Pneumonia detection in chest X-ray images using an ensemble of deep learning models

PLoS ONE, Journal Year: 2021, Volume and Issue: 16(9), P. e0256630 - e0256630

Published: Sept. 7, 2021

Pneumonia is a respiratory infection caused by bacteria or viruses; it affects many individuals, especially in developing and underdeveloped nations, where high levels of pollution, unhygienic living conditions, overcrowding are relatively common, together with inadequate medical infrastructure. causes pleural effusion, condition which fluids fill the lung, causing difficulty. Early diagnosis pneumonia crucial to ensure curative treatment increase survival rates. Chest X-ray imaging most frequently used method for diagnosing pneumonia. However, examination chest X-rays challenging task prone subjective variability. In this study, we developed computer-aided system automatic detection using images. We employed deep transfer learning handle scarcity available data designed an ensemble three convolutional neural network models: GoogLeNet, ResNet-18, DenseNet-121. A weighted average technique was adopted, wherein weights assigned base learners were determined novel approach. The scores four standard evaluation metrics, precision, recall, f1-score, area under curve, fused form weight vector, studies literature set experimentally, that error. proposed approach evaluated on two publicly datasets, provided Kermany et al. Radiological Society North America (RSNA), respectively, five-fold cross-validation scheme. achieved accuracy rates 98.81% 86.85% sensitivity 98.80% 87.02% RSNA respectively. results superior those state-of-the-art methods our performed better than widely techniques. Statistical analyses datasets McNemar's ANOVA tests showed robustness codes work at https://github.com/Rohit-Kundu/Ensemble-Pneumonia-Detection.

Language: Английский

---

A Comparative Performance Analysis of Data Resampling Methods on Imbalance Medical Data

IEEE Access, Journal Year: 2021, Volume and Issue: 9, P. 109960 - 109975

Published: Jan. 1, 2021

Medical datasets are usually imbalanced, where negative cases severely outnumber positive cases. Therefore, it is essential to deal with this data skew problem when training machine learning algorithms. This study uses two representative lung cancer datasets, PLCO and NLST, imbalance ratios (the proportion of samples in the majority class those minority class) 24.7 25.0, respectively, predict incidence. research performance 23 methods (resampling hybrid systems) three classical classifiers (logistic regression, random forest, LinearSVC) identify best techniques suitable for medical datasets. Resampling includes ten under-sampling (RUS, etc.), seven over-sampling (SMOTE, integrated sampling (SMOTEENN, SMOTE-Tomek). Hybrid systems include (Balanced Bagging, etc.). The results show that can improve classification ability model. Compared other imbalanced techniques, have highest standard deviation (SD), lowest SD. Over-sampling a stable method, AUC model generally higher than ways. Using ROS, forest performs predictive more used study. code available at https://mkhushi.github.io/.

Language: Английский

---

Deep-learning based detection of COVID-19 using lung ultrasound imagery

PLoS ONE, Journal Year: 2021, Volume and Issue: 16(8), P. e0255886 - e0255886

Published: Aug. 13, 2021

Background The COVID-19 pandemic has exposed the vulnerability of healthcare services worldwide, especially in underdeveloped countries. There is a clear need to develop novel computer-assisted diagnosis tools provide rapid and cost-effective screening places where massive traditional testing not feasible. Lung ultrasound portable, easy disinfect, low cost non-invasive tool that can be used identify lung diseases. Computer-assisted analysis imagery relatively recent approach shown great potential for diagnosing pulmonary conditions, being viable alternative COVID-19. Objective To evaluate compare performance deep-learning techniques detecting infections from imagery. Methods We adapted different pre-trained deep learning architectures, including VGG19, InceptionV3, Xception, ResNet50. publicly available POCUS dataset comprising 3326 frames healthy, COVID-19, pneumonia patients training fine-tuning. conducted two experiments considering three classes (COVID-19, pneumonia, healthy) (COVID-19 versus non-COVID-19) predictive models. obtained results were also compared with POCOVID-net model. For evaluation, we calculated per-class classification metrics (Precision, Recall, F1-score) overall (Accuracy, Balanced Accuracy, Area Under Receiver Operating Characteristic Curve). Lastly, performed statistical using ANOVA Friedman tests followed by post-hoc Wilcoxon signed-rank test Holm’s step-down correction. Results InceptionV3 network achieved best average accuracy (89.1%), balanced (89.3%), area under receiver operating curve (97.1%) detection bacterial healthy data. found statistically significant differences between models accuracy, curve. Post-hoc showed InceptionV3-based model POCOVID-net, VGG19-, ResNet50-based No InceptionV3- Xception-based Conclusions Deep promising avenue diagnosis. Particularly, provides most all AI-based evaluated this work. further based on

Language: Английский

---

High-precision multiclass classification of lung disease through customized MobileNetV2 from chest X-ray images

Computers in Biology and Medicine, Journal Year: 2023, Volume and Issue: 155, P. 106646 - 106646

Published: Feb. 10, 2023

In this study, multiple lung diseases are diagnosed with the help of Neural Network algorithm. Specifically, Emphysema, Infiltration, Mass, Pleural Thickening, Pneumonia, Pneumothorax, Atelectasis, Edema, Effusion, Hernia, Cardiomegaly, Pulmonary Fibrosis, Nodule, and Consolidation, studied from ChestX-ray14 dataset. A proposed fine-tuned MobileLungNetV2 model is employed for analysis. Initially, pre-processing done on X-ray images dataset using CLAHE to increase image contrast. Additionally, a Gaussian Filter, denoise images, data augmentation methods used. The pre-processed fed into several transfer learning models; such as InceptionV3, AlexNet, DenseNet121, VGG19, MobileNetV2. Among these models, MobileNetV2 performed highest accuracy 91.6% in overall classifying lesions Chest Images. This then optimise model. On data, model, MobileLungNetV2, achieves an extraordinary classification 96.97%. Using confusion matrix all classes, it determined that has high precision, recall, specificity scores 96.71%, 96.83% 99.78% respectively. study employs Grad-cam output determine heatmap disease detection. shows promising results images.

Language: Английский

---

A hybrid explainable ensemble transformer encoder for pneumonia identification from chest X-ray images

Journal of Advanced Research, Journal Year: 2022, Volume and Issue: 48, P. 191 - 211

Published: Sept. 7, 2022

Pneumonia is a microorganism infection that causes chronic inflammation of the human lung cells. Chest X-ray imaging most well-known screening approach used for detecting pneumonia in early stages. While chest-Xray images are mostly blurry with low illumination, strong feature extraction required promising identification performance. A new hybrid explainable deep learning framework proposed accurate disease using chest images. The workflow developed by fusing capabilities both ensemble convolutional networks and Transformer Encoder mechanism. backbone to extract features from raw input two different scenarios: (i.e., DenseNet201, VGG16, GoogleNet) B InceptionResNetV2, Xception). Whereas, built based on self-attention mechanism multilayer perceptron (MLP) identification. visual saliency maps derived emphasize crucial predicted regions end-to-end training process models over all scenarios performed binary multi-class classification scenarios. model recorded 99.21% performance terms overall accuracy F1-score task, while it achieved 98.19% 97.29% multi-classification task. For scenario, 97.22% 97.14% F1-score, 96.44% F1-score. multiclass 97.2% 95.8% 96.4% 94.9% could provide encouraging comparing individual, models, or even latest AI literature. code available here: https://github.com/chiagoziemchima/Pneumonia_Identificaton.

Language: Английский

---

LungNet22: A Fine-Tuned Model for Multiclass Classification and Prediction of Lung Disease Using X-ray Images

Journal of Personalized Medicine, Journal Year: 2022, Volume and Issue: 12(5), P. 680 - 680

Published: April 24, 2022

In recent years, lung disease has increased manyfold, causing millions of casualties annually. To combat the crisis, an efficient, reliable, and affordable diagnosis technique become indispensable. this study, a multiclass classification from frontal chest X-ray imaging using fine-tuned CNN model is proposed. The conducted on 10 classes lungs, namely COVID-19, Effusion, Tuberculosis, Pneumonia, Lung Opacity, Mass, Nodule, Pneumothorax, Pulmonary Fibrosis, along with Normal class. dataset collective gathered multiple sources. After pre-processing balancing eight augmentation techniques, total 80,000 images were fed to for purposes. Initially, pre-trained models, AlexNet, GoogLeNet, InceptionV3, MobileNetV2, VGG16, ResNet 50, DenseNet121, EfficientNetB7, employed dataset. Among these, VGG16 achieved highest accuracy at 92.95%. further improve accuracy, LungNet22 was constructed upon primary structure model. An ablation study used in work determine different hyper-parameters. Using Adam Optimizer, proposed commendable 98.89%. verify performance model, several matrices, including ROC curve AUC values, computed as well.

Language: Английский

---

A Review of Machine Learning Algorithms for Biomedical Applications

Annals of Biomedical Engineering, Journal Year: 2024, Volume and Issue: 52(5), P. 1159 - 1183

Published: Feb. 21, 2024

Language: Английский

---

Efficient pneumonia detection using Vision Transformers on chest X-rays

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: Jan. 30, 2024

Abstract Pneumonia is a widespread and acute respiratory infection that impacts people of all ages. Early detection treatment pneumonia are essential for avoiding complications enhancing clinical results. We can reduce mortality, improve healthcare efficiency, contribute to the global battle against disease has plagued humanity centuries by devising deploying effective methods. Detecting not only medical necessity but also humanitarian imperative technological frontier. Chest X-rays frequently used imaging modality diagnosing pneumonia. This paper examines in detail cutting-edge method detecting implemented on Vision Transformer (ViT) architecture public dataset chest available Kaggle. To acquire context spatial relationships from X-ray images, proposed framework deploys ViT model, which integrates self-attention mechanisms transformer architecture. According our experimentation with Transformer-based framework, it achieves higher accuracy 97.61%, sensitivity 95%, specificity 98% X-rays. The model preferable capturing context, comprehending relationships, processing images have different resolutions. establishes its efficacy as robust solution surpassing convolutional neural network (CNN) based architectures.

Language: Английский

---

Adversarial Attack and Defence through Adversarial Training and Feature Fusion for Diabetic Retinopathy Recognition

Sensors, Journal Year: 2021, Volume and Issue: 21(11), P. 3922 - 3922

Published: June 7, 2021

Due to the rapid growth in artificial intelligence (AI) and deep learning (DL) approaches, security robustness of deployed algorithms need be guaranteed. The susceptibility DL adversarial examples has been widely acknowledged. artificially created will lead different instances negatively identified by models that are humanly considered benign. Practical application actual physical scenarios with threats shows their features. Thus, attacks defense, including machine its reliability, have drawn growing interest and, recent years, a hot topic research. We introduce framework provides defensive model against speckle-noise attack, training, feature fusion strategy, which preserves classification correct labelling. evaluate analyze defenses on retinal fundus images for Diabetic Retinopathy recognition problem, is state-of-the-art endeavor. Results obtained images, prone attacks, 99% accurate prove proposed robust.

Language: Английский

---

VGG19 Network Assisted Joint Segmentation and Classification of Lung Nodules in CT Images

Diagnostics, Journal Year: 2021, Volume and Issue: 11(12), P. 2208 - 2208

Published: Nov. 26, 2021

Pulmonary nodule is one of the lung diseases and its early diagnosis treatment are essential to cure patient. This paper introduces a deep learning framework support automated detection nodules in computed tomography (CT) images. The proposed employs VGG-SegNet supported mining pre-trained DL-based classification detection. CT images implemented using attained features, then these features serially concatenated with handcrafted such as Grey Level Co-Occurrence Matrix (GLCM), Local-Binary-Pattern (LBP) Pyramid Histogram Oriented Gradients (PHOG) enhance disease accuracy. used for experiments collected from LIDC-IDRI Lung-PET-CT-Dx datasets. experimental results show that VGG19 architecture can achieve an accuracy 97.83% SVM-RBF classifier.

Language: Английский

---