ResDSda_U-Net: A Novel U-Net-Based Residual Network for Segmentation of Pulmonary Nodules in Lung CT Images

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 87775 - 87789

Published: Jan. 1, 2023

The timely detection and segmentation of pulmonary nodules in lung computed tomography (CT) images can aid the early diagnosis treatment cancer. However, manual by doctors is highly demanding terms operational requirements efficiency. To effectively improve nodule segmentation, this paper proposes a novel neural network, called ResDSda_U-Net, based on original U-Net network with following improvements: (1) combining Depthwise Over-parameterized Convolutional layer (DO-Conv) simple parameter-free attention module (SimAM), form newly designed ResDS block; (2) introducing dense atrous spatial pyramid pooling (DASPP) module, between encoder decoder, using modified dilated rates to extract multi-scale information more effectively; (3) channel mechanisms Convolution Channel Attention (CCA) Spatial (CSA) blocks, enhance global pixel attention, fully capture contextual information, enable decoder better eliminate differences pixels. conducted experiments demonstrate that proposed ResDSda_U-Net outperforms all existing networks (according evaluation metrics used) considered state-of-the-art half metrics), achieving corresponding values 86.65% for Dice Similarity Coefficient (DSC), 76.73% Intersection over Union (IoU), 86.30% sensitivity, 87.22% precision.

Language: Английский

---

A deep learning‐based x‐ray imaging diagnosis system for classification of tuberculosis, COVID‐19, and pneumonia traits using evolutionary algorithm

International Journal of Imaging Systems and Technology, Journal Year: 2023, Volume and Issue: 34(1)

Published: Dec. 21, 2023

Abstract To aid in detection of tuberculosis, researchers have concentrated on developing computer‐aided diagnostic technologies based x‐ray imaging. Since it generates noninvasive standard‐of‐care data, a chest image is one the most often used imaging modalities solutions. Due to their significant interclass similarities and low intra‐class variation abnormalities, pictures continue pose difficulty for proper diagnosis. In this paper, novel automated framework proposed classification COVID‐19, pneumonia from images using deep learning improved optimization technique. Two pre‐trained convolutional neural network models such as EfficientB0 ResNet50 been utilized fine‐tuned additional layers. Both are trained with fixed hyperparameters selected datasets obtained newly models. A feature selection technique has that selects best features. version, distance update position formulation modified. The features further fused serial standard deviation threshold function. experimental process conducted three an accuracy 98.2%, 99.0%, 98.7%, respectively. addition, detailed Wilcoxon signed‐rank analysis shows method significance performance. Based results, concluded after fusion process. comparison recent techniques more terms precision rate.

Language: Английский

---

Prediction of Omicron Virus Using Combined Extended Convolutional and Recurrent Neural Networks Technique on CT-Scan Images

Interdisciplinary Perspectives on Infectious Diseases, Journal Year: 2022, Volume and Issue: 2022, P. 1 - 11

Published: Nov. 30, 2022

COVID-19 has sparked a global pandemic, with variety of inflamed instances and deaths increasing on an everyday basis. Researchers are actively improving distinct mathematical ML algorithms to forecast the infection. The prediction detection Omicron variant brought new issues for health fraternity due its ubiquity in human beings. In this research work, two learning algorithms, namely, deep (DL) machine (ML), were developed virus infections. Automatic disease have become crucial medical science rapid population growth. study, combined Extended CNN-RNN model was chest CT-scan image dataset predict number +ve −ve cases proposed evaluated compared against existing system utilizing 16,733-sample training testing images collected from Kaggle repository. This article aims introduce DL technique based combination Convolutional Neural Network (ECNN) Recurrent (ERNN) diagnose virus-infected automatically using images. To overcome drawbacks system, proposes that is ECNN-ERNN, where ECNN used extraction features ERNN exploration extracted features. A 16,733 computer tomography as pilot assessment prototype. investigational experiment results show projected prototype provides 97.50% accuracy, 98.10% specificity, 98.80% AUC, 97.70% F1-score. last, study outlines advantages being offered by respect other models comparing different parameters validation such error rate, data size, time complexity, execution time.

Language: Английский

---

Towards unsupervised radiograph clustering for COVID-19: The use of graph-based multi-view clustering

Engineering Applications of Artificial Intelligence, Journal Year: 2024, Volume and Issue: 133, P. 108336 - 108336

Published: April 2, 2024

Automatic classification methods widely used for diagnosing and analyzing COVID-19 cases. These assume known labels rely on a single view of the dataset. Given prevalence cases extensive volume patient records lacking labels, this communication underscores our unique approach—conducting first study case diagnosis in an unsupervised manner. Our work operates under assumption prior knowledge regarding number classes, such as COVID-19, pneumonia, normal, study. By adopting learning paradigm, we leverage wealth unlabeled data, reducing dependence human experts annotating numerous images. This paper introduces enhanced version recent direct method where non-negative cluster indices spectral embeddings are jointly estimated. Beyond inherent advantages method, proposed model improvements through two additional types constraints: (i) ensuring consistent smoothing across all views (ii) imposing orthogonality constraint matrix assignments. The efficacy is demonstrated using public COVIDx dataset with three showcasing promising results categorizing radiographs. approach tested other image datasets to assess its effectiveness.

Language: Английский

---

Pulmonary Edema and Pleural Effusion Detection Using EfficientNet-V1-B4 Architecture and AdamW Optimizer from Chest X-Rays Images

Computers, materials & continua/Computers, materials & continua (Print), Journal Year: 2024, Volume and Issue: 80(1), P. 1055 - 1073

Published: Jan. 1, 2024

This paper presents a novel multiclass system designed to detect pleural effusion and pulmonary edema on chest X-ray images, addressing the critical need for early detection in healthcare. A new comprehensive dataset was formed by combining 28,309 samples from ChestX-ray14, PadChest, CheXpert databases, with 10,287, 6022, 12,000 representing Pleural Effusion, Pulmonary Edema, Normal cases, respectively. Consequently, preprocessing step involves applying Contrast Limited Adaptive Histogram Equalization (CLAHE) method boost local contrast of samples, then resizing images 380 × dimensions, followed using data augmentation technique. The classification task employs deep learning model based EfficientNet-V1-B4 architecture is trained AdamW optimizer. proposed achieved an accuracy (ACC) 98.3%, recall precision 98.7%, F1-score 98.7%. Moreover, robustness revealed Receiver Operating Characteristic (ROC) analysis, which demonstrated Area Under Curve (AUC) 1.00 normal cases 0.99 effusion. experimental results demonstrate superiority multi-class system, has potential assist clinicians timely accurate diagnosis, leading improved patient outcomes. Notably, ablation-CAM visualization at last convolutional layer portrayed further enhanced diagnostic capabilities heat maps will aid interpreting localizing abnormalities more effectively.

Language: Английский

---