However, it has some limitations that affect its quality. Med. Med. The Softmax activation function is used for this purpose because the output should be binary (positive COVID-19 negative COVID-19). This algorithm is tested over a global optimization problem. Eurosurveillance 18, 20503 (2013). Background The large volume and suboptimal image quality of portable chest X-rays (CXRs) as a result of the COVID-19 pandemic could post significant challenges for radiologists and frontline physicians. CNNs are more appropriate for large datasets. Cohen, J.P., Morrison, P. & Dao, L. Covid-19 image data collection. 2 (right). Adv. To survey the hypothesis accuracy of the models. Image Underst. & Baby, C.J. Emphysema medical image classification using fuzzy decision tree with fuzzy particle swarm optimization clustering. EMRes-50 model . Szegedy, C. et al. kharrat and Mahmoud32proposed an FS method based on a hybrid of Simulated Annealing (SA) and GA to classify brain tumors using MRI. The MPA starts with the initialization phase and then passing by other three phases with respect to the rational velocity among the prey and the predator. Covid-19 dataset. arXiv preprint arXiv:1711.05225 (2017). ISSN 2045-2322 (online). Technol. 6 (left), for dataset 1, it can be seen that our proposed FO-MPA approach outperforms other CNN models like VGGNet, Xception, Inception, Mobilenet, Nasnet, and Resnet. (14)-(15) are implemented in the first half of the agents that represent the exploitation. Eng. Brain tumor segmentation with deep neural networks. Ge, X.-Y. However, the proposed IMF approach achieved the best results among the compared algorithms in least time. This study aims to improve the COVID-19 X-ray image classification using feature selection technique by the regression mutual information deep convolution neuron networks (RMI Deep-CNNs). Accordingly, the FC is an efficient tool for enhancing the performance of the meta-heuristic algorithms by considering the memory perspective during updating the solutions. Vis. Lilang Zheng, Jiaxuan Fang, Xiaorun Tang, Hanzhang Li, Jiaxin Fan, Tianyi Wang, Rui Zhou, Zhaoyan Yan: PVT-COV19D: COVID-19 Detection Through Medical Image Classification Based on Pyramid Vision Transformer. a cough chills difficulty breathing tiredness body aches headaches a new loss of taste or smell a sore throat nausea and vomiting diarrhea Not everyone with COVID-19 develops all of these. \end{aligned} \end{aligned}$$, $$\begin{aligned} WF(x)=\exp ^{\left( {\frac{x}{k}}\right) ^\zeta } \end{aligned}$$, $$\begin{aligned}&Accuracy = \frac{\text {TP} + \text {TN}}{\text {TP} + \text {TN} + \text {FP} + \text {FN}} \end{aligned}$$, $$\begin{aligned}&Sensitivity = \frac{\text {TP}}{\text{ TP } + \text {FN}}\end{aligned}$$, $$\begin{aligned}&Specificity = \frac{\text {TN}}{\text {TN} + \text {FP}}\end{aligned}$$, $$\begin{aligned}&F_{Score} = 2\times \frac{\text {Specificity} \times \text {Sensitivity}}{\text {Specificity} + \text {Sensitivity}} \end{aligned}$$, $$\begin{aligned} Best_{acc} = \max _{1 \le i\le {r}} Accuracy \end{aligned}$$, $$\begin{aligned} Best_{Fit_i} = \min _{1 \le i\le r} Fit_i \end{aligned}$$, $$\begin{aligned} Max_{Fit_i} = \max _{1 \le i\le r} Fit_i \end{aligned}$$, $$\begin{aligned} \mu = \frac{1}{r} \sum _{i=1}^N Fit_i \end{aligned}$$, $$\begin{aligned} STD = \sqrt{\frac{1}{r-1}\sum _{i=1}^{r}{(Fit_i-\mu )^2}} \end{aligned}$$, https://doi.org/10.1038/s41598-020-71294-2. \end{aligned} \end{aligned}$$, $$\begin{aligned} \begin{aligned} U_{i}(t+1)&= \frac{1}{1!} 101, 646667 (2019). I. S. of Medical Radiology. PubMed is applied before larger sized kernels are applied to reduce the dimension of the channels, which accordingly, reduces the computation cost. For this motivation, we utilize the FC concept with the MPA algorithm to boost the second step of the standard version of the algorithm. COVID-19 image classification using deep features and fractional-order marine predators algorithm. The memory terms of the prey are updated at the end of each iteration based on first in first out concept. In this paper, Inception is applied as a feature extractor, where the input image shape is (229, 229, 3). Lilang Zheng, Jiaxuan Fang, Xiaorun Tang, Hanzhang Li, Jiaxin Fan, Tianyi Wang, Rui Zhou, Zhaoyan Yan. However, WOA showed the worst performances in these measures; which means that if it is run in the same conditions several times, the same results will be obtained. For the exploration stage, the weibull distribution has been applied rather than Brownian to bost the performance of the predator in stage 2 and the prey velocity in stage 1 based on the following formula: Where k, and \(\zeta\) are the scale and shape parameters. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 770778 (2016). Google Scholar. used a dark Covid-19 network for multiple classification experiments on Covid-19 with an accuracy of 87% [ 23 ]. In this work, the MPA is enhanced by fractional calculus memory feature, as a result, Fractional-order Marine Predators Algorithm (FO-MPA) is introduced. COVID-19 (coronavirus disease 2019) is a new viral infection disease that is widely spread worldwide. Based on Standard Deviation measure (STD), the most stable algorithms were SCA, SGA, BPSO, and bGWO, respectively. In the meantime, to ensure continued support, we are displaying the site without styles Bukhari, S. U.K., Bukhari, S. S.K., Syed, A. A. et al. FP (false positives) are the positive COVID-19 images that were incorrectly labeled as negative COVID-19, while FN (false negatives) are the negative COVID-19 images that were mislabeled as positive COVID-19 images. This study presents an investigation on 16 pretrained CNNs for classification of COVID-19 using a large public database of CT scans collected from COVID-19 patients and non-COVID-19 subjects. 43, 302 (2019). It noted that all produced feature vectors by CNNs used in this paper are at least bigger by more than 300 times compared to that produced by FO-MPA in terms of the size of the featureset. In our experiment, we randomly split the data into 70%, 10%, and 20% for training, validation, and testing sets, respectively. Image segmentation is a necessary image processing task that applied to discriminate region of interests (ROIs) from the area of outsides. (23), the general formulation for the solutions of FO-MPA based on FC memory perspective can be written as follows: After checking the previous formula, it can be detected that the motion of the prey becomes based on some terms from the previous solutions with a length of (m), as depicted in Fig. Harikumar, R. & Vinoth Kumar, B. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. 42, 6088 (2017). It is obvious that such a combination between deep features and a feature selection algorithm can be efficient in several image classification tasks. According to the formula10, the initial locations of the prey and predator can be defined as below: where the Elite matrix refers to the fittest predators. Mirjalili, S. & Lewis, A. J. 121, 103792 (2020). Syst. The results showed that the proposed approach showed better performances in both classification accuracy and the number of extracted features that positively affect resource consumption and storage efficiency. Currently, we witness the severe spread of the pandemic of the new Corona virus, COVID-19, which causes dangerous symptoms to humans and animals, its complications may lead to death. The proposed approach selected successfully 130 and 86 out of 51 K features extracted by inception from dataset 1 and dataset 2, while improving classification accuracy at the same time. 25, 3340 (2015). Abadi, M. et al. Johnson, D.S., Johnson, D. L.L., Elavarasan, P. & Karunanithi, A. The results of max measure (as in Eq. Classification of COVID19 using Chest X-ray Images in Keras 4.6 33 ratings Share Offered By In this Guided Project, you will: Learn to Build and Train the Convolutional Neural Network using Keras with Tensorflow as Backend Learn to Visualize Data in Matplotlib Learn to make use of the Trained Model to Predict on a New Set of Data 2 hours Rajpurkar, P. etal. Faramarzi et al.37 implement this feature via saving the previous best solutions of a prior iteration, and compared with the current ones; the solutions are modified based on the best one during the comparison stage. (14)(15) to emulate the motion of the first half of the population (prey) and Eqs. Convolutional neural networks were implemented in Python 3 under Google Colaboratory46, commonly referred to as Google Colab, which is a research project for prototyping machine learning models on powerful hardware options such as GPUs and TPUs. Evaluation outcomes showed that GA based FS methods outperformed traditional approaches, such as filter based FS and traditional wrapper methods. Sci Rep 10, 15364 (2020). This combination should achieve two main targets; high performance and resource consumption, storage capacity which consequently minimize processing time. Chowdhury, M.E. etal. Google Scholar. A., Fan, H. & Abd ElAziz, M. Optimization method for forecasting confirmed cases of covid-19 in china. (iii) To implement machine learning classifiers for classification of COVID and non-COVID image classes. All authors discussed the results and wrote the manuscript together. ), such as \(5\times 5\), \(3 \times 3\), \(1 \times 1\). For more analysis of feature selection algorithms based on the number of selected features (S.F) and consuming time, Fig. Johnson et al.31 applied the flower pollination algorithm (FPA) to select features from CT images of the lung, to detect lung cancers. The results indicate that all CNN-based architectures outperform the ViT-based architecture in the binary classification of COVID-19 using CT images. J. Clin. In 2019 26th National and 4th International Iranian Conference on Biomedical Engineering (ICBME), 194198 (IEEE, 2019). COVID 19 X-ray image classification. Donahue, J. et al. These images have been further used for the classification of COVID-19 and non-COVID-19 images using ResNet50 and AlexNet convolutional neural network (CNN) models. 40, 2339 (2020). Therefore, in this paper, we propose a hybrid classification approach of COVID-19. The whole dataset contains around 200 COVID-19 positive images and 1675 negative COVID19 images. PubMedGoogle Scholar. & SHAH, S. S.H. The diagnostic evaluation of convolutional neural network (cnn) for the assessment of chest x-ray of patients infected with covid-19. The optimum path forest (OPF) classifier was applied to classify pulmonary nodules based on CT images. Moreover, the \(R_B\) parameter has been changed to depend on weibull distribution as described below. Furthermore, using few hundreds of images to build then train Inception is considered challenging because deep neural networks need large images numbers to work efficiently and produce efficient features. Inspired by this concept, Faramarzi et al.37 developed the MPA algorithm by considering both of a predator a prey as solutions. Moreover, other COVID-19 positive images were added by the Italian Society of Medical and Interventional Radiology (SIRM) COVID-19 Database45. One of the best methods of detecting. Pool layers are used mainly to reduce the inputs size, which accelerates the computation as well. Figure7 shows the most recent published works as in54,55,56,57 and44 on both dataset 1 and dataset 2. In this paper, after applying Chi-square, the feature vector is minimized for both datasets from 51200 to 2000. In order to normalize the values between 0 and 1 by dividing by the sum of all feature importance values, as in Eq. The Weibull Distribution is a heavy-tied distribution which presented as in Fig. To obtain Nguyen, L.D., Lin, D., Lin, Z. Key Definitions. For each of these three categories, there is a number of patients and for each of them, there is a number of CT scan images correspondingly. FC provides a clear interpretation of the memory and hereditary features of the process. However, some of the extracted features by CNN might not be sufficient, which may affect negatively the quality of the classification images. Afzali et al.15 proposed an FS method based on principal component analysis and contour-based shape descriptors to detect Tuberculosis from lung X-Ray Images. By filtering titles, abstracts, and content in the Google Scholar database, this literature review was able to find 19 related papers to answer two research questions, i.e. (22) can be written as follows: By using the discrete form of GL definition of Eq. Sign up for the Nature Briefing newsletter what matters in science, free to your inbox daily. Med. Eng. The prey follows Weibull distribution during discovering the search space to detect potential locations of its food. J. Refresh the page, check Medium 's site status, or find something interesting. We adopt a special type of CNN called a pre-trained model where the network is previously trained on the ImageNet dataset, which contains millions of variety of images (animal, plants, transports, objects,..) on 1000 classe categories. They concluded that the hybrid method outperformed original fuzzy c-means, and it had less sensitive to noises. This task is achieved by FO-MPA which randomly generates a set of solutions, each of them represents a subset of potential features. It can be concluded that FS methods have proven their advantages in different medical imaging applications19. By submitting a comment you agree to abide by our Terms and Community Guidelines. The data was collected mainly from retrospective cohorts of pediatric patients from Guangzhou Women and Childrens medical center. The proposed IFM approach is summarized as follows: Extracting deep features from Inception, where about 51 K features were extracted. The model was developed using Keras library47 with Tensorflow backend48. Correspondence to Adv. Cite this article. Fung, G. & Stoeckel, J. Svm feature selection for classification of spect images of alzheimers disease using spatial information. Provided by the Springer Nature SharedIt content-sharing initiative, Environmental Science and Pollution Research (2023), Archives of Computational Methods in Engineering (2023), Arabian Journal for Science and Engineering (2023). and JavaScript. Harikumar et al.18 proposed an FS method based on wavelets to classify normality or abnormality of different types of medical images, such as CT, MRI, ultrasound, and mammographic images. In this paper, we apply a convolutional neural network (CNN) to extract features from COVID-19 X-Ray images. Inspired by our recent work38, where VGG-19 besides statistically enhanced Salp Swarm Algorithm was applied to select the best features for White Blood Cell Leukaemia classification. The dataset consists of 21,165 chest X-ray (CXR) COVID-19 images distributed on four categories which are COVID19, lung opacity, viral pneumonia, and NORMAL (Non-COVID). The accuracy measure is used in the classification phase. Using X-ray images we can train a machine learning classifier to detect COVID-19 using Keras and TensorFlow. It classifies the chest X-ray images into three categories that includes Covid-19, Pneumonia and normal. (8) can be remodeled as below: where \(D^1[x(t)]\) represents the difference between the two followed events. where r is the run numbers. The given Kaggle dataset consists of chest CT scan images of patients suffering from the novel COVID-19, other pulmonary disorders, and those of healthy patients. The symbol \(r\in [0,1]\) represents a random number. Efficient classification of white blood cell leukemia with improved swarm optimization of deep features. We can call this Task 2. https://doi.org/10.1016/j.future.2020.03.055 (2020). & Carlsson, S. Cnn features off-the-shelf: an astounding baseline for recognition. PubMed The announcement confirmed that from May 8, following Japan's Golden Week holiday period, COVID-19 will be officially downgraded to Class 5, putting the virus on the same classification level as seasonal influenza. arXiv preprint arXiv:1704.04861 (2017). 78, 2091320933 (2019). A properly trained CNN requires a lot of data and CPU/GPU time. The survey asked participants to broadly classify the findings of each chest CT into one of the four RSNA COVID-19 imaging categories, then select which imaging features led to their categorization. Chong, D. Y. et al. Moreover, the Weibull distribution employed to modify the exploration function. They were also collected frontal and lateral view imagery and metadata such as the time since first symptoms, intensive care unit (ICU) status, survival status, intubation status, or hospital location. Automated detection of covid-19 cases using deep neural networks with x-ray images. After applying this technique, the feature vector is minimized from 2000 to 459 and from 2000 to 462 for Dataset1 and Dataset 2, respectively. Computational image analysis techniques play a vital role in disease treatment and diagnosis. By achieving 98.7%, 98.2% and 99.6%, 99% of classification accuracy and F-Score for dataset 1 and dataset 2, respectively, the proposed approach outperforms several CNNs and all recent works on COVID-19 images. Med. 132, 8198 (2018). In this work, we have used four transfer learning models, VGG16, InceptionV3, ResNet50, and DenseNet121 for the classification tasks. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate. E. B., Traina-Jr, C. & Traina, A. J. IRBM https://doi.org/10.1016/j.irbm.2019.10.006 (2019). Thank you for visiting nature.com. Use of chest ct in combination with negative rt-pcr assay for the 2019 novel coronavirus but high clinical suspicion. IEEE Trans. In my thesis project, I developed an image classification model to detect COVID-19 on chest X-ray medical data using deep learning models such . A. In the last two decades, two famous types of coronaviruses SARS-CoV and MERS-CoV had been reported in 2003 and 2012, in China, and Saudi Arabia, respectively3. The next process is to compute the performance of each solution using fitness value and determine which one is the best solution. (33)), showed that FO-MPA also achieved the best value of the fitness function compared to others. Rep. 10, 111 (2020). \(\Gamma (t)\) indicates gamma function. Comput. Also, some image transformations were applied, such as rotation, horizontal flip, and scaling. J. Med. Access through your institution. 4b, FO-MPA algorithm selected successfully fewer features than other algorithms, as it selected 130 and 86 features from Dataset 1 and Dataset 2, respectively. According to the best measure, the FO-MPA performed similarly to the HHO algorithm, followed by SMA, HGSO, and SCA, respectively. 517 PDF Ensemble of Patches for COVID-19 X-Ray Image Classification Thiago Chen, G. Oliveira, Z. Dias Medicine Feature selection based on gaussian mixture model clustering for the classification of pulmonary nodules based on computed tomography. The code of the proposed approach is also available via the following link [https://drive.google.com/file/d/1-oK-eeEgdCMCnykH364IkAK3opmqa9Rvasx/view?usp=sharing]. Figure6 shows a comparison between our FO-MPA approach and other CNN architectures. Li et al.34 proposed a self-adaptive bat algorithm (BA) to address two problems in lung X-ray images, rebalancing, and feature selection. Going deeper with convolutions. Comput. AMERICAN JOURNAL OF EMERGENCY MEDICINE COVID-19: Facemask use prevalence in international airports in Asia, Europe and the Americas, March 2020 Toaar, M., Ergen, B. Average of the consuming time and the number of selected features in both datasets. wrote the intro, related works and prepare results. Duan et al.13 applied the Gaussian mixture model (GMM) to extract features from pulmonary nodules from CT images. 2020-09-21 . As seen in Fig. where CF is the parameter that controls the step size of movement for the predator. Mobilenets: Efficient convolutional neural networks for mobile vision applications. Also, As seen in Fig. Also, image segmentation can extract critical features, including the shape of tissues, and texture5,6. A. 2. Google Scholar. All classication models ever, the virus mutates, and new variants emerge and dis- performed better in classifying the Non-COVID-19 images appear. Faramarzi et al.37 divided the agents for two halves and formulated Eqs. In54, AlexNet pre-trained network was used to extract deep features then applied PCA to select the best features by eliminating highly correlated features. }\delta (1-\delta ) U_{i}(t-1)+ \frac{1}{3! Dhanachandra and Chanu35 proposed a hybrid method of dynamic PSO and fuzzy c-means to segment two types of medical images, MRI and synthetic images. HIGHLIGHTS who: Yuan Jian and Qin Xiao from the Fukuoka University, Japan have published the Article: Research and Application of Fine-Grained Image Classification Based on Small Collar Dataset, in the Journal: (JOURNAL) what: MC-Loss drills down on the channels to effectively navigate the model, focusing on different distinguishing regions and highlighting diverse features. Finally, the sum of the features importance value on each tree is calculated then divided by the total number of trees as in Eq. J. Med. Eng. Syst. Inception architecture is described in Fig. https://www.sirm.org/category/senza-categoria/covid-19/ (2020). International Conference on Machine Learning647655 (2014). Mutation: A mutation refers to a single change in a virus's genome (genetic code).Mutations happen frequently, but only sometimes change the characteristics of the virus. The definitions of these measures are as follows: where TP (true positives) refers to the positive COVID-19 images that were correctly labeled by the classifier, while TN (true negatives) is the negative COVID-19 images that were correctly labeled by the classifier. PVT-COV19D: COVID-19 Detection Through Medical Image Classification Based on Pyramid Vision Transformer. Afzali, A., Mofrad, F.B. Faramarzi, A., Heidarinejad, M., Mirjalili, S. & Gandomi, A. H. Marine predators algorithm: a nature-inspired metaheuristic. 152, 113377 (2020). Table2 depicts the variation in morphology of the image, lighting, structure, black spaces, shape, and zoom level among the same dataset, as well as with the other dataset. Internet Explorer). implemented the FO-MPA swarm optimization and prepared the related figures and manuscript text. The second one is based on Matlab, where the feature selection part (FO-MPA algorithm) was performed. Apostolopoulos, I. D. & Mpesiana, T. A. Covid-19: automatic detection from x-ray images utilizing transfer learning with convolutional neural networks. Furthermore, the proposed GRAY+GRAY_HE+GRAY_CLAHE image representation was evaluated on two different datasets, SARS-CoV-2 CT-Scan and New_Data_CoV2, where it was found to be superior to RGB . CAS Eur. 41, 923 (2019). Propose similarity regularization for improving C. 51, 810820 (2011). In this paper, we proposed a novel COVID-19 X-ray classification approach, which combines a CNN as a sufficient tool to extract features from COVID-19 X-ray images. Detection of lung cancer on chest ct images using minimum redundancy maximum relevance feature selection method with convolutional neural networks. Duan, H. et al. The largest features were selected by SMA and SGA, respectively. The proposed COVID-19 X-ray classification approach starts by applying a CNN (especially, a powerful architecture called Inception which pre-trained on Imagnet dataset) to extract the discriminant features from raw images (with no pre-processing or segmentation) from the dataset that contains positive and negative COVID-19 images. While, MPA, BPSO, SCA, and SGA obtained almost the same accuracy, followed by both bGWO, WOA, and SMA. Deep cnns for microscopic image classification by exploiting transfer learning and feature concatenation. 92, 103662. https://doi.org/10.1016/j.engappai.2020.103662 (2020). and pool layers, three fully connected layers, the last one performs classification. Therefore, several pre-trained models have won many international image classification competitions such as VGGNet24, Resnet25, Nasnet26, Mobilenet27, Inception28 and Xception29. (24). With the help of numerous algorithms in AI, modern COVID-19 cases can be detected and managed in a classified framework. and A.A.E. Besides, the binary classification between two classes of COVID-19 and normal chest X-ray is proposed. Detecting COVID-19 at an early stage is essential to reduce the mortality risk of the patients. The authors declare no competing interests. Memory FC prospective concept (left) and weibull distribution (right). In ancient India, according to Aelian, it was . Imaging 35, 144157 (2015). In Proceedings of the IEEE Conference on computer vision and pattern recognition workshops, 806813 (2014). A combination of fractional-order and marine predators algorithm (FO-MPA) is considered an integration among a robust tool in mathematics named fractional-order calculus (FO). (9) as follows. So, based on this motivation, we apply MPA as a feature selector from deep features that produced from CNN (largely redundant), which, accordingly minimize capacity and resources consumption and can improve the classification of COVID-19 X-ray images. all above stages are repeated until the termination criteria is satisfied.