AU2020103779A4 - Covid-19 classification recognition method based on ct images of lungs - Google Patents
Covid-19 classification recognition method based on ct images of lungs Download PDFInfo
- Publication number
- AU2020103779A4 AU2020103779A4 AU2020103779A AU2020103779A AU2020103779A4 AU 2020103779 A4 AU2020103779 A4 AU 2020103779A4 AU 2020103779 A AU2020103779 A AU 2020103779A AU 2020103779 A AU2020103779 A AU 2020103779A AU 2020103779 A4 AU2020103779 A4 AU 2020103779A4
- Authority
- AU
- Australia
- Prior art keywords
- images
- conv
- covid
- lungs
- softmax
- 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.)
- Ceased
Links
- 210000004072 lung Anatomy 0.000 title claims abstract description 60
- 238000000034 method Methods 0.000 title claims abstract description 32
- 208000025721 COVID-19 Diseases 0.000 title claims abstract description 30
- 238000012549 training Methods 0.000 claims abstract description 37
- 208000037841 lung tumor Diseases 0.000 claims abstract description 24
- 238000013526 transfer learning Methods 0.000 claims abstract description 14
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 12
- 238000013480 data collection Methods 0.000 claims abstract description 10
- 238000007636 ensemble learning method Methods 0.000 claims abstract description 6
- 230000010354 integration Effects 0.000 claims description 4
- 230000035945 sensitivity Effects 0.000 abstract description 23
- 238000011156 evaluation Methods 0.000 abstract description 18
- 238000011176 pooling Methods 0.000 abstract description 14
- 238000012360 testing method Methods 0.000 abstract description 9
- 208000020816 lung neoplasm Diseases 0.000 abstract description 7
- 206010028980 Neoplasm Diseases 0.000 abstract description 5
- 238000003745 diagnosis Methods 0.000 abstract description 4
- 238000010276 construction Methods 0.000 abstract description 2
- 238000012217 deletion Methods 0.000 abstract description 2
- 230000037430 deletion Effects 0.000 abstract description 2
- 238000002591 computed tomography Methods 0.000 description 46
- 238000004422 calculation algorithm Methods 0.000 description 8
- 238000002474 experimental method Methods 0.000 description 7
- 238000001514 detection method Methods 0.000 description 6
- 238000012935 Averaging Methods 0.000 description 5
- 230000003902 lesion Effects 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 150000007523 nucleic acids Chemical class 0.000 description 3
- 102000039446 nucleic acids Human genes 0.000 description 3
- 108020004707 nucleic acids Proteins 0.000 description 3
- 239000012445 acidic reagent Substances 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 238000013145 classification model Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 125000001475 halogen functional group Chemical group 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 230000003211 malignant effect Effects 0.000 description 2
- 238000000638 solvent extraction Methods 0.000 description 2
- 241001678559 COVID-19 virus Species 0.000 description 1
- 206010011409 Cross infection Diseases 0.000 description 1
- 206010029803 Nosocomial infection Diseases 0.000 description 1
- 241000700605 Viruses Species 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000002146 bilateral effect Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 239000005337 ground glass Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 230000001717 pathogenic effect Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 208000024891 symptom Diseases 0.000 description 1
- 238000005303 weighing Methods 0.000 description 1
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
- A61B6/02—Arrangements for diagnosis sequentially in different planes; Stereoscopic radiation diagnosis
- A61B6/03—Computed tomography [CT]
- A61B6/032—Transmission computed tomography [CT]
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
- A61B6/50—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment specially adapted for specific body parts; specially adapted for specific clinical applications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
- A61B6/52—Devices using data or image processing specially adapted for radiation diagnosis
- A61B6/5211—Devices using data or image processing specially adapted for radiation diagnosis involving processing of medical diagnostic data
- A61B6/5217—Devices using data or image processing specially adapted for radiation diagnosis involving processing of medical diagnostic data extracting a diagnostic or physiological parameter from medical diagnostic data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
- A61B6/52—Devices using data or image processing specially adapted for radiation diagnosis
- A61B6/5258—Devices using data or image processing specially adapted for radiation diagnosis involving detection or reduction of artifacts or noise
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- Theoretical Computer Science (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Heart & Thoracic Surgery (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Veterinary Medicine (AREA)
- Public Health (AREA)
- Artificial Intelligence (AREA)
- Animal Behavior & Ethology (AREA)
- Surgery (AREA)
- High Energy & Nuclear Physics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Optics & Photonics (AREA)
- Pathology (AREA)
- Radiology & Medical Imaging (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Dentistry (AREA)
- Pulmonology (AREA)
- Physiology (AREA)
- Image Analysis (AREA)
Abstract
The present invention discloses a COVID-19 classification recognition
method based on CT images of lungs, which comprises the following steps:
collecting CT images of normal lungs, CT images of lung tumors and CT images
of COVID-infected lungs to obtain three sample subsets respectively, and forming
a sample set; pre-training three convolutional neural networks (CNN) AlexNet,
GoogleNet and ResNet with a transfer learning method to obtain initialization
parameters of the three CNNs respectively; inputting the sample set into the three
pre-trained CNNs respectively to obtain three individual classifiers; and integrating
the three individual classifiers with an ensemble learning method to obtain an
ensemble classifier model. The ensemble model in the present invention has an
overall classification performance superior to that of the individual classifier, has
relatively high evaluation indexes such as specificity and sensitivity, and can better
meet the requirements for rapid recognition of the CT images of
COVID-19-infectedlungs.
Drawings of Description
| Grade-A Public News reports Academic
Data collection tertir hospital' databases journals
in China
Original data r u a a .c v sI
M 2500
2500
2933
Or ma dtanormal lung COVID
CT tumor CT
ra-ges CT Images
No
SWhether noise Image deletion
led Yes
2500 nrmallung
CT images
inclusion in Uniform Normalized
Data dat sarts - 2500lung tumor ajr46
renrocceatasets CT imnaaes namm
size 64*64
2500 COVID CT
images
Data set L.- Normal lung _ Lung Tumor CO
-- --- ---- --- --- ---
Construction of a Training set Testing set
training set and a I 6000 1500
testing sample I _I
spare L - - - - - - - - - - - - - - - - - - - - - - - - - - - I - I
I------------------------- ---------------
Five-fold
crossing Fdd23Fld Fd
Transfer learning Transfer learning
CT-AlexNetNet CT-GoogleNet CT-ResNet18
Input Input Input
ttx11, Conv, 96 7x7, Conv 3x3, pooling
3x3, pooling 3x3, pooling residual
blocksa 3x3, Conv, 64
tat, Conv blocks 3x3, Cony, 64
5x5, Conv, 256 3x3, Conv 2
3x3, pooling 3x3, pooling residual 3x3, Conv, 128
blocks 3x3, Cony, 128
3x3, Cony, 384 tx1 1x1 3x3
t3x3, Conv, 384t Conv Conv pool 2 3x3, Conv, 256
sn5. 33,v 3x3 5x5 tnt residual
3x3, Conv, 384 Cony Conv Conv blocks 3x3, Conv, 256
3x3, pooling
2 3x3, Conv, 512
F4096 residual 3x3, Conv, 512
C,4096 f
oncat blocks
FC,1000 pooling
FC,1000 FC,4096
Softmax Softmax Softmax
Constuctin ofIntegration with a "relative
an ensemble majority voting method"
classifier
Classification Diagnosis result
recognition
Fig. 1
1
Description
Drawings of Description
| Grade-A Public News reports Academic Data collection tertir hospital' databases journals in China
2933 Original data Or r ma dtanormal uaa.cvsI lung 2500 COVID 2500 M CT tumor CT ra-ges CT Images
No SWhether noise Image deletion led Yes 2500 nrmallung CT images inclusion in Uniform Normalized Data dat sarts - 2500lung tumor ajr46 CT imnaaes namm size 64*64 renrocceatasets 2500 COVID CT images
Data set L.- Normal lung _ Lung Tumor CO
-- --- --- ---- --- --- Construction of a Training set Testing set training set and a I 6000 1500 testing sample I _I spare L - - - - - - - - - - - - - - - - - - - - - - - - - - -I - I
I------------------------- --------------- Five-fold crossing Fdd23Fld Fd
Transfer learning Transfer learning
CT-AlexNetNet CT-GoogleNet CT-ResNet18
Input Input Input
ttx11, Conv, 96 7x7, Conv 3x3, pooling 3x3, pooling 3x3, pooling residual blocksa 3x3, Conv, 64 tat, Conv blocks 3x3, Cony, 64 5x5, Conv, 256 3x3, Conv 2 3x3, pooling 3x3, pooling residual 3x3, Conv, 128 blocks 3x3, Cony, 128 3x3, Cony, 384 tx1 1x1 3x3 t3x3, Conv, 384t Conv Conv pool 2 3x3, Conv, 256 sn5. 33,v 3x3 5x5 tnt residual 3x3, Conv, 384 Cony Conv Conv blocks 3x3, Conv, 256 3x3, pooling 2 3x3, Conv, 512 F4096 residual 3x3, Conv, 512 f oncat blocks C,4096 FC,1000 pooling FC,1000 FC,4096 Softmax Softmax Softmax
Constuctin ofIntegration with a "relative an ensemble majority voting method" classifier
Classification Diagnosis result recognition
Fig. 1
Description
COVID-19 CLASSIFICATION RECOGNITION METHOD BASED ON CT IMAGES OF LUNGS
Technical Field
The present invention relates to the technical field of medical image recognition, in particular to a COVID-19 classification recognition method based on CT (Computed Tomography) images of lungs.
Background
Due to the characteristic of strong infectivity of COVID-19, rapid and accurate detection of pathogenic viruses is vital to the subsequent treatment and quarantine. Common detection methods for COVID-19 comprise nucleic acid reagent testing and CT examination. Clinical studies have shown that the nucleic acid reagent testing has a high rate of false negative for patients with suspected symptoms for the first time, and has the problems of high requirements for the detection environment, strict process and long consumed time, thereby being difficult to be widely used. The CT detection of COVID-19 has the features of high sensitivity, low rate of missed diagnosis and high popularity of equipment, and can achieve a good complementary effect on nucleic acid testing. The early lung imaging of patients with COVID-19 was mainly featured by ground-glass opacity (GGO), and had apparent "crazy-paving sign"; the density of the lesions increased apparently with halo sign and reversed halo sign after a few days; with the aggravation of the disease, bilateral lung lesions appeared similar to "white lung"; and the density of lesions was gradually decreased, and the scope of lesions was reduced in the later stage. Therefore, the imaging of lung lesions in patients with COVID-19 can be obtained according to the features of CT images of lungs of the patients with COVID-19. At present, COVID-19 virus seriously threatens human life and health. With
Description
the rapid increase in demand for medical treatment, the existing medical resources and diagnosis and treatment capabilities are insufficient to cope with it. In addition, the increasing density of personnel in hospitals in the core epidemic areas greatly increases the risk of cross-infection. The development of a computer recognition learning model of COVID-19 based on the CT images of lungs is urgently needed. Therefore, how to provide a COVID-19 classification recognition method based on the CT images of lungs is a problem to be urgently solved by those skilled in the art.
Summary In view of this, the present invention provides a COVID-19 classification recognition method based on CT images of lungs, which realizes the rapid recognition of the CT images of COVID-19-infected lung by selecting individual classifiers and finally constructing an ensemble classifier EDL-COVID. To achieve the above purpose, the present invention provides the following technical solution: A COVID-19 classification recognition method based on CT images of lungs comprises the following steps: data collection: collecting CT images of normal lungs, CT images of lung tumors and CT images of COVID-infected lungs to obtain three sample subsets respectively, and forming a sample set; sub-model pre-training: pre-training three convolutional neural networks (CNN) AlexNet, GoogleNet and ResNet with a transfer learning method to obtain initialization parameters of the three CNNs respectively; individual classifier training: inputting the sample set into the three pre-trained CNNs respectively to obtain three individual classifiers; and classifier integration: integrating the three individual classifiers with an ensemble learning method to obtain an ensemble classifier model.
Description
Preferably, the step of data collection further comprises a process of uniformly naming the collected CT images of normal lungs, CT images of lung tumors and CT images of COVID-infected lungs and then normalizing the image size to obtain the three sample subsets. Preferably, the step of data collection further comprises a process of de-noising the collected CT images of normal lungs, CT images of lung tumors and CT images of COVID-infected lungs. Preferably, the ensemble learning method comprises a relative majority voting method. Compared with the prior art, the COVID-19 classification recognition method based on the CT images of lungs designed in the present invention has the advantages that: Three individual classifiers AlexNet-Softmax, GoogleNet-Softmax and ResNet-Softmax are constructed, and the ensemble classifier EDL-COVID is obtained by the ensemble learning method; and the experimental results show that the EDL-COVID model proposed by the present invention has a better overall classification performance than a single individual classifier, a fastest detection speed of 342.92 seconds, a detection accuracy of up to 97% and an integration accuracy of up to 99.05%, also has relatively high evaluation indexes such as specificity and sensitivity, and can better meet the requirements for rapid recognition of the CT images of COVID-19-infected lungs.
Description of Drawings
To more clearly describe the technical solution in the embodiments of the present invention or in the prior art, the drawings required to be used in the description of the embodiments or the prior art will be simply presented below. Apparently, the drawings in the following description are merely the embodiments of the present invention, and for those ordinary skilled in the art, other drawings
Description
can also be obtained according to the provided drawings without contributing creative labor. Fig. 1 is a flow chart of a COVID-19 classification recognition method based on CT images of lungs provided by the present invention; Fig. 2 is a schematic diagram of the collected CT images of lungs provided by an embodiment of the present invention; and Fig. 3 is a comparison diagram of evaluation indexes for lung image recognition under different classification models provided by the embodiment of the present invention.
Detailed Description
The technical solution in the embodiments of the present invention will be clearly and fully described below in combination with the drawings in the embodiments of the present invention. Apparently, the described embodiments are merely part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by those ordinary skilled in the art without contributing creative labor will belong to the protection scope of the present invention. Embodiments of the present invention disclose a COVID-19 classification recognition method based on CT images of lungs. The flow chart is shown in Fig. 1; and the method specifically comprises steps as follows. Data collection: collecting 2500 CT images of original normal lungs and 2500 CT images of lung tumors from a grade-A tertiary hospital in Ningxia; and acquiring a total of 2500 CT images of COVID-19-infected lungs from academic publications, news reports, public databases and other channels, wherein Fig. 2(a) shows the CT images NormalLung of normal lungs, Fig. 2(b) shows the CT images LungTumor of lung tumors, and Fig. 2(c) shows the CT images COVID of COVID-infected lungs.
Description
Sample set division: making the sample size |Sample-Lung| of a lung medical image sample set Sample-Lung equal to 7500; dividing the lung medical image sample set Sample-lung into three sample subsets according to the types (NormalLung, LungTumor and COVID) of medical images: Sample-NormalLung, Sample-LungTumor and Sample-COVID with the sample sizes of Sample-NormalLung=2500, |Sample-LungTumorl=2500, |Sample-COVIDI=2500, i.e., Sample-Lung={Sample-NormalLung Sample-LungTumor Sample-COVID} |Sample-NormalLung|=|Sample-LungTumorl=|Sample-COVIDI=2500; adjusting the size of the CT image: sample-lung=resize(Sample-Lung); constructing a training sample set and a testing sample set with a five-fold crossing method; respectively partitioning each sample subset into 5 uniform parts, each containing 500 samples, by using a partitioning algorithm in the three sample subsets Sample-NormalLung, Sample-LungTumor and Sample-COVID, to obtain a five-fold crossed sample partition set, {Sample-Lung-TrainingSet Sample-Lung-TestingSet}=fiveCross(Sample-Lung). Sub-model pre-training: pre-training networks by transfer learning to obtain individual classifiers; pre-training AlexNet by the transfer learning, and taking parameters in the pre-trained network as initialization parameters of the network, wherein AlexNet-Softmax=TransferLearning(AlexNet,Softmax); pre-training GoogleNet by the transfer learning, and taking parameters in the pre-trained network as initialization parameters of the network, wherein GoogleNet-Softmax=TransferLearning(GoogleNet,Softmax); pre-training ResNet18 by the transfer learning, and taking parameters in the pre-trained network as initialization parameters of the network, wherein ResNet-Softmax=TransferLearning(ResNet18,Softmax).
Description
Individual classifier training: respectively training the AlexNet-Softmax, GoogleNet-Softmax and ResNet-Softmax in the training sample set Sample-Lung-TrainingSet to obtain individual classifiers, i.e., AlexNet-Softmax=Training(AlexNet-Softmax,Sample-Lung-TrainingSet); GoogleNet-Softmax=Training(GoogleNet-Softmax,Sample-Lung-TrainingSet );and ResNet-Softmax=Training(ResNet-Softmax,Sample-Lung-TrainingSet). ResNet-NRC classifier integration: integrating the three individual classifiers with a relative majority voting method, i.e., EDL-COVID=Ensemble(AlexNet-Softmax,GoogleNet-Softmax,ResNet-Soft max). The experimental environments for performing the above steps are given below: Software environment: Windows10 operating system, MatlabR2019a; and Hardware environment: Intel(R)Core(TM)i5-7200U CPU @2.50GHz 2.70GHz, 4.0GB of memory, and 500GB of hard disk. 4.2 Evaluation indexes The evaluation indexes herein comprise accuracy, sensitivity, specificity, F-score and Mathews correlation coefficient (MCC), which are specifically introduced as follows. The accuracy is the most common evaluation index; the higher the accuracy is, the better the classifier performance is. The calculation formula is as follows: Accuracy = TP+TN ' TP+TN+FP+FN
The sensitivity and the specificity are used to evaluate the recognition capabilities of the classifiers for positive cases and negative cases, respectively; the larger the value is, the better the recognition performance is; and the calculation formula is as follows: Sensitivity= TP TP+FN
Description TN Specificity = TN FP
The F-score is a weighted harmonic average of recall and accuracy, for weighing the accuracy and the recall; and the calculation formula is as follows: F= 2xTP 2x TP +FP +FN
MCC is a correlation coefficient for describing the relationship between the actual classification and the predictive classification; the MCC comprehensively considers the true positive (TP), true negative (TN), false positive (FP) and false negative (FN), and is a more balanced index within a value range of [-1, 1]; the closer the value is to 1, the more accurate the prediction of the subject is; and the calculation formula is as follows: TP xTN-FPxFN
M(TP+FP)(P+FN)(TN+FP)(TN+FV)
wherein the TP represents that the subject is actually benign and correctly predicted in the number of samples; the FP represents that the subject is actually malignant but incorrectly predicted in the number of samples; the TN represents that the subject is actually malignant and correctly predicted in the number of samples; and the FN represents the subject is actually benign but incorrectly predicted in the number of samples. Specific data of embodiments will be given below. Experimental data collection: In the present embodiment, a total of 2933 CT images of lungs of the patients with COVID-19 were obtained from public publications, news reports and public databases, respectively. The CT images of lungs of the patients with COVID-19 were obtained from a third-party platform; the images from different platforms were different in size and format, and contained different levels of noise, such as watermarks and marks; in addition, the research directions of various platforms were different, such as research on statistical analysis for COVID-19 cases, research on tracking analysis for conditions of one patient, and research on analysis for patients of different ages and sexes and image feature comparison of different
Description
clinical types, causing different data modes, such as transverse section and coronal section. Therefore, in the present embodiment, the data were preprocessed; the images containing loud noise and coronal section were deleted; the unified format of the image was .JPG; the images are normalized to be 64*64 in size; and finally, a total of 2500 high-quality CT images of the patients with COVID-19 were obtained. Algorithm simulation: In the present embodiment, the five-fold crossing method was used for training; and the final experimental result was obtained by averaging the results every time; i.e., the number of training samples was 2000x3=6000, and the number of test samples was 500x3=1500, the results of five experiments were averaged. The experiments were performed on the CT image data sets of normal lungs, lung tumors and COVID-19 respectively, which were respectively recognized and classified in the three individual classifiers AlexNet-Softmax, GoogleNet-Softmax and ResNet-Softmax, and then were integrated to obtain the classification results; and finally, the algorithm was evaluated with the accuracy, the sensitivity, the specificity, the F-score and the MCC. Experiment 1: Classifier AlexNet-Softmax The experimental contents contained the recognition accuracy and the training time of the individual classifier AlexNet-Softmax in the training and recognition of the CT image data sets of normal lungs, lung tumors and COVID; and the algorithm was evaluated with the sensitivity, the specificity, the F-score and the MCC. The specific classification results were as shown in Table 1; and the index evaluation was as shown in Table 2. Table 1 Classification Results of AlexNet-Softmax
Five-fold Accurac Normal Lung COVID Time lungs tumors(S crossing y (%) tr s True False (S)us True False True False Tu as
Description
Fold1 97.00 473 27 487 13 495 5 342.92 Fold2 98.47 495 5 484 16 498 2 383.89 Fold3 98.07 488 12 486 14 497 3 350.25 Fold4 98.93 494 6 493 7 497 3 347.70 Fold5 98.33 498 2 480 20 497 3 347.60 Results 98.16 2448 52 2430 70 2484 16 354.47
Table 2 Classification Evaluation Indexes of AlexNet-Softmax Five-fold crossing SEN(%) SPE(%) F(%) MCC(%)
Fold1 97.33 98.4 96.09 94.13 Fold 2 98.47 99.6 97.74 96.62 Fold 3 97.67 99.0 96.59 94.88
Fold 4 98.93 99.0 98.12 97.17 Fold 5 99.07 99.4 97.55 96.32 A-AVE 98.16 99.36 97.3 95.95
Table 1 showed the experimental accuracy after the five-fold crossing and the final averaging; the average classification accuracy of AlexNet-Softmax was 98.16%; and the running time was 354.47 seconds, i.e., about 6 minutes. Table 2 showed the classification evaluation indexes; the sensitivity (SEN), the specificity (SPE), the F-score and the MCC were 98.16%, 99.36%, 97.3 and 95.95, respectively, which indicated that the model had excellent effects in the recognition of positive classes and negative classes and the description of the correlation between the actual classification and the predictive classification. Experiment 2: Classifier GoogLeNet-Softmax The experimental contents contained the recognition accuracy and the training time of the individual classifier GoogLeNet-Softmax in the training and recognition of the CT image data sets of normal lungs, lung tumors and COVID; and the algorithm was evaluated with the sensitivity, the specificity, the F-score
Description
and the MCC. The specific classification results were as shown in Table 3; and the index evaluation was as shown in Table 4. Table 3 Classification Results of GoogleNet-Softmax Five-fold Accura Normal lungs Lung tumors COVID Time crossing cy True False True False True False Fold1 97.33 471 29 497 3 492 8 934.31 Fold2 98.47 492 8 487 13 498 2 937.04 Fold3 97.67 488 12 482 18 495 5 930.6 Fold4 98.73 499 1 487 13 495 5 924.04 Fold5 99.07 498 2 488 12 500 0 917.75 Results 98.25 2448 52 2441 59 2480 20 928.74
Table 4 Classification Evaluation Indexes of GoogleNet-Softmax Five-fold SEN(%) SPE(%) F(%) MCC(%) crossing Fold 1 97.33 98.4 96.09 94.13
Fold2 98.47 99.6 97.74 96.62
Fold 3 97.67 99.0 96.59 94.88
Fold4 98.73 99.0 98.12 97.17 Fold 5 99.07 100 98.62 97.94 G-AVE 98.25 99.2 97.43 96.14
Table 3 showed the experimental accuracy after the five-fold crossing and the final averaging; the average classification accuracy of GoogleNet-Softmax was 98.25%; and the running time was 928.74 seconds, i.e., about 15 minutes. Table 4 showed the classification evaluation indexes; the SEN, the SPE, the F-score and the MCC were 98.25, 99.2 97.43 and 96.14, respectively, which indicated that the model had excellent effects in the recognition of positive classes and negative classes and the description of the correlation between the actual classification and the predictive classification.
Description
Experiment 3: Classifier ResNet-Softmax The experimental contents contained the recognition accuracy and the training time of the individual classifier ResNet-Softmax in the training and recognition of the CT image data sets of normal lungs, lung tumors and COVID; and the algorithm was evaluated with the sensitivity, the specificity, the F-score and the MCC. The specific classification results were as shown in Table 5; and the index evaluation was as shown in Table 6. Table 5 Classification Results of ResNet-Softmax Five-fold Accura Normal Lung tumors COVID crossing cy lungs Time True False True False True False Fold1 98.00 481 19 495 5 494 6 998.46 Fold2 98.53 490 10 490 10 498 2 1023.85 Fold3 98.53 495 5 483 17 500 0 985.42 Fold4 99.00 494 6 492 8 499 1 978.73 Fold5 98.73 495 5 492 8 494 6 966.46 Results 98.56 2455 45 2452 48 2485 15 990.584
Table 6 Classification Index Evaluation of ResNet-Softmax Five-fold . SEN(%) SPE(%) F(%) MCC(%) crossing Fold 1 98.0 98.8 97.05 95.57 Fold2 98.53 99.6 97.84 97.76 Fold 3 98.53 100 97.85 96.79 Fold4 99.0 99.8 98.52 97.78 Fold 5 98.73 98.8 98.11 97.17 R-AVE 98.56 99.4 97.87 96.81
Table 5 showed the experimental accuracy after the five-fold crossing and the final averaging; the average classification accuracy of ResNet-Softmax was
Description
98.56%; and the running time was 990.584 seconds, i.e., about 16 minutes. Table 6 showed the classification evaluation indexes; the SEN, the SPE, the F-score and the MCC were 98.56, 99.4, 97.87 and 96.81, respectively, which indicated that the model had excellent effects in the recognition of positive classes and negative classes and the description of the correlation between the actual classification and the predictive classification. In the above three experiments, three different classification models were respectively adopted, i.e., AlexNet-Softmax, GoogleNet-Softmax and ResNet-Softmax. The comparison of Table 1, Table 3 and Table 5 showed that the classification accuracy of ResNet-Softmax is 0.4% higher than that of AlexNet-Softmax. The comparison of Table 2, Table 4 and Table 6 showed that the SEN, the SPE, the F-score and the MCC were increased by 0.4%, 0.4%, 0.04%, 0.57% and 0.86% respectively, and the training time was increased by 603.144 seconds. It was not difficult to see that compared with AlexNet, the residual neural network ResNet had more network layers, richer extracted image features and higher classification accuracy, at the cost of greatly increasing the training time. Experiment 4: Ensemble classifier EDL-COVID The experimental contents contained the recognition accuracy and the training time of the individual classifier EDL-COVID in the training and recognition of the CT image data sets of normal lungs, lung tumors and COVID; and the algorithm was evaluated with the sensitivity, the specificity, the F-score and the MCC. The specific classification results were as shown in Table 7; and the index evaluation was as shown in Table 8. Table 7 Classification Results of EDL-COVID Five-fold Accura Normal lungs Lung tumors COVID Time crossing cy True False True False True False Fold1 98.53 486 14 496 4 496 4 2275.81 Fold2 99.07 496 4 492 8 498 2 2234.81
Description
Fold3 98.93 497 3 489 11 498 2 2266.31 Fold4 99.27 500 0 490 10 499 1 2250.53 Fold5 99.47 500 0 493 7 499 1 2231.85 Results 99.05 2479 21 2460 40 2490 10 2251.86
Table 8 Classification Index Evaluation of EDL-COVID Five-fold SEN(%) SPE(%) F(%) MCC(%) crossing Fold 1 98.53 99.2 97.83 96.74 Fold 2 99.07 99.6 98.61 97.92 Fold 3 98.93 99.6 98.42 97.63 Fold 4 99.27 99.8 98.91 98.37 Fold 5 99.47 99.8 99.2 98.81 R-AVE 99.05 99.6 98.59 97.89
As shown in Fig. 3, in order to more clearly represent the differences of different algorithms on each index, the average values of the five indexes were plotted into a line chart. Table 7 showed the experimental accuracy after the five-fold crossing and the final averaging; the average classification accuracy of EDL-COVID was 99.05%; and the running time was 2251.86 seconds, i.e., about 37 minutes. Table 8 showed the classification evaluation indexes; the SEN, the SPE, the F-score and the MCC were 99.05, 99.6, 98.59 and 97.89, respectively, which indicated that the model had excellent effects in the recognition of positive classes and negative classes and the description of the correlation between the actual classification and the predictive classification. From the above experimental results of view, the comparison of the three individual classifiers indicated that the ResNet had higher accuracy, sensitivity, specificity, F-score and MCC than the other two models, but took the most time among the three models; the AlexNet took the least time among the three models,
Description
with an average duration of 354.475 seconds, but had the lowest accuracy, sensitivity, specificity, F-Score and MCC. The comparison of the ensemble classifier and the single classifier indicated that the classification accuracy of the EDL-COVID model was improved by 0.89%, 0.80% and 0.49% respectively over AlexNet-Softmax, GoogleNet-Softmax and ResNet-18-Softmax, and the training time was increased by 1897.39 seconds, 1323.12 seconds and 1261.28 seconds respectively. It was not difficult to see that the classification accuracy of the EDL-COVID model was better than that of the single classifier in different network models; and the ensemble learning can improve the classification accuracy at the same cost of greatly increasing the training time. Each embodiment in the description is described in a progressive way. The difference of each embodiment from each other is the focus of explanation. The same and similar parts among all of the embodiments can be referred to each other. For a device disclosed by the embodiments, because the device corresponds to a method disclosed by the embodiments, the device is simply described. Refer to the description of the method part for the related part. The above description of the disclosed embodiments enables those skilled in the art to realize or use the present invention. Many modifications to these embodiments will be apparent to those skilled in the art. The general principle defined herein can be realized in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principle and novel features disclosed herein.
Claims (4)
1. A COVID-19 classification recognition method based on CT images of lungs, comprising the following steps: data collection: collecting CT images of normal lungs, CT images of lung tumors and CT images of COVID-infected lungs to obtain three sample subsets respectively, and forming a sample set; sub-model pre-training: pre-training three convolutional neural networks (CNN) AlexNet, GoogleNet and ResNet with a transfer learning method to obtain initialization parameters of the three CNNs respectively; individual classifier training: inputting the sample set into the three pre-trained CNNs respectively to obtain three individual classifiers; and classifier integration: integrating the three individual classifiers with an ensemble learning method to obtain an ensemble classifier model.
2. The COVID-19 classification recognition method based on CT images of lungs according to claim 1, wherein the step of data collection further comprises a process of uniformly naming the collected CT images of normal lungs, CT images of lung tumors and CT images of COVID-infected lungs and then normalizing the image size to obtain the three sample subsets.
3. The COVID-19 classification recognition method based on CT images of lungs according to claim 1, wherein the step of data collection further comprises a process of de-noising the collected CT images of normal lungs, CT images of lung tumors and CT images of COVID-infected lungs.
4. The COVID-19 classification recognition method based on CT images of lungs according to claim 1, wherein the ensemble learning method comprises a relative majority voting method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| AU2020103779A AU2020103779A4 (en) | 2020-11-30 | 2020-11-30 | Covid-19 classification recognition method based on ct images of lungs |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| AU2020103779A AU2020103779A4 (en) | 2020-11-30 | 2020-11-30 | Covid-19 classification recognition method based on ct images of lungs |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| AU2020103779A4 true AU2020103779A4 (en) | 2021-02-11 |
Family
ID=74502341
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| AU2020103779A Ceased AU2020103779A4 (en) | 2020-11-30 | 2020-11-30 | Covid-19 classification recognition method based on ct images of lungs |
Country Status (1)
| Country | Link |
|---|---|
| AU (1) | AU2020103779A4 (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113223723A (en) * | 2021-05-11 | 2021-08-06 | 胡敏雄 | Method for predicting multi-modal kidney tumor kidney protection operation difficulty and complications |
| CN113299392A (en) * | 2021-05-25 | 2021-08-24 | 河南八六三软件股份有限公司 | New coronary pneumonia CT sign identification and rapid diagnosis system |
| CN113537394A (en) * | 2021-04-08 | 2021-10-22 | 中国农业大学 | Method for evaluating freshness of iced pomfret by improving VGG-19 |
-
2020
- 2020-11-30 AU AU2020103779A patent/AU2020103779A4/en not_active Ceased
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113537394A (en) * | 2021-04-08 | 2021-10-22 | 中国农业大学 | Method for evaluating freshness of iced pomfret by improving VGG-19 |
| CN113223723A (en) * | 2021-05-11 | 2021-08-06 | 胡敏雄 | Method for predicting multi-modal kidney tumor kidney protection operation difficulty and complications |
| CN113223723B (en) * | 2021-05-11 | 2023-08-25 | 福建省立医院 | Method for predicting difficulty and complications of kidney-protecting operation of multi-mode kidney tumor |
| CN113299392A (en) * | 2021-05-25 | 2021-08-24 | 河南八六三软件股份有限公司 | New coronary pneumonia CT sign identification and rapid diagnosis system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| AU2020103779A4 (en) | Covid-19 classification recognition method based on ct images of lungs | |
| JP2023164839A (en) | Method for analysis of cough sound using disease signature to diagnose respiratory disease | |
| Mohammed et al. | Review on Nasopharyngeal Carcinoma: Concepts, methods of analysis, segmentation, classification, prediction and impact: A review of the research literature | |
| CN111681219B (en) | New coronavirus infection CT image classification method, system and equipment based on deep learning | |
| Albalawi et al. | Classification of breast cancer mammogram images using convolution neural network | |
| CN111933281B (en) | Disease typing determination system, method, device and storage medium | |
| de Sousa Costa et al. | Classification of malignant and benign lung nodules using taxonomic diversity index and phylogenetic distance | |
| CN113643261B (en) | Lung disease diagnosis method based on frequency attention network | |
| Tobias et al. | CNN-based deep learning model for chest X-ray health classification using tensorflow | |
| Xu et al. | COVID-19 diagnosis using chest CT scans and deep convolutional neural networks evolved by IP-based sine-cosine algorithm | |
| Khanna et al. | Radiologist-level two novel and robust automated computer-aided prediction models for early detection of COVID-19 infection from chest X-ray images | |
| Trivizakis et al. | Advancing COVID‑19 differentiation with a robust preprocessing and integration of multi‑institutional open‑repository computer tomography datasets for deep learning analysis | |
| Mangeri et al. | Chest diseases prediction from X-ray images using CNN models: a study | |
| Chen et al. | Auxiliary diagnosis for COVID-19 with deep transfer learning | |
| Gürsoy et al. | An overview of deep learning techniques for COVID-19 detection: methods, challenges, and future works | |
| Ali et al. | COVID-19 pneumonia level detection using deep learning algorithm and transfer learning | |
| Ghafoor | COVID-19 pneumonia level detection using deep learning algorithm | |
| TUNCER et al. | An automated Covid-19 respiratory sound classification method based on novel local symmetric Euclidean distance pattern and ReliefF iterative MRMR feature selector | |
| Chiwariro et al. | Comparative analysis of deep learning convolutional neural networks based on transfer learning for pneumonia detection | |
| CN111582328A (en) | COVID-19 classification identification method based on lung CT image | |
| Zhou et al. | Stool image analysis for precision health monitoring by smart toilets | |
| Mellal et al. | CNN Models Using Chest X-Ray Images for COVID-19 Detection: A Survey. | |
| Nneji et al. | COVID-19 Identification Using Deep Capsule Network: A Perspective of Super-Resolution CNN on Low-Quality CXR Images | |
| TWI818558B (en) | System and method for pathological voice recognition and computer-readable storage medium | |
| He et al. | Intestinal polyp recognition based on salient codebook locality-constrained linear coding with annular spatial pyramid matching |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| FGI | Letters patent sealed or granted (innovation patent) | ||
| MK22 | Patent ceased section 143a(d), or expired - non payment of renewal fee or expiry |