Detection of pneumonia using convolutional neural networks and deep learning

Journal of Applied Biomedicine, Journal Year: 2022, Volume and Issue: 42(3), P. 1012 - 1022

Published: July 1, 2022

The objective and automated detection of pneumonia represents a serious challenge in medical imaging, because the signs illness are not obvious CT or X-ray scans. Further on, it is also an important task, since millions people die every year. main goal this paper to propose solution for above mentioned problem, using novel deep neural network architecture. proposed novelty consists use dropout convolutional part network. method was trained tested on set 5856 labeled images available at one Kaggle’s many imaging challenges. chest (anterior-posterior) were selected from retrospective cohorts pediatric patients, aged between five years, Guangzhou Women Children’s Medical Center, Guangzhou, China. Results achieved by our would have placed first Kaggle competition with following metrics: 97.2% accuracy, 97.3% recall, 97.4% precision AUC=0.982, they competitive current state-of-the-art solutions.

Language: Английский

---

A comprehensive survey of deep learning research on medical image analysis with focus on transfer learning

Clinical Imaging, Journal Year: 2022, Volume and Issue: 94, P. 18 - 41

Published: Nov. 12, 2022

Language: Английский

---

Voltammetric sensor based on bimetallic nanocomposite for determination of favipiravir as an antiviral drug

Microchimica Acta, Journal Year: 2021, Volume and Issue: 188(12)

Published: Nov. 27, 2021

Language: Английский

---

Fusion of convolutional neural networks based on Dempster–Shafer theory for automatic pneumonia detection from chest X‐ray images

International Journal of Imaging Systems and Technology, Journal Year: 2021, Volume and Issue: 32(2), P. 658 - 672

Published: Sept. 13, 2021

Abstract Deep learning‐based applications for disease detection are essential tools experts to effectively diagnose diseases at different stages. In this article, a new approach based on an evidence fusion theory is proposed, allowing the combination of set deep learning classifiers provide more accurate results. The main contribution work application Dempster–Shafer five pre trained convolutional neural networks including VGG16, Xception, InceptionV3, ResNet50, and DenseNet201 diagnosis pneumonia from chest X‐ray images. To evaluate approach, experiments conducted using publicly available dataset containing than 5800 obtained results demonstrate that our provides excellent performance compared other state‐of‐the‐art methods; it achieves precision 97.5%, recall 98%, f1‐score 97.8%, accuracy 97.3%.

Language: Английский

---

Uncertainty-driven ensembles of multi-scale deep architectures for image classification

Information Fusion, Journal Year: 2022, Volume and Issue: 89, P. 53 - 65

Published: Aug. 13, 2022

The use of automatic systems for medical image classification has revolutionized the diagnosis a high number diseases. These alternatives, which are usually based on artificial intelligence (AI), provide helpful tool clinicians, eliminating inter and intra-observer variability that diagnostic process entails. Convolutional Neural Network (CNNs) have proved to be an excellent option this purpose, demonstrating large performance in wide range contexts. However, it is also extremely important quantify reliability model's predictions order guarantee confidence classification. In work, we propose multi-level ensemble system Bayesian Deep Learning approach maximize while providing uncertainty each decision. This combines information extracted from different architectures by weighting their results according predictions. Performance evaluated real scenarios: first one, aim differentiate between pulmonary pathologies: controls vs bacterial pneumonia viral pneumonia. A two-level decision tree employed divide 3-class into two binary classifications, yielding accuracy 98.19%. second context, assessed Parkinson's disease, leading 95.31%. reduced preprocessing needed obtaining performance, addition provided about evidence applicability used as aid clinicians.

Language: Английский

---

A novel deep neural network model based Xception and genetic algorithm for detection of COVID-19 from X-ray images

Annals of Operations Research, Journal Year: 2022, Volume and Issue: 328(1), P. 617 - 641

Published: Dec. 25, 2022

Language: Английский

---

MBi-GRUMCONV: A novel Multi Bi-GRU and Multi CNN-Based deep learning model for social media sentiment analysis

Journal of Cloud Computing Advances Systems and Applications, Journal Year: 2023, Volume and Issue: 12(1)

Published: Jan. 10, 2023

Abstract Today, internet and social media is used by many people, both for communication expressing opinions about various topics in domains of life. Various artificial intelligence technologies-based approaches on analysis these have emerged natural language processing the name different tasks. One tasks Sentiment analysis, which a popular method aiming task analyzing people’s provides powerful tool making decisions companies, governments, researchers. It desired to investigate effect using multi-layered neural networks together performance model be developed sentiment task. In this study, new, deep learning-based was proposed IMDB movie reviews dataset. This performs classification vectorized two methods Word2Vec, namely, Skip Gram Continuous Bag Words, three vector sizes (100, 200, 300), with help 6 Bidirectional Gated Recurrent Units 2 Convolution layers (MBi-GRUMCONV). experiments conducted model, dataset split into 80%-20% 70%-30% training-test sets, 10% training splits were validation purposes. Accuracy F1 score criteria evaluate performance. The 95.34% accuracy has outperformed studies literature. As result experiments, it found that better contribution success.

Language: Английский

---

AUTO-HAR: An adaptive human activity recognition framework using an automated CNN architecture design

Heliyon, Journal Year: 2023, Volume and Issue: 9(2), P. e13636 - e13636

Published: Feb. 1, 2023

Convolutional neural networks (CNNs) have demonstrated exceptional results in the analysis of time- series data when used for Human Activity Recognition (HAR). The manual design such architectures is an error-prone and time-consuming process. search optimal CNN considered a revolution networks. By means Neural Architecture Search (NAS), network can be designed optimized automatically. Thus, architecture representation found automatically because its ability to overcome limitations human experience thinking modes. Evolution algorithms, which are derived from evolutionary mechanisms as natural selection genetics, been widely employed develop optimize NAS they handle blackbox optimization process designing appropriate solution representations paradigms without explicit mathematical formulations or gradient information. Genetic algorithm (GA) find near-optimal solutions difficult problems. Considering these characteristics, efficient activity recognition (AUTO-HAR) presented this study. Using GA select architecture, current study proposes novel encoding schema structure space with much broader range operations effectively best HAR tasks. In addition, proposed provides reasonable degree depth it does not limit maximum length devised task architecture. To test effectiveness framework tasks, three datasets were utilized: UCI-HAR, Opportunity, DAPHNET. Based on study, has that method efficiently recognize average accuracy 98.5% (∓1.1), 98.3%, 99.14% (∓0.8) DAPHNET, respectively.

Language: Английский

---

Comparative investigation of physical, X-ray and neutron radiation shielding properties for B2O3-MnO2-CdO borate glasses

Ceramics International, Journal Year: 2023, Volume and Issue: 49(19), P. 30915 - 30923

Published: July 17, 2023

Language: Английский

---

COVID-19-The Role of Artificial Intelligence, Machine Learning, and Deep Learning: A Newfangled

Archives of Computational Methods in Engineering, Journal Year: 2023, Volume and Issue: 30(4), P. 2667 - 2682

Published: Jan. 17, 2023

Language: Английский

---