Use deep transfer learning for efficient time-series updating of subsurface flow surrogate model

Engineering Applications of Artificial Intelligence, Год журнала: 2025, Номер 153, С. 110873 - 110873

Опубликована: Апрель 18, 2025

Язык: Английский

---

COVID-19 detection from Chest X-ray images using a novel lightweight hybrid CNN architecture

Multimedia Tools and Applications, Год журнала: 2024, Номер unknown

Опубликована: Май 21, 2024

Язык: Английский

---

Automatic Detection of COVID-19 from Chest X-Ray Images Using EfficientNet-B7 CNN Model with Channel-wise Attention

International Journal of Computing and Digital Systems, Год журнала: 2024, Номер 15(1), С. 1443 - 1456

Опубликована: Март 14, 2024

Since the outbreak of global COVID-19 pandemic in Wuhan, China, 2019, its impact has been seen worldwide.Early identification is very crucial, as it keeps infected people isolated from other people, thus minimizing risk further transmission.The standard diagnostic approach based on RT-PCR.However, due to scarcity PCR kits some regions and costs associated with this technique, there a growing demand for alternative solutions.Recently, diagnosis by medical imaging recognized valid clinical practice.Meanwhile, massive increase cases put considerable pressure radiologists responsible interpreting these scans.This paper introduces an automated detection rapid diagnosis.We present deep CNN model differentiate between normal pneumonia cases, well patients COVID-19.Our EfficientNet-B7 architecture improved Squeeze Excitation block attention mechanism.In addition, we propose innovative that combines SVM achieve best performance.Experimental results show proposed framework provides better performance than existing SOTA methods, average accuracy 97.50%, while precision recall are both 100%.

Язык: Английский

---

GPU-based key-frame selection of pulmonary ultrasound images to detect COVID-19

Journal of Real-Time Image Processing, Год журнала: 2024, Номер 21(4)

Опубликована: Июнь 15, 2024

Abstract In the last decades, technological advances have led to a considerable increase in computing power constraints simulate complex phenomena various application fields, among which are climate, physics, genomics and medical diagnosis. Often, accurate results real time, or quasi needed, especially if related process requiring rapid interventions. To deal with such demands, more sophisticated approaches been designed, including GPUs, multicore processors hardware accelerators. Supercomputers manage high amounts of data at very speed; however, despite their performance, limitations due maintenance costs, obsolescence notable energy consumption. New processing architectures GPUs field can provide diagnostic therapeutic support whenever patient is subject risk. this context, image as an aid diagnosis, particular pulmonary ultrasound detect COVID-19, represents promising tool ability discriminate between different degrees disease. This technique has several advantages, no radiation exposure, low availability follow-up tests ease use even limited resources. work aims identify best approach optimize parallelize selection most significant frames video given input classification network that will differentiate healthy COVID patients. Three evaluated: histogram, entropy ResNet-50, followed by K-means clustering. Results highlight third accurate, simultaneously showing significantly lowering all times.

Язык: Английский

---

A high-accuracy lightweight network model for X-ray image diagnosis: A case study of COVID detection

PLoS ONE, Год журнала: 2024, Номер 19(6), С. e0303049 - e0303049

Опубликована: Июнь 18, 2024

The Coronavirus Disease 2019(COVID-19) has caused widespread and significant harm globally. In order to address the urgent demand for a rapid reliable diagnostic approach mitigate transmission, application of deep learning stands as viable solution. impracticality many existing models is attributed excessively large parameters, significantly limiting their utility. Additionally, classification accuracy model with few parameters falls short desirable levels. Motivated by this observation, present study employs lightweight network MobileNetV3 underlying architecture. This paper incorporates dense block capture intricate spatial information in images, well transition layer designed reduce size channel number feature map. Furthermore, label smoothing loss inter-class similarity effects uses class weighting tackle problem data imbalance. applies pruning technique eliminate unnecessary structures further parameters. As result, improved achieves an impressive 98.71% on openly accessible database, while utilizing only 5.94 million Compared previous method, maximum improvement reaches 5.41%. Moreover, research successfully reduces parameter count up 24 times, showcasing efficacy our approach. demonstrates benefits regions limited availability medical resources.

Язык: Английский

---