Application of Fuzzy Deep Neural Networks for Covid 19 diagnosis through chest Radiographs

Priyanka Yadlapalli; Bhavana D

doi:10.12688/f1000research.126197.1

Home Browse Application of Fuzzy Deep Neural Networks for Covid 19 diagnosis through...

ALL Metrics

-

Views

-

Downloads

Get PDF

Get XML

Export

▬

✚

Research Article

Application of Fuzzy Deep Neural Networks for Covid 19 diagnosis through chest Radiographs

[version 1; peer review: 1 approved with reservations, 1 not approved]

Priyanka Yadlapalli ^1,2, Bhavana D³

PUBLISHED 16 Jan 2023

Author details Author details

¹ Electronics and Communications Engineering, Gokarjurangara Institute of Engineering and Technology, Hyderabad, Telangana, 500090, India
² Research Scholar, KL University, Viyayawada, Andhrapradesh, 522 302, India
³ Electronics and Communications Engineering, KL University, Vijayawada, Andhrapradesh, 522 302, India

Priyanka Yadlapalli
Roles: Conceptualization, Data Curation, Investigation, Methodology, Resources, Supervision, Validation, Writing – Original Draft Preparation, Writing – Review & Editing

Bhavana D
Roles: Supervision, Validation

OPEN PEER REVIEW

REVIEWER STATUS

This article is included in the Artificial Intelligence and Machine Learning gateway.

This article is included in the Emerging Diseases and Outbreaks gateway.

This article is included in the Computational Modelling and Numerical Aspects in Engineering collection.

Abstract

Background: The increasing number of COVID-19 patients around the world and the limited number of detection kits pose a challenge in determining the presence of the disease. Imaging modalities such as X-rays are commonly used because they are readily available and cost-effective. Deep learning has proved to be an excellent tool because of the abundance of online medical images in various medical modalities, such as X-Ray, computerized tomography (CT) Scan, and magnetic resonance imaging (MRI). A large number of medical research projects have been proposed and launched since early 2020 due to the overwhelming use of deep learning techniques in medical imaging.
Methods: We have used fuzzy logic and deep learning to determine if chest X-ray images belong to people who have pneumonia related to COVID-19 and people who have interstitial pneumonias that aren't related to COVID-19.
Results: In comparison to the current literature, the proposed transfer learning approach is more successful. It is possible to classify covid, viral, and bacterial pneumonia or a healthy patient using ResNet 18 Architecture's four-class classifiers. The proposed method achieved a 97% classification accuracy, 96% precision, and 98% recall in the case of COVID-19 detection using chest X-ray images, which demonstrates the importance of deep learning in medical image diagnosis. Furthermore, the results demonstrate that the proposed technique has the maximum sensitivity rate, with 97.1% ratio. Finally, with a 97.47% F1-score rate, the proposed strategy yields the highest value when compared to the others.
Conclusions: DeepLearning techniques and fuzzy features resulted in an improved classification ability, with an accuracy rate of up to 97.7% using ResNet 18, which is a better value when compared to the remaining techniques. Classification of COVID-19 scans and other pneumonia cases have been done successfully by demonstrating the potential for applying such deep learning techniques in the near future.

Keywords

Corona virus pandemic, Fuzzy logic, Deep learning, Residual Networks, Neural network

Corresponding author: Priyanka Yadlapalli

Competing interests: No competing interests were disclosed.

Grant information: The author(s) declared that no grants were involved in supporting this work.

Copyright: © 2023 Yadlapalli P and D B. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

How to cite: Yadlapalli P and D B. Application of Fuzzy Deep Neural Networks for Covid 19 diagnosis through chest Radiographs [version 1; peer review: 1 approved with reservations, 1 not approved]. F1000Research 2023, 12:60 (https://doi.org/10.12688/f1000research.126197.1) First published: 16 Jan 2023, 12:60 (https://doi.org/10.12688/f1000research.126197.1) Latest published: 16 Jan 2023, 12:60 (https://doi.org/10.12688/f1000research.126197.1)

Introduction

Coronavirus disease 2019 (COVID-19) is caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Since May 28 2020, across the globe many people have lost their lives due to COVID-19.¹^,² The majority of patients who had been affected from COVID-19 have been diagnosed with flu like symptoms, and quite a small number of patients developed chronic infections like pneumonia and acute respiratory distress syndrome (ARDS).³ No treatment will be effective until patients have been properly screened for COVID-19. Reverse transcription polymerase chain reaction (RT PCR), is considered as the golden standard screening method for identifying SARS-CoV-2, having a success rate of around 30-60% only.⁴

Over the last decade, medical image analysis has undergone a paradigm shift, owing largely to the phenomenal success of deep learning methods that achieve excellent performance on a variety of tasks. Medical image analysis has developed into a large and active field of research in recent decades as a result of its significant clinical impact and remaining challenges. Conventional radiography has several advantages over other imaging modalities, including a short examination time, high spatial resolution, and low image cost.⁵ In medicine, conventional radiography is one of the most ineffective imaging modalities, due to its wide range of applications for different body parts and diseases. Hence, chest X-rays through computer-aided diagnostic systems have been in great demand for the detection of COVID-19 pneumonia.⁶

Fuzzy logic is an effective technique for tackling uncertainty and imprecision. In addition to providing accurate data, uncertainty intervals allow for double control of the detection process, reducing false negatives and positives and improving COVID-19 patient detection.⁷ A fuzzy model based medical diagnosis system relies on expert knowledge, observation, and experience.⁸

Due to the increase in the number of cases, COVID-19 viruses have evolved and infected humans, making them more widely spread. COVID-19 has been diagnosed using a variety of artificial intelligence (AI) technologies and radiographic image databases in recent years. COVID-19-related lung computerized tomography (CT) images are publicly available for deep learning research.

Literature survey

Table 1 summarises the various methods developed for automated detection of infected cases. Chest X-ray images are trained using different deep learning techniques. These techniques have gained popularity in recent years⁹^,¹⁶ due to their low computational time and high accuracy efficiency.

Table 1. Research works related to COVID-19 and deep learning.

Name of the Author	Data Type	Method Adopted	Image classes	Training Model	Accuracy
Ieracitano et al. (2022)⁹	Chest -X Ray	Fuzzy based Deep Learning Models	Covid 19 vs Healthy and other pneumonia cases	MLP	81%
Alqudah et al. (2020)¹⁰	Chest -X Ray	Deep Learning Models	COVID-19 vs Normal and pneumonia cases	ResNetV2	98%
Cohen et al. (2020)¹¹	Chest -X Ray	Deep Learning Models	Covid 19 vs Normal	Convolutional Neural Networks	94.8%
Apostolopoulos and Bessiana (2020)¹²	Chest -X Ray	Deep Transfer Learning	COVID-19, Pneumonia, Normal	Mobile Net	97.8%
Togacar et al.¹³	Chest -X Ray	Deep Learning Models	COVID-19 vs	SqueezeNet	99.27%
Ozturk et al. (2020)¹⁴	Chest -X Ray	Deep Learning Models	Covid+, Pneumonia, No-findings	DarkCovidNet	87.0%
Wang et al. (2020)¹⁵	Chest -X Ray	Deep Learning Models	Covid 19 vs Normal, pneumonia	CovidNet	93.3%

To summarise, deep learning is favored for the reasons listed below:

a) Effective use of unstructured data.
b) Reduced computational time.
c) Ability to deliver accurate results.

Deep learning has many advantages. For the purposes of comparing the accuracy of their performance in detecting COVID-19 cases, the researchers used a variety of training models including Inception V3, ResNet 50, ResNetV2, MobileNet, VGG19, and others (Table 1).

Ieracitano et al.⁹ proposed a fuzzy enhanced deep learning approach for the detection of COVID-19 with an accuracy of 81%.

Using ResNetV2 architecture, Alqudah has achieved 98% accuracy.¹⁰ Using the VGG19 architecture, Apostolopoulos and Bessiana¹² classified virus cases with pneumonia and normal cases with a 97.8% success rate. SqueezeNet Architecture was developed by Togacar et al¹³ classified COVID-19 with normal cases, resulting in an accuracy percentage of 99.27%. To classify COVID-19 as pneumonia, Ozturk et al.¹⁴ used Dark CovidNet and achieved an accuracy rate of 87.9%.¹⁴ By using the COVID-NET Training Model and classifying COVID-19 into the Normal and Pneumonia categories, Linda Wang et al.¹⁵ has achieved an accuracy of 93.3%.

From the above mentioned methods, we can deduce that the highest accuracy that was obtained was around 96.84%. In order to increase the accuracy levels of the above areas, four different transfer learning techniques have been adopted for this study and their performances has been analysed. CNN-based transfer-learning method provides greater accuracy for detection and classification. Radiographic findings, on the other hand, have so far been ineffective in determining the underlying cause of pneumonia.¹⁷ Machine learning can be used to diagnose pneumonia by analyzing a radiograph which is shown in Figure 1, and it can also be used to distinguish viral and bacterial pneumonia more accurately [31]. The objective of the study is to evaluate the performance of various deep learning architectures and to classify and detect the chest radiographs for detection of Covid 19 along with other pneumonial diseases which is depicted in Figure 2.

Figure 1. Procedure of using deep learning techniques in chest X-rays.

Figure 2. Combined dataset of both coronavirus disease 2019 (COVID-19) and pneumonia X-ray images.

The analysis in this study was done using chest radiograph images from two different sources.¹¹^,¹⁸ To compile an archive of COVID-19 images, Cohen JP used images which are freely available online.¹¹ The images on this site are constantly being updated by researchers from all over the world. In the database, more than 350 X-ray images have been classified as COVID-19. By selecting 397 frontal chest X-ray images from Kaggle,¹⁸ we can avoid the problem of data imbalance as the data we are selecting is random. Table 2 shows the specifics of the data set.

Table 2. Details of the datasets used.

Type of Dataset	Type of disease	Number of Images
Covid Chest X ray data set	COVID-19	350
Chest X ray images	Viral	1493
	Bacterial	2780
	Normal	1583

Figure 3 shows the data distribution has been graphically represented using four different classes. Chest X-rays of healthy people, patients with bacterial pneumonia, viral pneumonia, and patients with COVID-19 are depicted in Figure 4.

Figure 3. Distribution of data for final four class classification.

Figure 4. A) Normal B) bacterial pneumonia C) viral pneumonia D) COVID-19.

The significant contribution of the proposed research is:

• Limited and imbalanced data sets are made available to the general public.
• For this reason, we used multi-operation data augmentation to balance the COVID-19 and normal classes.
• VGG-16, AlexNet, SqueezeNet and ResNet 18 are the four most effective pre-trained deep CNN models that have been thoroughly compared.
• In the end, we have found the best model for designing a more efficient CNN-based solution for the early detection of COVID-19 infection.

The rest of the article is organised as follows. Literature Survey section contains various methods describing current research on COVID-19 virus detection methods using machine learning. It also describes the samples used to validate various CNN models. The ‘Methods in testing’ section discusses the proposed methodology for accurately detecting COVID-19 virus in chest X-ray frames. The ‘Results’ section is where various classification schemes are described and compared to the proposed method. The last section focuses on the conclusion and areas for future research.

Methods in testing

The three stages of the proposed methodology are

a) Preprocessing of images.
b) Augmentation of data.
c) Training of a deep learning CNN.

Figure 5 depicts the feature extraction and classification stages of convolution neural networks.

Figure 5. Convolution neural network architecture.

Image preprocessing

All the changes made to the raw data prior to feeding it to the machine learning or deep learning algorithm are referred to as preprocessing. The resized images are preprocessed in order to increase the efficiency and accuracy of the deep learning models by using ImageNet database.¹⁹ It is a freely accessible data set for computer vision which includes millions of images and thousands of image classes. Transfer learning refers to the training of the network on a new dataset based on an earlier trained network on an ImageNet index.

Data augmentation

The augmented data helps deep learning algorithms to improve their classification accuracy. The results of deep learning techniques can be improved by enhancing existing data rather than collecting new data, because the amount of input data can be increased by improving images of existing data. It also helps to avoid the over fitting problem when training a machine model using Google Colab.

Rotation, translation, scaling, and horizontal flipping of the training set are used to address the issue of data scarcity. There are several techniques that can be applied to each image in training.

Rotation

The image can be rotated in either a clockwise or counterclockwise direction using Google Colab, at an angle ranging from 0 to 15 degrees. The image's proportions are lost when it is rotated. The final image size may also be affected by finer angle rotations.

Image scaling

Images can be scaled in two ways: outwards and inwards. Image size increases when the image is scaled outward, while image size decreases when the image is scaled inward. Samples between 90 and 110% of the image's frame size can be taken at random.

Translation

Image translation by -10% to 10% on the horizontal (X) and vertical (Y) axis. In this method, the entities can be identified anywhere in the image.

Horizontal flip

The image is flipped horizontally from left to right in this operation. 0.5% of the time, the image is flipped horizontally.

Gaussian blur

Using a Gaussian filter with a kernel size of 5x5, one can blur or smooth images by removing high frequency components. Training data includes the blurred images obtained from the Gaussian filter.

Fuzzy enhanced CNN -based transfer learning

Fuzzy logic (FL) can transfer an input space to an output space and is one approach for complex systems.⁵ The results of the study can aid in patient diagnosis and community self-identification, making it simpler for everyone to deal with Covid-19. Fuzzy algorithms tend to be robust and reasoning processes tend to be simple, FL was chosen as the soft computing approach for implementing the proposed COVID-19 diagnosis system. This means that FL uses less computing power and takes less time to develop than more traditional methods. FL is adaptable and is simple to incorporate into machine learning techniques, which makes it an especially useful characteristic for real-time systems like online diagnostic apps.²⁰

An enormous amount of data and a long period of training are necessary. For many deep learning models, the most fundamental presupposition is that both testing and training data should be collected from the same distribution.²¹ Due to its prohibitive costs and data labelling requirements, it cannot be used in practical applications. In CNN, the transfer learning technique is used for training the datasets. The ImageNet database contains a transfer learning network that has been pre-trained, which is used in this study to train on the dataset. The derivative of the activation function is used to build the neural network and leads to the vanishing gradient problem. This problem is alleviated by using this transfer learning method. Transfer learning is depicted in Figure 6, where a large dataset like ImageNet can be used to train a model for a smaller dataset application.

Figure 6. Concept of transfer learning.

Transfer learning has the advantage of already learning basic features like recognizing shapes and edges in an image. The main advantage of the pre-trained model is that it learns basic features from the images in its database²²^,²³ As a further benefit, only the network's final layers need to be trained, which reduces the amount of computation time. In addition to Dense networks, Squeeze Net, VGG, Inception, and Residual Networks, CNN already has a variety of architectures. When it comes to accuracy, computational time, and various other challenges, residual networks have surpassed the rest. Residual networks have proved the best when compared to other learning techniques as they solve the degradation problem by using skip connections to connect shallow to deep layers.²⁴

Convolution layer has been divided into the following further layers:

a) Adaptive ConcatPool2D layer
b) Flattened Layer
c) Blocks of rectified Linear Unit (ReLu)
d) Dropout layer
e) Linear layer
f) Batch Norm1D

In the final linear layer block, we would get exactly the same quantity of outputs as equal to the same quantity of classes in the dataset, which is very important. Pareto's 80/20 rule dictates that 70% of the preprocessed Chest X-ray images are used for training; 20% of the images are used for network testing; and the remaining 10% of the preprocessed images are used for validation. Classification accuracy is improved by using variable learning rates that do not have fixed values. In the Fast Ai library, the learning rate (LR) find function can be used to find the best method for training a network model's learning rate.

VGG 16 Architecture

The VGG architecture comprises of numerous convolutional layers activated by ReLU (rectified linear unit), with a kernel size of 3x3.²⁵ The VGG models VGG-11, VGG-16, and VGG-19 are three versions of the VGG model that are structurally similar. They are made up of three fully connected layers, followed by subsequent convolution and pooling layers. The sole difference between them is the number of convolution layers (11, 16, or 19), as indicated by their titles.

Alex Net Architecture

AlexNet nearly uses 650k neurons and 60 million parameters thus classifying more than 1000 classes. One Softmax layer, three pooling layers, two FLCs, and five convolutional layers (CLs) comprise this network.²⁶ A 227x227 three AlexNet input image must have 96 kernels sized at 11 x11x 3 which is given as the input to the second Convolutional Layer, so that it can produce four number of pixels.

SqueezeNet

In addition to ImageNet, SqueezeNet is a CNN trained on this database. In comparison, SqueezeNet has 50 times fewer parameters than AlexNet, which was trained on over 1 million images during its development.²⁷ Both the expand and squeeze layers are included in this network's structure. Extend layer contains a mix of 1x1 filters and 3x3 convolution filters, which feed into this layer. The suggested fire module, which consists of two sections, the squeeze portion and the expand part, is at the heart of SqueezeNet. A 1x1 convolutional kernel and a 1x1 convolutional layer make up the squeeze component. 1x1 and 3x3 convolutional kernels and convolutional layers, respectively, make up the expand component. The 1x1 and 3x3 feature maps are concatenated in the expanded layer. A comparison of SqueezeNet and AlexNet's accuracy on the ImageNet dataset reveals that they are about equal.

Architecture of ResNet Model (ResNet 18)

Showkat et al.²⁸ has proposed ResNet which is one of the various deep learning models. As the shallower models are very much prone to copy learning parameters and setting the additional layers for identity mapping which in fact is the basic principle behind ResNet. As a result of which the deeper networks have found to be more difficult to optimize Residual models are built to fit into the residual mapping F(x) rather than the underlying desired mapping

H(x) in order to improve deeper models. By superimposing two 3x3 convolutional layers the ResNet architecture can be constructed.

ResNet incorporates the original input into the output feature map after it has been passed through one or more convolutional layers. Here, we'll take a closer look at Figure 7. The result is referred to as Relu, and it is either the first data or the data from the block before it. As long as the output and input sizes don't match, a constrained 1x1 convolution is used to reproject the filters, and a stride value of 2 is used to reduce the output size by half as much as possible through pooling.

Figure 7. Structure of Residual Block.

In order to establish a distinction between Covid-19 and other viral pneumonias, ResNet18 has more layered blocks. With ResNet18, you can train on over 11 million different variables. The ResNet18 model distinguishes between normal cases in addition to COVID-19 cases with bacterial and viral pneumonia. As seen in Figure 8, a lot of factors are required to generalise for the various virus classifications.

Figure 8. Schematic of transfer learning with ResNet 18.

For the identification of COVID-19 CT images, ResNet is a commonly used and preferred deep learning network. The accuracy of the model prediction decreases as the depth of the network goes beyond a specific number in ResNet and other DL networks,²⁹ therefore the model depth must be carefully adjusted. This difficulty was overcome by transferring features from lower layers to higher layers and creating an identity mapping between the network's higher and lower layers. The block use multiplier is the primary difference between ResNet-18 and ResNet-34.

Learning rate assessment

The learning rate is a critical factor for deep neural networks when it comes to the process of learning. By using equation.1, the formula outputs the most recent weights at the end of each epoch.

(1)

θ_{j + 1} = θ_{j} - α \frac{\partial j (θ)}{\partial θ_{j}}

j –It's the number of epochs that have been run.

j( $ϴ$ ) – loss function

α - learning rate

θ_j+1 updated new weight

$\frac{\partial J (ϴ)}{\partial ϴ_{i}}$ gradient weight of θ_j

In some cases, choosing the optimal learning rate can be difficult. Converging of weights too quickly results in sub-optimal weight loss and an unstable training process when the rate of learning is high. When the network-based model is constantly tweaked with different learning rates, the process requires manual intervention. Having a low learning rate slows down the training process and causes the network to be delayed. This study describes an experiment where the optimal learning rate may be started by cyclically changing the learner's speed in line with their boundaries. Remaining networks are trained using X-ray images of patient’s lungs with bacterial and viral infections and healthy lungs to distinguish between them.

Testing methods

Although models of deep learning are regarded as black box techniques, most researchers and practitioners are unable to understand where the network is drawing its input data activation from, or how the network has reached its final destination; this is the reality in practice. For this purpose, a debugging technique called Grad-CAM (Gradient-Weighted Class Activation Mapping) algorithm,³⁰ which is a visual class activation map is commonly used in deep neural networks. To further aid in the study and understanding of deep learning models, Grad-CAM provides an evaluation method. In the final convolutional layer, Grad-CAM uses the inclinations of any objective idea to generate an image-based coarse limitation map that highlights the image's most critical regions for predicting the idea. If a chest radiograph assessment is required, the network can approve the location using Grad-CAM. The slope of the data flowing into the network's final convolutional layer is examined by Graduate CAM. This tool generates a heat map based on a particular class label.

Convolutional networks look for specific features in an image, which are represented by a heat map in the experimental analysis section. Visualization in Figure 9 shows the areas of an image where COVID-19 is believed to be presented.

Figure 9. ResNet18 network model predicted class activation heat maps.

The VGG 16, Alex Net, SqueezeNet, and ResNet 18 models were trained and assessed on a Tesla K80 GPU (graphics processing unit) with a memory capacity of approximately 12GB offered by the Google Colab.

Metrics for evaluating performance

Classifier performance on test data is frequently evaluated using the confusion matrix. In this table, you'll find all of the test data's actual and expected instances of the classes that are represented. F1-score, specificity and sensitivity, precision and classification accuracy can all be calculated using it. To calculate these parameters from the confusion matrix, a few terms need to be specified, such as

1. True Positive for correctly predicting the positive class (TP).
2. To accurately predict the negative class, use (b) True Negative (TN).
3. True Negative (TN) if the positive class is estimated incorrectly.
4. Negative class prediction error causes a False Positive (FP).

Classification problems can be summarized in a confusion matrix, which includes both the predicted and the actual labels. In addition to reporting on the errors, the classifier provides information on the types of errors it has made. Fig 10 shows a diagram of the confusion matrix.

Figure 10. Confusion matrix of 4 classes (normal, viral pneumonia, bacterial pneumonia).

Confusion matrix

Figure 10 shows the confusion matrix, with true positive and true negative values, as well as false positive and false negative values. Due to an uneven distribution of images in the test set, with different image instances for each of the testing classes. We can use an AUC (Area Under the Curve) score for binary classification and a weighted F1 score for four-class classification to evaluate the model. Four -class classification requires that precision (P) and recall (R) for each class be calculated independently and averaged to calculate the F1 score. To evaluate models on the imbalance dataset, the F1 score is a weighted harmonic mean of P and R. The values listed can be calculated using the following equations:

Accuracy:

Accuracy is defined as the ratio of correct predictions (COVID-19 cases) from the overall number of cases for each type of case is mathematically conveyed as:

(2)

Accuracy = \frac{(TP + TN)}{(TP + FP + FN + TN)}

TP = No. Correct Positive Case Predictions.

TN = No. of correct negative case predictions.

FP = Percentage of positive cases correctly predicted.

FN = No. of in correct predictions of negative cases.

Precision:

The percentage of positive cases that were correctly predicted. A lower false positive rate is directly related to greater precision

(3)

Precision = \frac{TP}{(TP + FP) (TP + FP)}

Recall:

The percentage of correctly predicted positive class observations that actually occurred is nothing but recall.

(4)

Recall = \frac{TP}{(TP + FN)}

Specificity:

The percentage of predicted negative events that actually occurred is called specificity.

(5)

Specificity = \frac{TN}{(FP + TN)}

F- Measure:

The F-measure is calculated when there are a large number of true negative observations in the class distribution. As a result, it finds a good balance between the two.

(6)

F Measure = 2 * \frac{(Precision * recall)}{(Precision + recall)}

Equations 2-6 gives us accuracy, recall, specificity and F-measure formulas are used for calculating the performance evaluation of various deep learning techniques.

Results

Table 3 shows the Comparison of different deep learning techniques for the four different types of viruses. When multiple models from the figure are compared, deeper networks outperform shallower networks. When it comes to accuracy, recall and specificity, we can conclude that ResNet18's design is far superior to other architectures as it is highlighted in green in Table 3. It is possible to develop prototypes that can automatically classify results into four categories (COVID-19, viral, bacterial pneumonia and normal cases) using the proposed method, which has the highest accuracy values among recent studies, according to our knowledge.

Table 3. Comparison of different deep learning techniques for the four different types of virus.

Type	Models	Accuracy	Recall	Specificity	Precision	F1 Scores
Normal	VGG 16	0.912	0.92	0.932	0.91	0.935
	AlexNet	0.945	0.953	0.926	0.931	0.943
	SqueezeNet	0.961	0.94	0.98	0.985	0.961
	Res Net 18	0.964	0.97	0.95	0.954	0.965
Viral Pneumonia	VGG 16	0.87	0.845	0.937	0.8	0.87
	AlexNet	0.884	0.883	0.941	0.886	0.885
	SqueezeNet	0.861	0.859	0.93	0.87	0.865
	Res Net 18	0.91	0.89	0.94	0.875	0.909
Bacterial Pneumonia	VGG 16	0.85	0.91	0.85	0.81	0.91
	AlexNet	0.9	0.94	0.845	0.86	0.921
	SqueezeNet	0.83	0.905	0.75	0.75	0.84
	Res Net 18	0.92	0.96	0.86	0.89	0.94
COVID-19	VGG 16	0.966	0.95	0.981	0.981	0.965
	AlexNet	0.973	0.98	0.952	0.963	0.976
	SqueezeNet	0.976	0.98	0.962	0.964	0.972
	Res Net 18	0.977	0.98	0.963	0.964	0.973

Conclusion

The detection of COVID-19, bacterial, and viral pneumonias, as well as healthy cases, was performed in this work utilising deep learning models. Because it is critical to detect COVID-19 that spreads swiftly and globally, artificial intelligence approaches are utilised to do so precisely and promptly. By comparing model visualisation across a large number of samples as well as accuracy scores, we demonstrated the importance of Grad-CAM heatmaps are used as the major model validation metric. The deep learning models employed in the proposed approach (“VGG16, Alex Net, Squeeze Net, and ResNet 18“) have less parameters than the other deep learning models which resulted in increasing speed and performance. Experiments indicated that combining chest X-ray and fuzzy features increases the classification ability, with an accuracy rate of up to 97.7% using ResNet 18, which is a higher value when compared to the other methodologies. The suggested model makes use of the end-to-end learning scheme, which is one of the major advantages of CNN models. In the future, deep learning-based analysis can be performed on images of other organs impacted by the virus, in accordance with the views of COVID-19 experts. Additionally, we can investigate using a bigger chest X-ray dataset to evaluate our suggested model in the near future. Our research is not focused on building a perfect detection system; rather, it is focused on developing cost-effective approaches to tackle this disease.

Data availability

Kaggle: COVID-19 Detection X-Ray Dataset, https://doi.org/10.34740/kaggle/dsv/3122958.³¹

The dataset is organized into 3 folders (train, test, val) and contains subfolders for each image category (Pneumonia/Normal). There are 5,863 X-Ray images (JPEG) and 2 categories (Pneumonia/Normal).

Data are available under the terms of the Creative Commons Attribution-Sharealike 4.0 International (CC BY-SA 4.0).

References

1. Rahmani AM, Mirmahaleh SYH: Coronavirus disease (COVID-19) prevention and treatment methods and effective parameters: A systematic literature review. Sustain. Cities Soc. 2021; 64: 102568. Publisher Full Text
2. Jain G, Mittal D, Thakur D, et al.: A deep learning approach to detect Covid-19 coronavirus with X-Ray images. Biocybernetics and biomedical engineering. 2020; 40(4): 1391–1405. PubMed Abstract | Publisher Full Text
3. Shoeibi A, Khodatars M, Alizadehsani R, et al.: Automated detection and forecasting of covid-19 using deep learning techniques: A review. arXiv preprint arXiv:2007.10785. 2020.
4. Saygılı A: A new approach for computer-aided detection of coronavirus (COVID-19) from CT and X-ray images using machine learning methods. Appl. Soft Comput. 2021 Jul 1; 105: 107323. PubMed Abstract | Publisher Full Text
5. Sakib S, Tazrin T, Fouda MM, et al.: DL-CRC: deep learning-based chest radiograph classification for COVID-19 detection: a novel approach. Ieee Access. 2020 Sep 18; 8: 171575–171589. PubMed Abstract | Publisher Full Text
6. Ismael AM, Şengür A: Deep learning approaches for COVID-19 detection based on chest X-ray images. Expert Syst. Appl. 2021 Feb 1; 164: 114054. PubMed Abstract | Publisher Full Text
7. Hu Q, Gois FN, Costa R, et al.: Explainable artificial intelligence-based edge fuzzy images for COVID-19 detection and identification. Appl. Soft Comput. 2022 Jul 1; 123: 108966. PubMed Abstract | Publisher Full Text
8. Choudhury SH, Aurin AJ, Mitaly TA, et al.: Predicting the possibility of COVID-19 infection using fuzzy logic system. International Journal of Intelligent Information and Database Systems. 2021; 14(3): 239–256. Publisher Full Text
9. Ieracitano, et al.: A Fuzzy-enhanced Deep Learning Approach for Early Detection of Covid-19 Pneumonia from Portable Chest X-Ray Images. Neurocomputing. 2022; 481: 202–215. PubMed Abstract | Publisher Full Text
10. Alqudah AM, Qazan S, Alqudah A: Automated systems for detection of COVID-19 using chest X-ray images and lightweight convolutional neural networks.2020.
11. Cohen JP, Morrison P, Dao L: COVID-19 image data collection. arXiv preprint arXiv:2003.11597. 2020.
12. Apostolopoulos ID, Bessiana T: Covid-19: Automatic detection from X-Ray images utilizing Transfer Learning with Convolutional Neural Networks. arXiv e-prints. 2020 March 01. arXiv preprint arXiv:2003.11617. 2020.
13. Toğaçar M, Ergen B, Cömert Z: COVID-19 detection using deep learning models to exploit Social Mimic Optimization and structured chest X-ray images using fuzzy color and stacking approaches. Comput. Biol. Med. 2020; 121: 103805. Publisher Full Text
14. Ozturk T, Talo M, Yildirim EA, et al.: Automated detection of COVID-19 cases using deep neural networks with X-ray images. Comput. Biol. Med. 2020; 121: 103792. PubMed Abstract | Publisher Full Text
15. Wang L, Lin ZQ, Wong A: Covid-net: A tailored deep convolutional neural network design for detection of covid-19 cases from chest x-ray images. Sci. Rep. 2020 Nov 11; 10(1): 1–2.
16. Smith LN: Cyclical learning rates for training neural networks. 2017 IEEE winter conference on applications of computer vision (WACV). IEEE;2017, March; (pp. 464–472).
17. Gilanie G, Bajwa UI, Waraich MM, et al.: Coronavirus (COVID-19) detection from chest radiology images using convolutional neural networks. Biomedical Signal Processing and Control. 2021 Apr 1; 66: 102490. PubMed Abstract | Publisher Full Text
18. GitHub:COVID-19. [dataset].Retrieved March 10, 2020.Reference Source
19. Akçay S, Kundegorski ME, Devereux M, et al.: Transfer learning using convolutional neural networks for object classification within X-ray baggage security imagery. 2016 IEEE International Conference on Image Processing (ICIP). IEEE;2016, September; (pp. 1057–1061).
20. Salehi S, Abedi A, Balakrishnan S, et al.: Coronavirus disease 2019 (COVID-19): a systematic review of imaging findings in 919 patients. AJR Am. J. Roentgenol. 2020; 215(1): 87–93. PubMed Abstract | Publisher Full Text
21. Minaee S, Kafieh R, Sonka M, et al.: Deep-COVID: Predicting COVID-19 from chest X-ray images using deep transfer learning. Med. Image Anal. 2020; 65: 101794. PubMed Abstract | Publisher Full Text
22. Yang H, Mei S, Song K, et al.: Transfer-learning-based online Mura defect classification. IEEE Trans. Semicond. Manuf. 2017; 31(1): 116–123. Publisher Full Text
23. Panwar H, Gupta PK, Siddiqui MK, et al.: Application of deep learning for fast detection of COVID-19 in X-Rays using nCOVnet. Chaos, Solitons Fractals. 2020; 138: 109944. PubMed Abstract | Publisher Full Text
24. Ismael SA, Mohammed A, Hefny H: An enhanced deep learning approach for brain cancer MRI images classification using residual networks. Artif. Intell. Med. 2020 Jan 1; 102: 101779. PubMed Abstract | Publisher Full Text
25. Belaid ON, Loudini M: Classification of brain tumor by combination of pre-trained vgg16 cnn. Journal of Information Technology Management. 2020; 12(2): 13–25.
26. Heidari A, Erfanian H: Online COVID-19 Infection Diagnoses via Chest X-Ray Images using Alexnet Deep Learning Model. International Journal of Web Research. 2022 Jun 1; 5(1): 50–55.
27. Syed Nizamudeen Ahmed J, Mohamed Sathik M, Nallaperumal K, et al.:Automated Diagnosis of COVID-19 Using Squeeze Net Architecture Based on Deep Learning. InEmergent Converging Technologies and Biomedical Systems. Singapore:Springer;2022; (pp. 651–662).
28. Showkat S, Qureshi S: Efficacy of Transfer Learning-based ResNet models in Chest X-ray image classification for detecting COVID-19 Pneumonia. Chemom. Intell. Lab. Syst. 2022 May 15; 224: 104534. PubMed Abstract | Publisher Full Text
29. Alom MZ, Yakopcic C, Nasrin M, et al.: Breast cancer classification from histopathological images with inception recurrent residual convolutional neural network. J. Digit. Imaging. 2019; 32(4): 605–617. PubMed Abstract | Publisher Full Text
30. Selvaraju RR, Cogswell M, Das A, et al.: Grad-cam: Visual explanations from deep networks via gradient-based localization. Proceedings of the IEEE international conference on computer vision. 2017; (pp. 618–626).
31. Tahir AM, Chowdhury MEH, Qiblawey Y, et al.: COVID-QU-Ex Dataset . [Data set]. Kaggle.2022. Publisher Full Text

Comments on this article Comments (0)

Version 1

VERSION 1 PUBLISHED 16 Jan 2023

Author details Author details

¹ Electronics and Communications Engineering, Gokarjurangara Institute of Engineering and Technology, Hyderabad, Telangana, 500090, India
² Research Scholar, KL University, Viyayawada, Andhrapradesh, 522 302, India
³ Electronics and Communications Engineering, KL University, Vijayawada, Andhrapradesh, 522 302, India

Priyanka Yadlapalli
Roles: Conceptualization, Data Curation, Investigation, Methodology, Resources, Supervision, Validation, Writing – Original Draft Preparation, Writing – Review & Editing

Bhavana D
Roles: Supervision, Validation

Competing interests

No competing interests were disclosed.

Grant information

The author(s) declared that no grants were involved in supporting this work.

Article Versions (1)

version 1

Published: 16 Jan 2023, 12:60

https://doi.org/10.12688/f1000research.126197.1

Copyright

© 2023 Yadlapalli P and D B. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Download

Export To

metrics

	Views	Downloads
F1000Research	-	-
PubMed Central Data from PMC are received and updated monthly.	-	-

Citations

0

SEE MORE DETAILS

CITE

how to cite this article

Yadlapalli P and D B. Application of Fuzzy Deep Neural Networks for Covid 19 diagnosis through chest Radiographs [version 1; peer review: 1 approved with reservations, 1 not approved] F1000Research 2023, 12:60 (https://doi.org/10.12688/f1000research.126197.1)

NOTE: it is important to ensure the information in square brackets after the title is included in all citations of this article.

track

receive updates on this article

Track an article to receive email alerts on any updates to this article.

Open Peer Review

Version 1

VERSION 1

PUBLISHED 16 Jan 2023

Views

1

Reviewer Report 24 Nov 2023

Ervin Gubin Moung, Faculty of Computing and Informatics, Universiti Malaysia Sabah, Sabah, Malaysia

Not Approved

https://doi.org/10.5256/f1000research.138580.r204708

The paper provides an overview of the methodology, but details like specific model architectures and configurations are not fully presented in the paper.
Specifics about Random's seed values or the context of

The paper provides an overview of the methodology, but details like specific model architectures and configurations are not fully presented in the paper.
Specifics about Random's seed values or the context of randomness aren't detailed in the paper. Kindly provide this information for reproducibility purpose.
Specific hyperparameters and model configurations were not found in the paper. Kindly tabulate all these information for reproducibility purpose.
Explicit information and settings about the preprocessing step is not mentioned in this paper. Kindly tabulate this information for reproducibility purpose.
The paper does mention splitting the data into training (70%), testing (20%), and validation (10%) sets, but specifics about each set or potential overlaps are not provided. Provide the "seed" information for random shuffling (if applied).
A comparison of this paper's performance against existing literature on the same topic is missing. Kindly add a table that compares and discusses the key metrics, methodologies, and results from both this study and relevant works in the field. This will provide readers with a clearer context and understanding of the research's significance and advancements.
The abstract and conclusion of the paper don’t clearly state what makes this research special. Please make sure to highlight the main contributions of your work in these sections. Additionally, it would be helpful to talk about any limitations of your study and suggest ideas for what might be researched next.

Is the work clearly and accurately presented and does it cite the current literature?
Yes
Is the study design appropriate and is the work technically sound?
Yes
Are sufficient details of methods and analysis provided to allow replication by others?
Yes
If applicable, is the statistical analysis and its interpretation appropriate?
Not applicable
Are all the source data underlying the results available to ensure full reproducibility?
No
Are the conclusions drawn adequately supported by the results?
Yes

Competing Interests: No competing interests were disclosed.

Reviewer Expertise: machine learning, deep learning, object detection, artificial intelligence.

I confirm that I have read this submission and believe that I have an appropriate level of expertise to state that I do not consider it to be of an acceptable scientific standard, for reasons outlined above.

CITE

Report a concern

Respond or Comment

Views

5

Reviewer Report 13 Oct 2023

Kogilavani Shanmugavadivel, Kongu Engineering College, Perundurai, Tamil Nadu, India

Approved with Reservations

https://doi.org/10.5256/f1000research.138580.r204714

Summary

The article proposes a novel approach that combines fuzzy logic with deep learning, specifically employing ResNet-18 architecture, to diagnose COVID-19 and distinguish it from other types of pneumonia and healthy cases through chest X-ray images. The ... Continue reading

Summary

The article proposes a novel approach that combines fuzzy logic with deep learning, specifically employing ResNet-18 architecture, to diagnose COVID-19 and distinguish it from other types of pneumonia and healthy cases through chest X-ray images. The introduction emphasizes the scarcity of COVID-19 detection kits and the potential of deep learning in medical imaging. The methodology section highlights the integration of fuzzy logic within the deep neural network and the use of a carefully selected dataset. The results demonstrate a high classification accuracy of 97% for COVID-19 detection, outperforming existing literature. The study concludes by affirming the effectiveness of the proposed approach and underscores the potential of employing deep learning techniques for accurate classification of COVID-19 and other pneumonia cases, suggesting promising applications in the near future.

Review Comments

The introduction could benefit from a clear statement of the research objectives and hypotheses to provide readers with a more focused understanding of the study's purpose.
Consider including a brief overview of the limitations and challenges associated with using chest X-ray images for COVID-19 diagnosis to provide a well-rounded context for the study.
Provide a thorough explanation of the dataset used, including its size, diversity, and any potential biases, to ensure transparency and replicability of the study.
Elaborate on the selection criteria for chest X-ray images to establish the rationale for including specific images in the dataset.
Clarify the rationale for choosing ResNet-18 architecture and why it was deemed suitable for the proposed classification task.
Discuss any pre-processing techniques applied to the chest X-ray images and justify their necessity for improving the model's performance.
Include comparisons with other deep learning models or traditional machine learning approaches to showcase the superiority of the proposed fuzzy deep neural network.
Provide more details on the fuzzy logic integration within the deep learning model and how it enhances the classification accuracy.
Incorporate visualizations of the fuzzy logic components and how they influence the network's decision-making process for improved comprehension.
Discuss potential challenges or limitations of the fuzzy logic integration, such as computational complexity or interpretability, and propose ways to mitigate them.
Clarify the evaluation metrics used (e.g., accuracy, precision, recall) and why they were chosen to assess the model's performance.
Suggest future directions for research, such as exploring the interpretability of the fuzzy logic components in the context of COVID-19 diagnosis.
Consider discussing the computational resources required to implement the proposed approach, which can be crucial for practical application and scalability.
Highlight any ethical considerations related to the use of medical imaging data and deep learning in healthcare, emphasizing the importance of patient privacy and data security.
Provide a brief discussion on the potential impact of the proposed approach on clinical practice and patient outcomes.
Ensure proper citation and acknowledgment of previous research in the field, especially regarding existing approaches to COVID-19 detection using medical imaging.
Review the manuscript for grammatical errors and clarity to enhance readability and comprehension.

Is the work clearly and accurately presented and does it cite the current literature?
Yes
Is the study design appropriate and is the work technically sound?
Partly
Are sufficient details of methods and analysis provided to allow replication by others?
Partly
If applicable, is the statistical analysis and its interpretation appropriate?
Partly
Are all the source data underlying the results available to ensure full reproducibility?
Yes
Are the conclusions drawn adequately supported by the results?
Yes

Competing Interests: No competing interests were disclosed.

Reviewer Expertise: Machine learning, Natural language Processing

I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.

CITE

Report a concern

Respond or Comment

Comments on this article Comments (0)

Version 1

VERSION 1 PUBLISHED 16 Jan 2023

Open Peer Review

Reviewer Status

Reviewer Reports

	Invited Reviewers
	1	2
Version 1 16 Jan 23	read	read

Kogilavani Shanmugavadivel, Kongu Engineering College, Perundurai, India
Ervin Gubin Moung, Universiti Malaysia Sabah, Sabah, Malaysia

Comments on this article

All Comments(0)

Add a comment

Sign up for content alerts

Browse by related subjects

Back to all reports

Reviewer Report

1 Views

24 Nov 2023 | for Version 1

Ervin Gubin Moung, Faculty of Computing and Informatics, Universiti Malaysia Sabah, Sabah, Malaysia

1 Views Cite this report Responses(0)

Not Approved

The paper provides an overview of the methodology, but details like specific model architectures and configurations are not fully presented in the paper.
Specifics about Random's seed values or the context of randomness aren't detailed in the paper. Kindly provide this information for reproducibility purpose.
Specific hyperparameters and model configurations were not found in the paper. Kindly tabulate all these information for reproducibility purpose.
Explicit information and settings about the preprocessing step is not mentioned in this paper. Kindly tabulate this information for reproducibility purpose.
The paper does mention splitting the data into training (70%), testing (20%), and validation (10%) sets, but specifics about each set or potential overlaps are not provided. Provide the "seed" information for random shuffling (if applied).
A comparison of this paper's performance against existing literature on the same topic is missing. Kindly add a table that compares and discusses the key metrics, methodologies, and results from both this study and relevant works in the field. This will provide readers with a clearer context and understanding of the research's significance and advancements.
The abstract and conclusion of the paper don’t clearly state what makes this research special. Please make sure to highlight the main contributions of your work in these sections. Additionally, it would be helpful to talk about any limitations of your study and suggest ideas for what might be researched next.

Is the work clearly and accurately presented and does it cite the current literature?
Yes
Is the study design appropriate and is the work technically sound?
Yes
Are sufficient details of methods and analysis provided to allow replication by others?
Yes
If applicable, is the statistical analysis and its interpretation appropriate?
Not applicable
Are all the source data underlying the results available to ensure full reproducibility?
No
Are the conclusions drawn adequately supported by the results?
Yes

Competing Interests

No competing interests were disclosed.

Reviewer Expertise

machine learning, deep learning, object detection, artificial intelligence.

I confirm that I have read this submission and believe that I have an appropriate level of expertise to state that I do not consider it to be of an acceptable scientific standard, for reasons outlined above.

Respond to this report

Responses (0)

Back to all reports

Reviewer Report

5 Views

13 Oct 2023 | for Version 1

Kogilavani Shanmugavadivel, Kongu Engineering College, Perundurai, Tamil Nadu, India

5 Views Cite this report Responses(0)

Approved With Reservations

Summary

The article proposes a novel approach that combines fuzzy logic with deep learning, specifically employing ResNet-18 architecture, to diagnose COVID-19 and distinguish it from other types of pneumonia and healthy cases through chest X-ray images. The introduction emphasizes the scarcity of COVID-19 detection kits and the potential of deep learning in medical imaging. The methodology section highlights the integration of fuzzy logic within the deep neural network and the use of a carefully selected dataset. The results demonstrate a high classification accuracy of 97% for COVID-19 detection, outperforming existing literature. The study concludes by affirming the effectiveness of the proposed approach and underscores the potential of employing deep learning techniques for accurate classification of COVID-19 and other pneumonia cases, suggesting promising applications in the near future.

Review Comments

The introduction could benefit from a clear statement of the research objectives and hypotheses to provide readers with a more focused understanding of the study's purpose.
Consider including a brief overview of the limitations and challenges associated with using chest X-ray images for COVID-19 diagnosis to provide a well-rounded context for the study.
Provide a thorough explanation of the dataset used, including its size, diversity, and any potential biases, to ensure transparency and replicability of the study.
Elaborate on the selection criteria for chest X-ray images to establish the rationale for including specific images in the dataset.
Clarify the rationale for choosing ResNet-18 architecture and why it was deemed suitable for the proposed classification task.
Discuss any pre-processing techniques applied to the chest X-ray images and justify their necessity for improving the model's performance.
Include comparisons with other deep learning models or traditional machine learning approaches to showcase the superiority of the proposed fuzzy deep neural network.
Provide more details on the fuzzy logic integration within the deep learning model and how it enhances the classification accuracy.
Incorporate visualizations of the fuzzy logic components and how they influence the network's decision-making process for improved comprehension.
Discuss potential challenges or limitations of the fuzzy logic integration, such as computational complexity or interpretability, and propose ways to mitigate them.
Clarify the evaluation metrics used (e.g., accuracy, precision, recall) and why they were chosen to assess the model's performance.
Suggest future directions for research, such as exploring the interpretability of the fuzzy logic components in the context of COVID-19 diagnosis.
Consider discussing the computational resources required to implement the proposed approach, which can be crucial for practical application and scalability.
Highlight any ethical considerations related to the use of medical imaging data and deep learning in healthcare, emphasizing the importance of patient privacy and data security.
Provide a brief discussion on the potential impact of the proposed approach on clinical practice and patient outcomes.
Ensure proper citation and acknowledgment of previous research in the field, especially regarding existing approaches to COVID-19 detection using medical imaging.
Review the manuscript for grammatical errors and clarity to enhance readability and comprehension.

Is the work clearly and accurately presented and does it cite the current literature?
Yes
Is the study design appropriate and is the work technically sound?
Partly
Are sufficient details of methods and analysis provided to allow replication by others?
Partly
If applicable, is the statistical analysis and its interpretation appropriate?
Partly
Are all the source data underlying the results available to ensure full reproducibility?
Yes
Are the conclusions drawn adequately supported by the results?
Yes

Competing Interests

No competing interests were disclosed.

Reviewer Expertise

Machine learning, Natural language Processing

I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.

Respond to this report

Responses (0)

[1] 1. Rahmani AM, Mirmahaleh SYH: Coronavirus disease (COVID-19) prevention and treatment methods and effective parameters: A systematic literature review. Sustain. Cities Soc. 2021; 64: 102568. Publisher Full Text

[2] 2. Jain G, Mittal D, Thakur D, et al.: A deep learning approach to detect Covid-19 coronavirus with X-Ray images. Biocybernetics and biomedical engineering. 2020; 40(4): 1391–1405. PubMed Abstract | Publisher Full Text

[3] 3. Shoeibi A, Khodatars M, Alizadehsani R, et al.: Automated detection and forecasting of covid-19 using deep learning techniques: A review. arXiv preprint arXiv:2007.10785. 2020.

[4] 4. Saygılı A: A new approach for computer-aided detection of coronavirus (COVID-19) from CT and X-ray images using machine learning methods. Appl. Soft Comput. 2021 Jul 1; 105: 107323. PubMed Abstract | Publisher Full Text

[5] 5. Sakib S, Tazrin T, Fouda MM, et al.: DL-CRC: deep learning-based chest radiograph classification for COVID-19 detection: a novel approach. Ieee Access. 2020 Sep 18; 8: 171575–171589. PubMed Abstract | Publisher Full Text

[6] 6. Ismael AM, Şengür A: Deep learning approaches for COVID-19 detection based on chest X-ray images. Expert Syst. Appl. 2021 Feb 1; 164: 114054. PubMed Abstract | Publisher Full Text

[7] 7. Hu Q, Gois FN, Costa R, et al.: Explainable artificial intelligence-based edge fuzzy images for COVID-19 detection and identification. Appl. Soft Comput. 2022 Jul 1; 123: 108966. PubMed Abstract | Publisher Full Text

[8] 8. Choudhury SH, Aurin AJ, Mitaly TA, et al.: Predicting the possibility of COVID-19 infection using fuzzy logic system. International Journal of Intelligent Information and Database Systems. 2021; 14(3): 239–256. Publisher Full Text

[9] 9. Ieracitano, et al.: A Fuzzy-enhanced Deep Learning Approach for Early Detection of Covid-19 Pneumonia from Portable Chest X-Ray Images. Neurocomputing. 2022; 481: 202–215. PubMed Abstract | Publisher Full Text

[10] 10. Alqudah AM, Qazan S, Alqudah A: Automated systems for detection of COVID-19 using chest X-ray images and lightweight convolutional neural networks.2020.

[11] 11. Cohen JP, Morrison P, Dao L: COVID-19 image data collection. arXiv preprint arXiv:2003.11597. 2020.

[12] 12. Apostolopoulos ID, Bessiana T: Covid-19: Automatic detection from X-Ray images utilizing Transfer Learning with Convolutional Neural Networks. arXiv e-prints. 2020 March 01. arXiv preprint arXiv:2003.11617. 2020.

[13] 13. Toğaçar M, Ergen B, Cömert Z: COVID-19 detection using deep learning models to exploit Social Mimic Optimization and structured chest X-ray images using fuzzy color and stacking approaches. Comput. Biol. Med. 2020; 121: 103805. Publisher Full Text

[14] 14. Ozturk T, Talo M, Yildirim EA, et al.: Automated detection of COVID-19 cases using deep neural networks with X-ray images. Comput. Biol. Med. 2020; 121: 103792. PubMed Abstract | Publisher Full Text

[15] 15. Wang L, Lin ZQ, Wong A: Covid-net: A tailored deep convolutional neural network design for detection of covid-19 cases from chest x-ray images. Sci. Rep. 2020 Nov 11; 10(1): 1–2.

[16] 16. Smith LN: Cyclical learning rates for training neural networks. 2017 IEEE winter conference on applications of computer vision (WACV). IEEE;2017, March; (pp. 464–472).

[17] 17. Gilanie G, Bajwa UI, Waraich MM, et al.: Coronavirus (COVID-19) detection from chest radiology images using convolutional neural networks. Biomedical Signal Processing and Control. 2021 Apr 1; 66: 102490. PubMed Abstract | Publisher Full Text

[18] 18. GitHub:COVID-19. [dataset].Retrieved March 10, 2020.Reference Source

[19] 19. Akçay S, Kundegorski ME, Devereux M, et al.: Transfer learning using convolutional neural networks for object classification within X-ray baggage security imagery. 2016 IEEE International Conference on Image Processing (ICIP). IEEE;2016, September; (pp. 1057–1061).

[20] 20. Salehi S, Abedi A, Balakrishnan S, et al.: Coronavirus disease 2019 (COVID-19): a systematic review of imaging findings in 919 patients. AJR Am. J. Roentgenol. 2020; 215(1): 87–93. PubMed Abstract | Publisher Full Text

[21] 21. Minaee S, Kafieh R, Sonka M, et al.: Deep-COVID: Predicting COVID-19 from chest X-ray images using deep transfer learning. Med. Image Anal. 2020; 65: 101794. PubMed Abstract | Publisher Full Text

[22] 22. Yang H, Mei S, Song K, et al.: Transfer-learning-based online Mura defect classification. IEEE Trans. Semicond. Manuf. 2017; 31(1): 116–123. Publisher Full Text

[23] 23. Panwar H, Gupta PK, Siddiqui MK, et al.: Application of deep learning for fast detection of COVID-19 in X-Rays using nCOVnet. Chaos, Solitons Fractals. 2020; 138: 109944. PubMed Abstract | Publisher Full Text

[24] 24. Ismael SA, Mohammed A, Hefny H: An enhanced deep learning approach for brain cancer MRI images classification using residual networks. Artif. Intell. Med. 2020 Jan 1; 102: 101779. PubMed Abstract | Publisher Full Text

[25] 25. Belaid ON, Loudini M: Classification of brain tumor by combination of pre-trained vgg16 cnn. Journal of Information Technology Management. 2020; 12(2): 13–25.

[26] 26. Heidari A, Erfanian H: Online COVID-19 Infection Diagnoses via Chest X-Ray Images using Alexnet Deep Learning Model. International Journal of Web Research. 2022 Jun 1; 5(1): 50–55.

[27] 27. Syed Nizamudeen Ahmed J, Mohamed Sathik M, Nallaperumal K, et al.:Automated Diagnosis of COVID-19 Using Squeeze Net Architecture Based on Deep Learning. InEmergent Converging Technologies and Biomedical Systems. Singapore:Springer;2022; (pp. 651–662).

[28] 28. Showkat S, Qureshi S: Efficacy of Transfer Learning-based ResNet models in Chest X-ray image classification for detecting COVID-19 Pneumonia. Chemom. Intell. Lab. Syst. 2022 May 15; 224: 104534. PubMed Abstract | Publisher Full Text

[29] 29. Alom MZ, Yakopcic C, Nasrin M, et al.: Breast cancer classification from histopathological images with inception recurrent residual convolutional neural network. J. Digit. Imaging. 2019; 32(4): 605–617. PubMed Abstract | Publisher Full Text

[30] 30. Selvaraju RR, Cogswell M, Das A, et al.: Grad-cam: Visual explanations from deep networks via gradient-based localization. Proceedings of the IEEE international conference on computer vision. 2017; (pp. 618–626).

[31] 31. Tahir AM, Chowdhury MEH, Qiblawey Y, et al.: COVID-QU-Ex Dataset . [Data set]. Kaggle.2022. Publisher Full Text

Application of Fuzzy Deep Neural Networks for Covid 19 diagnosis through chest Radiographs

Abstract

Keywords

Introduction

Literature survey

Table 1. Research works related to COVID-19 and deep learning.

Figure 1. Procedure of using deep learning techniques in chest X-rays.

Figure 2. Combined dataset of both coronavirus disease 2019 (COVID-19) and pneumonia X-ray images.

Table 2. Details of the datasets used.

Figure 3. Distribution of data for final four class classification.

Figure 4. A) Normal B) bacterial pneumonia C) viral pneumonia D) COVID-19.

Methods in testing

Figure 5. Convolution neural network architecture.

Image preprocessing

Data augmentation

Rotation

Fuzzy enhanced CNN -based transfer learning

Figure 6. Concept of transfer learning.

VGG 16 Architecture

Alex Net Architecture

SqueezeNet

Architecture of ResNet Model (ResNet 18)

Figure 7. Structure of Residual Block.

Figure 8. Schematic of transfer learning with ResNet 18.

Learning rate assessment

(1)

Testing methods

Figure 9. ResNet18 network model predicted class activation heat maps.

Metrics for evaluating performance

Figure 10. Confusion matrix of 4 classes (normal, viral pneumonia, bacterial pneumonia).

Confusion matrix

(2)

(3)

(4)

(5)

(6)

Results

Table 3. Comparison of different deep learning techniques for the four different types of virus.

Conclusion

Data availability

References

Comments on this article Comments (0)

Open Peer Review

Comments on this article Comments (0)

Open Peer Review

Reviewer Status

Reviewer Reports

Comments on this article

Browse by related subjects

Competing Interests Policy

Stay Updated