Research on the classification of CT images of COVID-19 by pre-training model based on self-attention mechanism

Nowadays, with the outbreak of COVID-19, the prevention and treatment of COVID-19 has gradually become the focus of social disease prevention, and most patients are also more concerned about the symptoms. COVID-19 has symptoms similar to the common cold, and it cannot be diagnosed based on the symptoms shown by the patient, so it is necessary to observe medical images of the lungs to finally determine whether they are COVID-19 positive. As the number of patients with symptoms similar to pneumonia increases, more and more medical images of the lungs need to be generated. At the same time, the number of physicians at this stage is far from meeting the needs of patients, resulting in patients unable to detect and understand their own conditions in time. In this regard, we have performed image augmentation, data cleaning, and designed a deep learning classification network based on the data set of COVID-19 lung medical images. accurate classification judgment. The network can achieve 95.76% classification accuracy for this task through a new fine-tuning method and hyperparameter tuning we designed, which has higher accuracy and less training time than the classic convolutional neural network model. © 2023 SPIE.

Mixup-Inf-Net: A data augmentation algorithm for segmentation of new coronary pneumonia infections

In 2020, the global spread of Coronavirus Disease 2019 exposed entire world to a severe health crisis. This has limited fast and accurate screening of suspected cases due to equipment shortages and and harsh testing environments. The current diagnosis of suspected cases has benefited greatly from the use of radiographic brain imaging, also including X-ray and scintigraphy, as a crucial addition to screening tests for new coronary pneumonia disease. However, it is impractical to gather enormous volumes of data quickly, which makes it difficult for depth models to be trained. To solve these problems, we obtained a new dataset by data augmentation Mixup method for the used chest CT slices. It uses lung infection segmentation (Inf-Net [1]) in a deep network and adds a learning framework with semi-supervised to form a Mixup-Inf-Net semi-supervised learning framework model to identify COVID-19 infection area from chest CT slices. The system depends primarily on unlabeled data and merely a minimal amount of annotated data is required;therefore, the unlabeled data generated by Mixup provides good assistance. Our framework can be used to improve improve learning and performance. The SemiSeg dataset and the actual 3D CT images that we produced are used in a variety of tests, and the analysis shows that Mixup-Inf-Net semi-supervised outperforms most SOTA segmentation models learning framework model in this study, which also enhances segmentation performance. © 2023 SPIE.

COVID-19 Image Segmentation Algorithms Based on Conditional Generative Adversarial Network

The size and location of the lesions in CT images of novel corona virus pneumonia (COVID-19) change all the time, and the lesion areas have low contrast and blurred boundaries, resulting in difficult segmentation. To solve this problem, a COVID-19 image segmentation algorithm based on conditional generative adversarial network (CGAN) is proposed. Uses the improved DeeplabV3+ network as a generator, which enhances the extraction of multi-scale contextual features, reduces the number of network parameters and improves the training speed. A Markov discriminator with 6 fully convolutional layers is proposed instead of a common discriminator, with the aim of focusing more on the local features of the CT image. By continuously adversarial training between the generator and the discriminator, the network weights are optimised so that the final segmented image generated by the generator is infinitely close to the ground truth. On the COVID-19 CT public dataset, the area under the curve of ROC, F1-Score and dice similarity coefficient achieved 96.64%, 84.15% and 86.14% respectively. The experimental results show that the proposed algorithm is accurate and robust, and it has the possibility of becoming a safe, inexpensive, and time-saving medical assistant tool in clinical diagnosis, which provides a reference for computer-aided diagnosis. © 2022 ACM.

Semi-Supervised Modified-UNet for Lung Infection Image Segmentation

Automatic lung infection segmentation in computed tomography (CT) scans can offer great assistance in radiological diagnosis by improving accuracy and reducing time required for diagnosis. The biggest challenges for deep learning (DL) models in segmenting infection region are the high variances in infection characteristics, fuzzy boundaries between infected and normal tissues, and the troubles in getting large number of annotated data for training. To resolve such issues, we propose a Modified U-Net (Mod-UNet) model with minor architectural changes and significant modifications in the training process of vanilla 2D UNet. As part of these modifications, we updated the loss function, optimization function, and regularization methods, added a learning rate scheduler and applied advanced data augmentation techniques. Segmentation results on two Covid-19 Lung CT segmentation datasets show that the performance of Mod-UNet is considerably better than the baseline U-Net. Furthermore, to mitigate the issue of lack of annotated data, the Mod-UNet is used in a semi-supervised framework (Semi-Mod-UNet) which works on a random sampling approach to progressively enlarge the training dataset from a large pool of unannotated CT slices. Exhaustive experiments on the two Covid-19 CT segmentation datasets and on a real lung CT volume show that the Mod-UNet and Semi-Mod-UNet significantly outperform other state-of-theart approaches in automated lung infection segmentation. IEEE

Multi-Joint Unit Network for COVID-19 Diagnosis using X-Ray and CT Images

The global COVID-19 pandemic has caused a health crisis globally. Automated diagnostic methods can control the spread of the pandemic, as well as assists physicians to tackle high workload conditions through the quick treatment of affected patients. Owing to the scarcity of medical images and from different resources, the present image heterogeneity has raised challenges for achieving effective approaches to network training and effectively learning robust features. We propose a multi-joint unit network for the diagnosis of COVID-19 using the joint unit module, which leverages the receptive fields from multiple resolutions for learning rich representations. Existing approaches usually employ a large number of layers to learn the features, which consequently requires more computational power and increases the network complexity. To compensate, our joint unit module extracts low-, same-, and high-resolution feature maps simultaneously using different phases. Later, these learned feature maps are fused and utilized for classification layers. We observed that our model helps to learn sufficient information for classification without a performance loss and with faster convergence. We used three public benchmark datasets to demonstrate the performance of our network. Our proposed network consistently outperforms existing state-of-the-art approaches by demonstrating better accuracy, sensitivity, and specificity and F1-score across all datasets. © 2022 ACM.

Towards Lifting the Trade-off between Accuracy and Adversarial Robustness of Deep Neural Networks with Application on COVID 19 CT Image Classification and Medical Image Segmentation

Deep neural networks (DNNs) are vulnerable to adversarial noises. Adversarial training is a general strategy to improve DNN robustness. But training a DNN model with adversarial noises may result in a much lower accuracy on clean data, which is termed the trade-off between accuracy and adversarial robustness. Towards lifting this trade-off, we propose an adversarial training method that generates optimal adversarial training samples. We evaluate our methods on PathMNIST and COVID-19 CT image classification tasks, where the DNN model is ResNet-18, and Heart MRI and Prostate MRI image segmentation tasks, where the DNN model is nnUnet. All these four datasets are publicly available. The experiment results show that our method has the best robustness against adversarial noises and has the least accuracy degradation compared to the other defense methods. © 2023 SPIE.

COVID-19 Instance Segmentation Network Based on Semantic Information Generating Proposals

The CT images of lungs of COVID-19 patients have distinct pathological features, segmenting the lesion area accurately by the method of deep learning, which is of great significance for the diagnosis and treatment of COVID-19 patients. Instance segmentation has higher sensitivity and can output the Bounding Boxes of the lesion region, however, the traditional instance segmentation method is weak in the segmentation of small lesions, and there is still room for improvement in the segmentation accuracy. We propose a instance segmentation network which is called as Semantic R-CNN. Firstly, a semantic segmentation branch is added on the basis of Mask-RCNN, and utilizing the image processing tool Skimage in Python to label the connected domain for the result of semantic segmentation, extracting the rectangular boundaries of connected domain and using them as Proposals, which will replace the Regional Proposal Network in the instance segmentation. Secondly, the Atrous Spatial Pyramid Pooling is introduced into the Feature Pyramid Network, then improving the feature fusion method in FPN. Finally, the cascade method is introduced into the detection branch of the network to optimize the Proposals. Segmentation experiments were carried out on the pathological lesion segmentation data set of CC-CCII, the average accuracy of the semantic segmentation is 40.56mAP, and compared with the Mask-RCNN, it has improved by 9.98mAP. After fusing the results of semantic segmentation and instance segmentation, the Dice coefficient is 80.7%, the sensitivity is 85.8%, and compared with the Inf-Net, it has increased by 1.6% and 8.06% respectively. The proposed network has improved the segmentation accuracy and reduced the false-negatives. © 2022 ACM.

LLDSNet: Limited labeled data segmentation network for COVID-19 CT lesion segmentation

With the COVID-19 outbreak in 2019, the world is facing a major crisis and people's health is at serious risk. Accurate segmentation of lesions in CT images can help doctors understand disease infections, prescribe the right medicine and control patients' conditions. Fast and accurate diagnosis not only can make the limited medical resources get reasonable allocation, but also can control the spread of disease, and computer-aided diagnosis can achieve this purpose, so this paper proposes a deep learning segmentation network LLDSNet based on a small amount of data, which is divided into two modules: contextual feature-aware module (CFAM) and shape edge detection module (SEDM). Due to the different morphology of lesions in different CT, lesions with dispersion, small lesion area and background area imbalance, lesion area and normal area boundary blurred, etc. The problem of lesion segmentation in COVID-19 poses a major challenge. The CFAM can effectively extract the overall and local features, and the SEDM can accurately find the edges of the lesion area to segment the lesions in this area. The hybrid loss function is used to avoid the class imbalance problem and improve the overall network performance. It is demonstrated that LLDSNet dice achieves 0.696 for a small number of data sets, and the best performance compared to five currently popular segmentation networks. © 2023 SPIE.

Classic Augmentation based Classifier Generative Adversarial Network (CACGAN) for COVID-19 Diagnosis

Computer-aided diagnosis of COVID-19 from lung medical images has received increasing attention in previous clinical practice and research. However, developing such automatic model is usually challenging due to the requirement of a large amount of data and sufficient computer power. With only 317 training images, this paper presents a Classic Augmentation based Classifier Generative Adversarial Network (CACGAN) for data synthetising. In order to take into account, the feature extraction ability and lightness of the model for lung CT images, the CACGAN network is mainly constructed by convolution blocks. During the training process, each iteration will update the discriminator's network parameters twice and the generator's network parameters once. For the evaluation of CACGAN. This paper organized multiple comparison between each pair from CACGAN synthetic data, classic augmented data, and original data. In this paper, 7 classifiers are built, ranging from simple to complex, and are trained for the three sets of data respectively. To control the variable, the three sets of data use the exact same classifier structure and the exact same validation dataset. The result shows the CACGAN successfully learned how to synthesize new lung CT images with specific labels. © 2023 SPIE.

Transfer Learning Model in Deep Neural Network for COVID-19 Prediction from CT Images

Deep learning models are often used to process radi-ological images automatically and can accurately train networks' weights on appropriate datasets. One of the significant benefits of the network is that it is possible to use the weight of a pre-trained network for other applications by fine-tuning the current weight. The primary purpose of this work is to employ a pre-trained deep neural framework known as transfer learning to detect and diagnose COVID-19 in CT images automatically. This paper uses a popular deep neural model, ResNet152, as a neural transfer approach. The presented framework uses the weight obtained from the ImageNet dataset, fine-tuned by the dataset used in the work. The effectiveness of the suggested COVID-19 prediction system is evaluated experimentally and compared with DenseNet, another transfer learning model. The recommended ResNet152 transfer learning model exhibits improved performance and has a 99% accuracy when analogized with the DenseNet201 transfer learning model. © 2022 IEEE.

Hybrid-Patch-Alex: A new patch division and deep feature extraction-based image classification model to detect COVID-19, heart failure, and other lung conditions using medical images

COVID-19, chronic obstructive pulmonary disease (COPD), heart failure (HF), and pneumonia can lead to acute respiratory deterioration. Prompt and accurate diagnosis is crucial for effective clinical management. Chest X-ray (CXR) and chest computed tomography (CT) are commonly used for confirming the diagnosis, but they can be time-consuming and biased. To address this, we developed a computationally efficient deep feature engineering model called Hybrid-Patch-Alex for automated COVID-19, COPD, and HF diagnosis. We utilized one CXR dataset and two CT image datasets, including a newly collected dataset with four classes: COVID-19, COPD, HF, and normal. Our model employed a hybrid patch division method, transfer learning with pre-trained AlexNet, iterative neighborhood component analysis for feature selection, and three standard classifiers (k-nearest neighbor, support vector machine, and artificial neural network) for automated classification. The model achieved high accuracy rates of 99.82%, 92.90%, and 97.02% on the respective datasets, using kNN and SVM classifiers.

Improved Residual U-Net for COVID-19 Lung Infection Multi-Class Segmentation in CT Image

COVID-19, the new coronavirus, is a threat to global public health. Today, there is an urgent need for automatic COVID-19 infection detection tools. This work proposes an automatic COVID-19 infection detection system based on CT image segmentation. A deep learning network developed from an improved Residual U-net architecture extracts infected areas from a CT lung image. We tested the system on COVID-19 public CT images. An evaluation using the F1 score, sensitivity, specificity and accuracy proved the effectiveness of the proposed network. Besides, experimental results showed that the proposed network performed well in extracting infection regions so, it can assist experts in COVID-19 infection detection. © 2022 IEEE.

Mathematical Morphology Adopted Automatic Detection of Ground Glass Opacities in Lung CT Images of COVID-19 Patients

The severity of the nCOVID infection relies on the presence of Ground Glass Opacities (GGO) present in the patient's chest CT scan images. Although, detecting and delineating the precise boundaries of GGO in the chest CT images is challenging. Here, we proposed a fast and novel technique to automatically segment the regions containing GGO in lung CT images using mathematical morphology. We have tested our algorithm on the chest CT images of 145 Covid-positive cases. This unique segmentation approach correctly segments the lung field from chest CT images and identifies GGO with average sensitivity, specificity, and accuracy of 96.89%, 95.23%, and 97.22%, respectively. We used expert radiologists' hand-curated segmentation of GGO as ground truth for quantificational performance analysis. Our research results indicate that this algorithm performs well found in the literature. © 2023 IEEE.

CdcSegNet: Automatic COVID-19 Infection Segmentation from CT Images

It has been more than two years since the outbreak of COVID-19, which has spread to almost every corner of the world and killed a great number of people. Rapid detection and screening have become an important means of controlling the spread of COVID-19. Segmentation of COVID-19 infected tissue from computed tomography (CT) images of a patient’s lungs can provide clinicians with important information to quantify and diagnose COVID-19. However, the accuracy of medical image segmentation is seriously affected by such factors as the low contrast between the infected tissue and the edge of the surrounding environment, the large variation of the infected tissue and the lack of labeling data. Therefore, a deep learning model called CdcSegNet to accurately segment lung lesions from CT images infected by COVID-19 is proposed. In our method, transfer learning is introduced to solve the problem of lack of annotation data, and three modules, i.e., continuous dilated convolution module (CDC), parallel dual attention module (PDA) and additional multi-core pooling layer (AMP) are innovatively proposed to solve the problem of fuzzy segmentation boundary and to segment effectively infected tissues. Extensive experiments and comparison studies are made, and demonstrate that our model CdcSegNet has high accuracy in COVID-19 segmentation, and is superior to the state-of-the-art models in terms of DICE, SEN, SPE, PPV, and VOE. IEEE

Testing and Verifying the Security of COVID-19 CT Images Deep Learning System with Adversarial Attack

The Coronavirus disease 2019 (COVID-19) is a pandemic that occurred in December 2019 and spread globally. Most of the current research is on how to apply deep learning to detect COVID-19, but little research has been done on the security of COVID-19 deep learning systems. Therefore, we test and verify the security of COVID-19 CT images deep learning system with adversarial attack. Firstly, we build a deep learning system for recognizing COVID-19 CT images. Secondly, adding imperceptible disturbance to CT images will lead to neural network classification errors. Finally, we discuss the application of formal methods and formal verification to deep learning systems. We hope to draw more attention from researchers to the application of formal methods and formal verification to artificial intelligence. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

Federated Learning for Detecting COVID-19 in Chest CT Images: A Lightweight Federated Learning Approach

The novel coronavirus is spreading rapidly worldwide, and finding an effective and rapid diagnostic method is apriority. Medical data involves patient privacy, and the centralized collection of large amounts of medical data is impossible. Federated learning is a privacy-preserving machine learning paradigm that can be well applied to smart healthcare by coordinating multiple hospitals to perform deep learning training without transmitting data. This paper demonstrates the feasibility of a federated learning approach for detecting COVID-19 through chest CT images. We propose a lightweight federated learning method that normalizes the local training process by globally averaged feature vectors. In the federated training process, the models' parameters do not need to be transmitted, and the local client only uploads the average of the feature vectors of each class. Clients can choose different local models according to their computing capabilities. We performed a comprehensive evaluation using various deep-learning models on COVID-19 chest CT images. The results show that our approach can effectively reduce the communication load of federated learning while having high accuracy for detecting COVID-19 on chest CT images. © 2022 IEEE.

Analysis of High-Resolution CT Images of COVID-19 Patients

This research work is carried out to quantify the COVID-19 disease and to explore whether the quantitative can be used to analyze the survivability of the patient during admission. In this method, a novel percentage split distribution (PSD), thresholding-based image segmentation method is proposed to quantify normal and lesion regions by analyzing the benign GGOs. The method segments the lung-CT image based on pixel distribution. The segmented regions are quantified as a fraction of region of interest with total number of pixels. The study is also extended to analyze the left and right lungs separately with some common findings on lesion distribution involved with COVID-19 disease. The performance of PSD method has been compared with two traditional image segmentation-based methods. From the results, it has been observed that the segments created by the PSD method are better than experimental methods and clearly identify the margins of lesion and normal regions. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

Identification of COVID-19 in CT Images using Deep Learning Model

Worldwide, COVID-19 has had a substantial impact on patients and hospital systems. Early identification and diagnosis are essential for regulating the growth of COVID-19. The input CT screening images are initially segmented into various regions using the Fuzzy C-means (FCM) clustering technique. Next, region-based image quality enhancement employs a histogram equalization method. Furthermore, certain necessary data is represented in a new image using the Local Directional Number technique. Lastly, the input images are portioned with the help of a traditional convolutional neural network model. The proposed convolutional neural network based system was able to give an accuracy of 98.60%, and the results revealed that methods for detecting COVID-19 impact from CT scan images must be developed significantly before considering it as a medical choice. Moreover, many diverse datasets are essential to assess the processes in a real-world setting. © 2022 IEEE.

Automated diagnosis of COVID stages using texture-based Gabor features in variational mode decomposition from CT images

COVID-19 is a deadly and fast-spreading disease that makes early death by affecting human organs, primarily the lungs. The detection of COVID in the early stages is crucial as it may help restrict the spread of the progress. The traditional and trending tools are manual, time-inefficient, and less accurate. Hence, an automated diagnosis of COVID is needed to detect COVID in the early stages. Recently, several methods for exploiting computed tomography (CT) scan pictures to detect COVID have been developed;however, none are effective in detecting COVID at the preliminary phase. We propose a method based on two-dimensional variational mode decomposition in this work. This proposed approach decomposes pre-processed CT scan pictures into sub-bands. The texture-based Gabor filter bank extracts the relevant features, and the student's t-value is used to recognize robust traits. After that, linear discriminative analysis (LDA) reduces the dimensionality of features and provides ranks for robust features. Only the first 14 LDA features are qualified for classification. Finally, the least square- support vector machine (SVM) (radial basis function) classifier distinguishes between COVID and non-COVID CT lung images. The results of the trial showed that our model outperformed cutting-edge methods for COVID classification. Using tenfold cross-validation, this model achieved an improved classification accuracy of 93.96%, a specificity of 95.59%, and an F1 score of 93%. To validate our proposed methodology, we conducted different relative experiments with deep learning and traditional machine learning-based models like random forest, K-nearest neighbor, SVM, convolutional neural network, and recurrent neural network. The proposed model is ready to help radiologists identify diseases daily. © 2023 Wiley Periodicals LLC.

Applied Enhanced Q-NAS for COVID-19 Detection in CT Images

The Deep Neural Networks are flexible and robust models that have gained attention from the machine learning community over the last decade. During the construction of a neural network, an expert can spend significant time designing a neural architecture with trial and error sessions. Because of the manual process, there is a greater interest in Neural Architecture Search (NAS), which is an automated method of architectural search in neural networks. Quantum-inspired evolutionary algorithms present propitious results regarding faster convergence when compared to other solutions with restricted search space and high computational costs. In this work, we enhance the Q-NAS model: a quantum-inspired algorithm to search for deep networks by assembling substructures. We present a new architecture that was designed automatically by the Q-NAS and applied to a case study for COVID-19 vs. healthy classification. For this classification, the Q-NAS algorithm was able to find a network architecture with only 1.23 M parameters that reached the accuracy of 99.44%, which overcame benchmark networks like Inception (GoogleLeNet), EfficientNet and VGG that were also tested in this work. The algorithm is publicly avaiable at https://github.com/julianoce/qnas. © 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.