CN115861292A - Pulmonary tuberculosis infectivity discrimination method based on CT image two-dimensional projection and deep learning - Google Patents
Pulmonary tuberculosis infectivity discrimination method based on CT image two-dimensional projection and deep learning Download PDFInfo
- Publication number
- CN115861292A CN115861292A CN202310023737.4A CN202310023737A CN115861292A CN 115861292 A CN115861292 A CN 115861292A CN 202310023737 A CN202310023737 A CN 202310023737A CN 115861292 A CN115861292 A CN 115861292A
- Authority
- CN
- China
- Prior art keywords
- image
- lung
- dimensional
- projection
- infectivity
- 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.)
- Pending
Links
Images
Abstract
The invention relates to the field related to medical intelligent analysis, discloses a method for judging infectivity of tuberculosis based on two-dimensional projection of a CT image and deep learning, and provides a method for analyzing infectivity of a tuberculosis CT image based on two-dimensional projection, wherein a three-dimensional model is not needed in the method, and the sample size can be greatly increased, so that the over-fitting problem of the deep learning model is greatly relieved; a double-flow two-dimensional ResNet model is provided, the model adopts a double-backbone network design, a residual error module is used for extracting features, and the model is excellent in performance on a two-dimensional projection tuberculosis CT image infectivity judging task; the projection in the Z, X and Y directions is adopted, and the mean value and the standard deviation are used as the projection attributes, and the experimental result shows that the method is effective; the mean value of the prediction probabilities in the three directions is used in the test stage, so that the final prediction result of the model is more robust.
Description
Technical Field
The invention relates to the field related to medical intelligent analysis, in particular to a pulmonary tuberculosis infectivity distinguishing method based on CT image two-dimensional projection and deep learning.
Background
Tuberculosis is a chronic infectious disease mainly caused by mycobacterium tuberculosis; the world health organization estimates that about 150 million people die of tuberculosis in 2018, a mortality rate higher than that caused by any other single infectious pathogen. The spread of mycobacterium tuberculosis (including coughing, sneezing, talking, singing, and even deep breathing) primarily through the respiratory tract, preventing its spread is a difficult long-term task, especially when drug-resistant mycobacterium tuberculosis is present; the world health organization recommends the diversion of patients with tuberculosis symptoms or diagnosed as tuberculosis and respiratory isolation of patients deemed or proven to be contagious to reduce the spread of mycobacterium tuberculosis among healthcare workers, healthcare workers in medical institutions, or other high risk environments; therefore, the development of new tuberculosis prevention and management tools is urgently needed; the current methods clinically used for detecting tubercle bacillus include sputum smear examination and sputum bacteria culture examination; sputum smear examination is fast, but repeatability is poor, single detection rate is low, and repeated examination is usually needed; the sputum bacteria culture inspection takes a long time, usually 2-6 weeks, and the cost is expensive; sputum smear examination and sputum bacteria culture examination both depend on the quality of the sputum sample; in addition, there are other methods for assessing the infectivity of tuberculosis patients, such as mathematical models for assessing the mode of transmission of tuberculosis populations; and a special molecular biological method for measuring the number of live mycobacterium tuberculosis in the air exhaled by a single tuberculosis patient; however, these methods are technically demanding, require long training periods and are costly.
In recent years, deep learning methods are widely used for diagnosis of various diseases; the application of the compound in the pulmonary tuberculosis is mainly combined with lung images to identify the pulmonary tuberculosis and the pneumonia, identify drug-resistant tuberculosis and non-drug-resistant tuberculosis, identify infection of mycobacterium tuberculosis and non-mycobacterium tuberculosis, or quickly screen active tuberculosis patients; few studies have focused on rapidly detecting and assessing the level of infectivity in tuberculosis patients to classify them; and the existing three-dimensional deep learning model usually has huge parameters, and the acquired clinical data for CT image diagnosis and analysis of tuberculosis bacteria elimination capability are less, so that overfitting usually occurs in model training.
Disclosure of Invention
The invention aims to provide a method for judging the infectivity of tuberculosis based on two-dimensional projection of a CT image and deep learning, so as to solve the problems in the background technology.
In order to achieve the purpose, the invention provides the following technical scheme:
the method for judging the infectivity of the tuberculosis based on the two-dimensional projection of the CT image and the deep learning comprises the following steps:
preprocessing training data, acquiring CT image data of a historical tuberculosis patient and a detection result of a tuberculosis sputum smear, and judging the infectivity of the patient based on the detection result for dividing the CT image data;
performing cutting projection on a lung image, performing lung segmentation processing on the CT image data through an open source model to obtain a three-dimensional segmentation result, performing projection on the left lung and the right lung in the direction of a spatial coordinate axis respectively based on the three-dimensional segmentation result, and generating a projection image, wherein the three-dimensional segmentation result represents a CT image only keeping the lung;
processing the projection image through a two-dimensional ResNet neural network to generate projection image characteristics, performing maximum value operation fusion on a plurality of characteristic vectors based on the projection image characteristics, and obtaining prediction probability through a full connection layer and a Sigmoid activation function;
performing secondary classification on the data set, performing secondary classification on a pre-divided test data set through a neural network processing step of a sample, drawing an ROC curve according to a probability value output by the test set, calculating RO-AUC as an evaluation index of a secondary classification model, selecting a threshold value when rated specificity is selected as a classification threshold value based on the ROC curve, and judging infectivity by using the classification threshold value;
judging the infectivity of the patient, obtaining a CT image of the lung of the tuberculosis patient to be diagnosed, obtaining a plurality of prediction probabilities through the steps of cutting projection of the lung image and neural network processing of a sample, averaging the prediction probabilities to obtain the final prediction probability of the patient, judging the prediction probability based on a classification threshold value in two classifications of the data set, and obtaining the infectivity of the patient.
As a further scheme of the invention: the training data preprocessing step further comprises:
and carrying out data format conversion on original CT image data of a patient, and converting DICOM format data into NII format data so as to realize data desensitization, wherein the division of the CT image data comprises an infectious strong group and an infectious weak group.
As a further scheme of the invention: before the step of performing the cutting projection of the lung image, the method further comprises the following steps:
and dividing the data into a training data set and a test data set based on the class proportion of the infectious strong group and the infectious weak group, wherein the training data set and the test data set are independent and do not interfere with each other, and the training data set and the test training set respectively account for eighty percent and twenty percent.
As a further scheme of the invention: the step of performing lung segmentation processing on the CT image data through the open source model to obtain a three-dimensional segmentation result specifically includes:
segmenting the lung through an open source model lungmask, multiplying the lung segmentation result by the original CT image to obtain a CT image only retaining the lung, namely a three-dimensional segmentation result, normalizing the CT value, wherein the range of the intercepted CT value is-1000 to 400, and linear interpolation is carried out to 0 to 1.
As a further scheme of the invention: the step of projecting the left and right lungs in the spatial coordinate axis direction based on the three-dimensional segmentation result further comprises a pre-step of:
and judging the three-dimensional segmentation result, if the left lung and the right lung of the segmentation result are not adhered, separating the two regions through a connected domain, judging the left lung and the right lung through the centroid position of the connected domain, if the left lung and the right lung are adhered, directly segmenting the left lung and the right lung along the central line of the cross section of the segmentation result, and when a projection is generated, projecting to remove the pixel point mean value and the standard deviation in the corresponding direction.
As a further scheme of the invention: the two-dimensional ResNet model comprises two main networks, wherein the two main networks are respectively used for extracting projection image characteristics of the left lung and the right lung and are fused through a maximum value operation, in the training of the two-dimensional ResNet model with double flows, the increase of training sample size and the improvement of model generalization energy are carried out through an online data amplification technology, the data amplification technology comprises random overturning, random scaling, random translation, random rotation and random noise, and a loss function of model training is a cross entropy loss function:
wherein the content of the first and second substances,
represents the number of samples, based on the number of samples>
Represents a sample +>
Based on the classification tag in (4), is selected>
Representing model prediction samples>
The loss function causes the model to predict a strongly contagious sample as 1 and a weakly contagious sample as 0 for the probability of strong contagious.
Compared with the prior art, the invention has the beneficial effects that: the method for analyzing the infectivity of the tuberculosis CT image based on the two-dimensional projection does not need to use a three-dimensional model, and can greatly improve the sample size, thereby greatly relieving the over-fitting problem of a deep learning model; a double-flow two-dimensional ResNet model is provided, the model adopts a double-backbone network design, a residual error module is used for extracting features, and the model is excellent in performance on a two-dimensional projection tuberculosis CT image infectivity judging task; the projection in the Z, X and Y directions is adopted, and the mean value and the standard deviation are used as the projection attributes, and the experimental result shows that the method is effective; the mean value of the prediction probabilities in the three directions is used in the test stage, so that the final prediction result of the model is more robust.
Drawings
Fig. 1 is a flow chart of a method for judging infectivity of tuberculosis based on two-dimensional projection of CT image and deep learning.
Fig. 2 is a flow chart of a two-dimensional projection method and deep learning prediction in a method for judging infectivity of tuberculosis based on two-dimensional projection of CT images and deep learning.
Fig. 3 is an example of CT images of patients with weak infectivity and strong infectivity in the method for distinguishing infectivity of tuberculosis based on two-dimensional projection of CT images and deep learning.
FIG. 4 is a schematic diagram of a CT image lung segmentation result in a method for judging infectivity of pulmonary tuberculosis based on two-dimensional projection of a CT image and deep learning.
Fig. 5 is a schematic diagram of a two-dimensional projection image in the method for judging infectivity of tuberculosis based on two-dimensional projection of CT image and deep learning.
Fig. 6 is a two-dimensional ResNet network structure of double flows in the method for judging infectivity of tuberculosis based on CT image two-dimensional projection and deep learning.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The following detailed description of specific embodiments of the present invention is provided in connection with specific embodiments.
As shown in fig. 1 to 6, a method for diagnosing infectivity of tuberculosis based on two-dimensional projection and deep learning of CT images according to an embodiment of the present invention is mainly aimed at analyzing three-dimensional CT images based on a two-dimensional projection method to alleviate overfitting phenomenon of a three-dimensional deep learning model on an infectivity diagnosis task, and thus improve the diagnostic performance of the deep learning model; the used projection method is that the mean value and the standard deviation of pixel points are respectively taken from the three directions of Z, X and Y for the left lung and the right lung of a three-dimensional CT image to obtain the two-dimensional projection of the three surfaces, and then a double-flow two-dimensional ResNet model is used for analyzing projection data to realize the diagnosis of the bacteria-removing capability of a patient; the technical problems to be solved by the invention comprise the preprocessing of a three-dimensional CT image, the generation of two-dimensional projection, the training of a double-flow two-dimensional ResNet model and the diagnosis of a tuberculosis patient by using the trained ResNet model; the method comprises the following steps:
s10, preprocessing training data, acquiring CT image data of a historical tuberculosis patient and a detection result of a tuberculosis sputum smear, and judging the infectivity strength of the patient based on the detection result so as to divide the CT image data.
S20, cutting and projecting lung images, performing lung segmentation processing on the CT image data through an open source model to obtain a three-dimensional segmentation result, respectively projecting the left lung and the right lung in the direction of a spatial coordinate axis based on the three-dimensional segmentation result to generate a projection image, wherein the three-dimensional segmentation result represents the CT image only keeping the lungs.
And S30, processing the projected image through a two-dimensional ResNet neural network to generate projected image characteristics, performing maximum value operation fusion on a plurality of characteristic vectors based on the projected image characteristics, and obtaining prediction probability through a full connection layer and a Sigmoid activation function.
S40, performing secondary classification on the data set, performing secondary classification on the pre-divided test data set through a neural network processing step of a sample, drawing an ROC curve according to the probability value output by the test set, calculating RO-AUC as an evaluation index of a secondary classification model, and selecting a threshold value of rated specificity as a classification threshold value based on the ROC curve, wherein the classification threshold value is used for judging infectivity.
S50, judging the infectivity of the patient, obtaining a CT image of the lung of the tuberculosis patient to be diagnosed, obtaining a plurality of prediction probabilities through the steps of cutting projection of the lung image and neural network processing of a sample, averaging the prediction probabilities to obtain the final prediction probability of the patient, and judging the prediction probability based on a classification threshold value in two classifications of the data set to obtain the infectivity of the patient.
Further, the preprocessing step of the training data further includes:
and carrying out data format conversion on original CT image data of a patient, and converting DICOM format data into NII format data so as to realize data desensitization, wherein the division of the CT image data comprises an infectious strong group and an infectious weak group.
Further, before the step of performing the segmentation projection of the lung image, the method further comprises the steps of:
and dividing the data into a training data set and a testing data set based on the class proportion of the infectious strong group to the infectious weak group, wherein the training data set and the testing data set are independent and do not interfere with each other, and the training data set and the testing training set respectively account for eighty percent and twenty percent.
Further, the step of performing lung segmentation processing on the CT image data through an open source model to obtain a three-dimensional segmentation result specifically includes:
segmenting the lung through an open source model lungmask, multiplying the lung segmentation result by the original CT image to obtain a CT image only retaining the lung, namely a three-dimensional segmentation result, normalizing the CT value, wherein the range of the intercepted CT value is-1000 to 400, and linear interpolation is carried out to 0 to 1.
Further, the step of projecting the left and right lungs in the spatial coordinate axis direction based on the three-dimensional segmentation result may further include a pre-step of:
and judging the three-dimensional segmentation result, if the left lung and the right lung of the segmentation result are not adhered, separating the two regions through a connected domain, judging the left lung and the right lung through the centroid position of the connected domain, if the left lung and the right lung are adhered, directly segmenting the left lung and the right lung along the cross section central line of the segmentation result, and when the projection is generated, projecting to remove the pixel point mean value and the standard deviation in the corresponding direction.
Further, the two-dimensional ResNet model includes two backbone networks, the two backbone networks are respectively used for extracting projection image features of the left lung and the right lung, and are fused through a maximum value operation, in the training of the two-dimensional ResNet model with double flow, the increase of training sample size and the improvement of model generalization energy are performed through an online data amplification technology, the data amplification technology includes random inversion, random scaling, random translation, random rotation and random noise, and a loss function of model training is a cross entropy loss function:
wherein, the first and the second end of the pipe are connected with each other,
represents the number of samples, based on the number of samples>
Indicates that a sample is->
Based on the classification tag in (4), is selected>
Representing model prediction samples>
Is the probability of strong infectivity. The loss function causes the model to predict a strongly contagious sample as 1 and a weakly contagious sample as 0.
In the embodiment, CT image data of a patient with tuberculosis through a tuberculosis sputum smear examination are collected, the original CT image data is generally in a DICOM format, and the DICOM format data is converted into NII format data to realize data desensitization; based on the tuberculosis sputum smear examination result of the tuberculosis patient, the specialist divides the CT image into a strong infectious group and a weak infectious group. If the tuberculosis sputum smear examination of the patient with the pulmonary tuberculosis is negative for three times or more within one month, the patient is classified as weak infectivity; patients with tuberculosis who are positive by a tuberculosis sputum smear test are classified as strongly contagious. FIG. 2 shows CT images of patients with weakly and strongly infectious tuberculosis; all data are divided into 80% for training and 20% for testing according to the category proportion, so that the training set and the testing set are mutually independent and mutually exclusive, and the problem of information leakage does not exist. Before analyzing the pulmonary tuberculosis, the segmentation of the lung is necessary, which not only can eliminate the interference of irrelevant tissues, but also can reduce the learning cost of a deep learning model. The lungs were segmented using the open source model lungmask (https:// github. Com/JoHof/lungmask), and the lung segmentation results were multiplied by the original CT image to obtain a CT image with only the lungs preserved, as shown in fig. 3. In order to facilitate deep learning model training, the CT value is normalized, firstly the CT value range is intercepted to be-1000-400, and then linear interpolation is carried out to be 0-1; the left lung and the right lung are needed to be distinguished before projection is generated, according to the three-dimensional lung division result obtained in the step S3, if the left lung and the right lung of the division result are not adhered, the two areas are separated through a connected domain, and then the left lung and the right lung are judged through the centroid position of the connected domain; if the left and right lungs are adhered, the left and right lungs are cut along the center line of the cross section of the division result and divided into left and right lungs according to the orientation. After distinguishing the left lung from the right lung, respectively projecting the left lung and the right lung along three directions of Z, X and Y, taking the mean value and the standard deviation of pixel points in the corresponding directions by projection, and finally zooming the obtained images to 192 multiplied by 192 sizes, as shown in figure 4; the deep learning network used by the invention is a double-flow two-dimensional ResNet model, and the network structure of the deep learning network is shown in FIG. 5. The double-flow two-dimensional ResNet model is provided with two main networks which are respectively used for extracting the projection image characteristics of the left lung and the right lung, the extracted feature vectors of the left lung and the right lung are fused through a maximum value operation, and finally the prediction probability is obtained through a full connection layer and a Sigmoid activation function. Two backbone networks adopt two-dimensional ResNet neural networks, and the structures of the two backbone networks are the same and do not share weight. Taking the right lung backbone network as an example, the input is a right lung projection image (i.e. the mean and standard deviation of the channel-spliced right lung projection; any one of the three directions Z, X, and Y), the image dimension is rapidly reduced by the convolution layer with the kernel size of 7 × 7 and the step size of 2 × 2 in the first layer and the maximum pooling layer in the second layer, and the memory requirement is reduced. The following 16 residual modules are used to extract the depth features of the projected image. A residual module is composed of 2 convolution layers, 2 Batch normalization layers (Batch normalization) and 2 linear rectification functions (ReLU), and input features are directly added to output features through a cross-layer connection structure, so that feature reuse is improved, and the problem of gradient disappearance during training is solved. The backbone network is finally connected with an adaptive maximum pooling layer (adaptive Maxpool) to compress the two-dimensional feature map into a 512-dimensional depth feature vector. Through the process, a 512-dimensional depth feature can be respectively extracted from the left lung projection and the right lung projection, the left lung feature and the right lung feature are fused into a 512-dimensional depth feature by using maximum value operation, and finally, a Full connection layer (Full connection) is used as a classifier; in a double-flow two-dimensional ResNet model training stage, an online data amplification technology is used for increasing the training sample size and improving the generalization capability of the model. The data amplification method used includes random inversion, random scaling, random translation, random rotation, random noise. In addition, projections of the left lung and the right lung in 3 directions are combined randomly to serve as model input, so that 9 groups of different training samples can be generated by using an original three-dimensional CT image sample, and the model training sample amount is directly enlarged by 9 times. The optimizer used was Adam, and the learning rate was set to 0.0001. The loss function of model training is a cross entropy loss function, and the expression is as follows:
wherein the content of the first and second substances,
represents the number of samples, based on the number of samples>
Represents a sample +>
Based on the classification tag in (4), is selected>
Representing model prediction samples>
Is the probability of strong infectivity. The loss boxThe number prompt model predicts a sample with strong infectivity as 1 and a sample with weak infectivity as 0; and inputting the test data set into the model obtained in the S5 for secondary classification, drawing an ROC curve according to the probability value output by the test set, and calculating RO-AUC as an evaluation index of the two-classification model. Selecting a threshold value with specificity of 0.7 on an ROC curve as a classification threshold value, and samples with specificity greater than or equal to the threshold value are positive, namely strong infectivity; samples below this threshold were negative, i.e. weakly infectious; and (3) processing the CT image of the tuberculosis patient to be diagnosed in the steps S3 and S4 to obtain projections in three directions of Z, X and Y, respectively inputting the projection in each direction into a two-dimensional ResNet model of double flows obtained by training in S5 for prediction, so as to obtain 3 prediction probabilities, and averaging the prediction probabilities in the 3 directions to obtain a final prediction result of the tuberculosis patient. Classifying the patient as strongly or weakly infectious according to the threshold determined at S6. />
Those skilled in the art will appreciate that all or part of the flow of the method implementing the above embodiments may be implemented by a computer program instructing associated hardware.
Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.
It will be understood that the present disclosure is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.
Claims (6)
1. The method for judging the infectivity of the tuberculosis based on the two-dimensional projection of the CT image and the deep learning is characterized by comprising the following steps of:
preprocessing training data, acquiring CT image data of a historical tuberculosis patient and a detection result of a tuberculosis sputum smear, and judging the infectivity of the patient based on the detection result for dividing the CT image data;
performing cutting projection on a lung image, performing lung segmentation processing on the CT image data through an open source model to obtain a three-dimensional segmentation result, performing projection on the left lung and the right lung in the direction of a spatial coordinate axis respectively based on the three-dimensional segmentation result, and generating a projection image, wherein the three-dimensional segmentation result represents a CT image only keeping the lung;
processing the projection image through a two-dimensional ResNet neural network to generate projection image characteristics, performing maximum value operation fusion on a plurality of characteristic vectors based on the projection image characteristics, and obtaining prediction probability through a full connection layer and a Sigmoid activation function;
performing secondary classification on the data set, performing secondary classification on the pre-divided test data set through a neural network processing step of a sample, drawing an ROC curve according to a probability value output by the test set, calculating RO-AUC as an evaluation index of a secondary classification model, selecting a threshold value of rated specificity based on the ROC curve as a classification threshold value, and judging infectivity by using the classification threshold value;
judging the infectivity of the patient, obtaining a CT image of the lung of the patient with pulmonary tuberculosis to be diagnosed, obtaining a plurality of prediction probabilities through the cutting projection of the lung image and the neural network processing step of a sample, averaging the prediction probabilities to obtain the final prediction probability of the patient, judging the prediction probability based on a classification threshold value in the two classifications of the data set, and obtaining the infectivity of the patient.
2. The method for judging infectivity of tuberculosis based on two-dimensional projection of CT image and deep learning of claim 1, wherein the pre-processing step of the training data further comprises:
and carrying out data format conversion on original CT image data of a patient, and converting DICOM format data into NII format data so as to realize data desensitization, wherein the division of the CT image data comprises an infectious strong group and an infectious weak group.
3. The method for judging infectivity of pulmonary tuberculosis based on two-dimensional projection of CT image and deep learning of claim 2, wherein before the step of performing the cutting projection of the lung image, the method further comprises the steps of:
and dividing the data into a training data set and a test data set based on the class proportion of the infectious strong group and the infectious weak group, wherein the training data set and the test data set are independent and do not interfere with each other, and the training data set and the test training set respectively account for eighty percent and twenty percent.
4. The method for judging infectivity of tuberculosis based on two-dimensional projection of CT image and deep learning of claim 3, wherein the step of performing segmentation process of lung on the CT image data through open source model to obtain three-dimensional segmentation result includes:
segmenting the lung through an open source model lungmask, multiplying the lung segmentation result by the original CT image to obtain a CT image only retaining the lung, namely a three-dimensional segmentation result, normalizing the CT value, wherein the range of the intercepted CT value is-1000 to 400, and linear interpolation is carried out to 0 to 1.
5. The method for judging infectivity of tuberculosis based on two-dimensional projection of CT image and deep learning as claimed in claim 4, wherein the step of projecting the left and right lungs in the direction of the spatial coordinate axis based on the three-dimensional segmentation result further comprises the steps of:
and judging the three-dimensional segmentation result, if the left lung and the right lung of the segmentation result are not adhered, separating the two regions through a connected domain, judging the left lung and the right lung through the centroid position of the connected domain, if the left lung and the right lung are adhered, directly segmenting the left lung and the right lung along the cross section central line of the segmentation result, and when the projection is generated, projecting to remove the pixel point mean value and the standard deviation in the corresponding direction.
6. The method for judging infectivity of pulmonary tuberculosis based on two-dimensional projection of CT image and deep learning of claim 5, wherein the two-dimensional ResNet model comprises two main networks, the two main networks are used for extracting the projection image characteristics of the left lung and the right lung respectively, and are fused through a maximum operation, in the training of the two-dimensional ResNet model with double flow, the increase of training sample size and the improvement of model generalization ability are performed through an online data augmentation technique, the data augmentation technique comprises random inversion, random scaling, random translation, random rotation and random noise, and the loss function of model training is a cross entropy loss function:
wherein the content of the first and second substances,
represents the number of samples, based on the number of samples>
Indicates that a sample is->
Based on the classification tag in (4), is selected>
Represents a model prediction sample->
Is the probability of strong infectivity. The loss function causes the model to predict a strongly contagious sample as 1 and a weakly contagious sample as 0./>
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310023737.4A CN115861292A (en) | 2023-01-06 | 2023-01-06 | Pulmonary tuberculosis infectivity discrimination method based on CT image two-dimensional projection and deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310023737.4A CN115861292A (en) | 2023-01-06 | 2023-01-06 | Pulmonary tuberculosis infectivity discrimination method based on CT image two-dimensional projection and deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115861292A true CN115861292A (en) | 2023-03-28 |
Family
ID=85657067
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310023737.4A Pending CN115861292A (en) | 2023-01-06 | 2023-01-06 | Pulmonary tuberculosis infectivity discrimination method based on CT image two-dimensional projection and deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115861292A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116433476A (en) * | 2023-06-09 | 2023-07-14 | 有方（合肥）医疗科技有限公司 | CT image processing method and device |
-
2023
- 2023-01-06 CN CN202310023737.4A patent/CN115861292A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116433476A (en) * | 2023-06-09 | 2023-07-14 | 有方（合肥）医疗科技有限公司 | CT image processing method and device |
| CN116433476B (en) * | 2023-06-09 | 2023-09-08 | 有方（合肥）医疗科技有限公司 | CT image processing method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112101451B (en) | Breast cancer tissue pathological type classification method based on generation of antagonism network screening image block | |
| CN111951246B (en) | Multidirectional X-ray chest radiography pneumonia diagnosis method based on deep learning | |
| Dasanayaka et al. | Deep learning methods for screening pulmonary tuberculosis using chest X-rays | |
| CN108549912A (en) | A kind of medical image pulmonary nodule detection method based on machine learning | |
| CN113095382B (en) | Interpretable tuberculosis classification network identification method based on CT image | |
| Mangeri et al. | Chest diseases prediction from X-ray images using CNN models: a study | |
| CN115861292A (en) | Pulmonary tuberculosis infectivity discrimination method based on CT image two-dimensional projection and deep learning | |
| R-Prabha et al. | Design of hybrid deep learning approach for covid-19 infected lung image segmentation | |
| Afshar et al. | Hybrid deep learning model for diagnosis of COVID-19 using CT scans and clinical/demographic data | |
| El-Shafai et al. | Classification framework for COVID-19 diagnosis based on deep cnn models. | |
| CN110782441A (en) | DR image pulmonary tuberculosis intelligent segmentation and detection method based on deep learning | |
| Sufian et al. | Deep learning feature extraction for COVID19 detection algorithm using computerized tomography scan | |
| CN112085742B (en) | NAFLD ultrasonic video diagnosis method based on context attention | |
| Bhansali et al. | CoronaNet: a novel deep learning model for COVID-19 detection in CT scans | |
| Zhang et al. | Multicontext 3D residual CNN for false positive reduction of pulmonary nodule detection | |
| Gupta et al. | Deep learning assisted COVID-19 detection using full CT-scans | |
| Riyadi et al. | Covid-19 Detection Based-On CT-Scan Images Using Inception Deep Learning | |
| Hijazi et al. | Ensemble deep learning for tuberculosis detection | |
| CN111652840B (en) | Turbid screening and classifying device for X-ray chest X-ray image lung | |
| Santos et al. | Detection of Fundus Lesions through a Convolutional Neural Network in Patients with Diabetic Retinopathy | |
| Sahib et al. | Lung Segmentation and Detection of COVID-19 in CT Images based on Deep Learning (CNN) | |
| CN117058467B (en) | Gastrointestinal tract lesion type identification method and system | |
| CN112508958B (en) | Lightweight multi-scale biomedical image segmentation method | |
| Rao et al. | Covid-19 X-Ray Image Detection using ResNet50 and VGG16 in Convolution Neural Network | |
| Hao et al. | Automatic detection of breast nodule in the ultrasound images using CNN |
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 |