer, the input of the second transducer layer is firstly passed through a multi-head attention machine sublayer and a normalization sublayer, then the random inactivation of neurons is carried out, and the obtained tensor is added with the output tensor of the first transducer layer to obtain the output tensor of the first part of the second transducer layer;
combining the last two dimensions of the output tensor of the first part of the second transducer layer into 1 dimension, and performing two full join operations: the first full-connection operation changes the number of neurons in the last dimension into 1/16 of the original number, then the second full-connection operation is carried out, the number of neurons in the last dimension is restored to the original number, and the tensor dimension is restored to the dimension before the first full-connection operation;
then outputting a tensor during the random inactivation operation of the neuron; this tensor is added to the output tensor of the first part of the second transducer layer to obtain the final output tensor of the transducer block.
6. The artificial intelligence based thyroid sampling data identification method of claim 3, wherein: the specific processing procedure of the convolution block in the feature fusion part is as follows:
after tensor is input into a convolution block, 3×3 convolution is performed to output a first tensor, the dimension of the first tensor is unchanged, and then 1×1 max pooling operation is performed to obtain a second tensor with the first dimension of 1 and the other two dimensions unchanged; multiplying the second tensor by the first tensor output by the 3×3 convolution to obtain a feature tensor with more focused attributes and word embedding dimensions, splicing the feature tensor with the corresponding tensor output by the transform block, performing 1×1 convolution on the spliced tensor, and restoring the first dimension of the spliced tensor to the dimension of the input convolution block without changing the other two dimensions, thereby obtaining the output result of the convolution block.
7. The artificial intelligence based thyroid sampling data identification method of claim 6, wherein: the specific operation of the thyroid recognition model is as follows:
inputting the data set subjected to dimension mapping into a thyroid identification model, and sequentially passing through a feature extraction part, a feature fusion part and a result output part;
firstly, the data set is sent to a first transducer block in a feature extraction part to obtain an output tensor, then the input of a second transducer block is the output tensor of the first transducer block, the output tensor of the second transducer block is obtained through the second transducer block, and then the 3 rd and 4 th transducer blocks perform the same operation to obtain corresponding output tensors, wherein the dimension of the output tensor of each transducer block is always the same as that of the input tensor;
then, entering a feature fusion part, wherein the input of the 1 st convolution block is the output tensor of the 4 th convolution block, when the 1 st convolution block feature tensor is obtained, splicing the feature tensor with the output tensor of the 3 rd convolution block, and then restoring the first dimension of the spliced tensor to the dimension size when the 1 st convolution block is input through the 1X 1 convolution of the last layer in the 1 st convolution block to obtain the output tensor of the 1 st convolution block; the input of the 2 nd convolution block is the output tensor of the 1 st convolution block, when the characteristic tensor of the 2 nd convolution block is obtained, the characteristic tensor is spliced with the output tensor of the 2 nd convolution block, and then the first dimension of the spliced tensor is restored to the dimension size when the 2 nd convolution block is input through the 1X 1 convolution of the last layer in the 2 nd convolution block, so that the output tensor of the 2 nd convolution block is obtained; the input of the 3 rd convolution block is the output tensor of the 2 nd convolution block, when the characteristic tensor of the 3 rd convolution block is obtained, the characteristic tensor is spliced with the output tensor of the 1 st convolution block, then the first dimension of the spliced tensor is restored to the dimension size when the 3 rd convolution block is input through the 1X 1 convolution of the last layer in the 3 rd convolution block, and finally the output tensor of the characteristic fusion part is obtained;
finally, the input of the result output part is the output tensor of the feature fusion part, wherein the last two-dimensional data of the input tensor of the result output part is combined firstly, and then the number of the last one-dimensional neurons is adjusted by using the first full-connection layer, namely, the number of the output neurons is 1/4 of the number of the input neurons; and then using a second full connection layer to adjust the output neurons to the final classification number, and finally calling a softmax function to adjust the output to the probability distribution.
8. The artificial intelligence based thyroid sampling data identification method of claim 1, wherein: the thyroid recognition model has the following loss functions in the training process:
wherein y represents sample tag data, and the tag data y is obtained in the step 1 and indicates whether thyroid abnormality exists in the object to be detected corresponding to the group of index data; y is pred Representing the prediction probability.
9. An electronic device comprising a memory and a processor configured to execute a thyroid identification model stored in the memory to implement an artificial intelligence based thyroid sampling data identification method as claimed in any one of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211101689.8A CN116186574B (en) | 2022-09-09 | 2022-09-09 | Thyroid sampling data identification method based on artificial intelligence |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211101689.8A CN116186574B (en) | 2022-09-09 | 2022-09-09 | Thyroid sampling data identification method based on artificial intelligence |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116186574A true CN116186574A (en) | 2023-05-30 |
| CN116186574B CN116186574B (en) | 2023-12-12 |
Family
ID=86444896
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211101689.8A Active CN116186574B (en) | 2022-09-09 | 2022-09-09 | Thyroid sampling data identification method based on artificial intelligence |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116186574B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105534524A (en) * | 2016-02-05 | 2016-05-04 | 思澜科技（成都）有限公司 | Device and method for quickly recognizing parathyroid gland in thyroid surgery |
| CN110321962A (en) * | 2019-07-09 | 2019-10-11 | 北京金山数字娱乐科技有限公司 | A kind of data processing method and device |
| CN113553856A (en) * | 2021-06-16 | 2021-10-26 | 吉林大学 | Deep neural network-based dispute focus identification method |
| CN114021628A (en) * | 2021-10-26 | 2022-02-08 | 中国海洋大学 | Transformer model-based manufacturing industry multi-source heterogeneous data fusion method |
| CN114169330A (en) * | 2021-11-24 | 2022-03-11 | 匀熵教育科技(无锡)有限公司 | Chinese named entity identification method fusing time sequence convolution and Transformer encoder |
-
2022
- 2022-09-09 CN CN202211101689.8A patent/CN116186574B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105534524A (en) * | 2016-02-05 | 2016-05-04 | 思澜科技（成都）有限公司 | Device and method for quickly recognizing parathyroid gland in thyroid surgery |
| CN110321962A (en) * | 2019-07-09 | 2019-10-11 | 北京金山数字娱乐科技有限公司 | A kind of data processing method and device |
| CN113553856A (en) * | 2021-06-16 | 2021-10-26 | 吉林大学 | Deep neural network-based dispute focus identification method |
| CN114021628A (en) * | 2021-10-26 | 2022-02-08 | 中国海洋大学 | Transformer model-based manufacturing industry multi-source heterogeneous data fusion method |
| CN114169330A (en) * | 2021-11-24 | 2022-03-11 | 匀熵教育科技(无锡)有限公司 | Chinese named entity identification method fusing time sequence convolution and Transformer encoder |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116186574B (en) | 2023-12-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111553892B (en) | Lung nodule segmentation calculation method, device and system based on deep learning | |
| CN112102237A (en) | Brain tumor recognition model training method and device based on semi-supervised learning | |
| CN111311703B (en) | Electrical impedance tomography image reconstruction method based on deep learning | |
| CN112070119A (en) | Ultrasonic tangent plane image quality control method and device and computer equipment | |
| CN108090509B (en) | Data length self-adaptive electrocardiogram classification method | |
| CN111598867B (en) | Method, apparatus, and computer-readable storage medium for detecting specific facial syndrome | |
| WO2020162837A2 (en) | Method and system for seizure detection | |
| Zhang et al. | ECG data compression using a neural network model based on multi-objective optimization | |
| CN116563533A (en) | Medical image segmentation method and system based on target position priori information | |
| CN114065825A (en) | Brain magnetic MEG source positioning method based on structural similarity | |
| CN116186574B (en) | Thyroid sampling data identification method based on artificial intelligence | |
| CN113116300A (en) | Physiological signal classification method based on model fusion | |
| CN116050691A (en) | Medical image report evaluation method, device, electronic equipment and storage medium | |
| CN115024816A (en) | Cold circulation microwave tumor ablation system | |
| CN116304781B (en) | Thyroid sampling data identification method based on cyclic neural network | |
| CN115546109B (en) | Thyroid sampling data identification method and device based on machine learning | |
| CN115481681B (en) | Mammary gland sampling data processing method based on artificial intelligence | |
| CN114176616A (en) | Venous thrombosis detection method, electronic device and storage medium | |
| CN116226702B (en) | Thyroid sampling data identification method based on bioelectrical impedance | |
| Chin et al. | CWD2GAN: Generative adversarial network of chronic wound depth detection for predicting chronic wound depth | |
| CN116186575B (en) | Mammary gland sampling data processing method based on machine learning | |
| CN112907551B (en) | Disease evolution method and device based on ultrasonic detection image | |
| CN117476219B (en) | Auxiliary method and auxiliary system for positioning CT (computed tomography) tomographic image based on big data analysis | |
| Akella et al. | Hybrid Edge-Artificial Intelligence Model for Identification and Classification of Brain Tumours from Computed Tomography Scans | |
| KR20220141597A (en) | Method and device generating fat suppression magnetic resonance image using generative adversarial neural network based on the bloch equation |
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 |