Optimizing Pulmonary Chest X-ray Classification with Stacked Feature Ensemble and Swin Transformer Integration DOI

Manas Ranjan Mohanty,

Pradeep Kumar Mallick, A. V. Reddy

et al.

Biomedical Physics & Engineering Express, Journal Year: 2024, Volume and Issue: 11(1), P. 015009 - 015009

Published: Oct. 29, 2024

Abstract This research presents an integrated framework designed to automate the classification of pulmonary chest x-ray images. Leveraging convolutional neural networks (CNNs) with a focus on transformer architectures, aim is improve both accuracy and efficiency image analysis. A central aspect this approach involves utilizing pre-trained such as VGG16, ResNet50, MobileNetV2 create feature ensemble. notable innovation adoption stacked ensemble technique, which combines outputs from multiple models generate comprehensive representation. In approach, each undergoes individual processing through three networks, pooled images are extracted just before flatten layer model. Consequently, in 2D grayscale format obtained for original image. These serve samples creating 3D resembling RGB stacking, intended classifier input subsequent analysis stages. By incorporating pooling layers facilitate ensemble, broader range features utilized while effectively managing complexities associated augmented pool. Moreover, study incorporates Swin Transformer architecture, known capturing local global features. The architecture further optimized using artificial hummingbird algorithm (AHA). fine-tuning hyperparameters patch size, multi-layer perceptron (MLP) ratio, channel numbers, AHA optimization technique aims maximize accuracy. proposed framework, featuring AHA-optimized features, evaluated diverse datasets—VinDr-CXR, PediCXR, MIMIC-CXR. observed accuracies 98.874%, 98.528%, 98.958% respectively, underscore robustness generalizability developed model across various clinical scenarios imaging conditions.

Language: Английский

Facial image analysis for automated suicide risk detection with deep neural networks DOI Creative Commons
Amr E. Eldin Rashed,

Ahmed E. Mansour Atwa,

Ali Mohammed Saleh Ahmed

et al.

Artificial Intelligence Review, Journal Year: 2024, Volume and Issue: 57(10)

Published: Sept. 3, 2024

Accurately assessing suicide risk is a critical concern in mental health care. Traditional methods, which often rely on self-reporting and clinical interviews, are limited by their subjective nature may overlook non-verbal cues. This study introduces an innovative approach to assessment using facial image analysis. The Suicidal Visual Indicators Prediction (SVIP) Framework leverages EfficientNetb0 ResNet architectures, enhanced through Bayesian optimization techniques, detect nuanced expressions indicating state. models' interpretability improved GRADCAM, Occlusion Sensitivity, LIME, highlight significant regions for predictions. Using datasets DB1 DB2, consist of full cropped images from social media profiles individuals with known outcomes, the method achieved 67.93% accuracy up 76.6% Bayesian-optimized Support Vector Machine model ResNet18 features DB2. provides less intrusive, accessible alternative video-based methods demonstrates substantial potential early detection intervention

Language: Английский

Citations

3
Enhanced COVID-19 Detection from X-ray Images with Convolutional Neural Network and Transfer Learning DOI Creative Commons
Qanita Bani Baker, Mahmoud Hammad, Mohammad AL-Smadi

et al.

Journal of Imaging, Journal Year: 2024, Volume and Issue: 10(10), P. 250 - 250

Published: Oct. 13, 2024

The global spread of Coronavirus (COVID-19) has prompted imperative research into scalable and effective detection methods to curb its outbreak. early diagnosis COVID-19 patients emerged as a pivotal strategy in mitigating the disease. Automated using Chest X-ray (CXR) imaging significant potential for facilitating large-scale screening epidemic control efforts. This paper introduces novel approach that employs state-of-the-art Convolutional Neural Network models (CNNs) accurate detection. employed datasets each comprised 15,000 images. We addressed both binary (Normal vs. Abnormal) multi-class (Normal, COVID-19, Pneumonia) classification tasks. Comprehensive evaluations were performed by utilizing six distinct CNN-based (Xception, Inception-V3, ResNet50, VGG19, DenseNet201, InceptionResNet-V2) As result, Xception model demonstrated exceptional performance, achieving 98.13% accuracy, 98.14% precision, 97.65% recall, 97.89% F1-score classification, while multi-classification it yielded 87.73% 90.20% an 87.49% F1-score. Moreover, other utilized models, such competitive performance compared with many recent works.

Language: Английский

Citations

3
Enhancing COVID-19 disease severity classification through advanced transfer learning techniques and optimal weight initialization schemes DOI
Tijana Geroski, Vesna Ranković, Ognjen Pavić

et al.

Biomedical Signal Processing and Control, Journal Year: 2024, Volume and Issue: 100, P. 107103 - 107103

Published: Oct. 23, 2024

Language: Английский

Citations

1
Mobile Diagnosis of COVID-19 by Biogeography-based Optimization-guided CNN DOI
Xue Han,

Zuojin Hu

Mobile Networks and Applications, Journal Year: 2024, Volume and Issue: unknown

Published: March 4, 2024

Language: Английский

Citations

1
Optimizing Pulmonary Chest X-ray Classification with Stacked Feature Ensemble and Swin Transformer Integration DOI

Manas Ranjan Mohanty,

Pradeep Kumar Mallick, A. V. Reddy

et al.

Biomedical Physics & Engineering Express, Journal Year: 2024, Volume and Issue: 11(1), P. 015009 - 015009

Published: Oct. 29, 2024

Abstract This research presents an integrated framework designed to automate the classification of pulmonary chest x-ray images. Leveraging convolutional neural networks (CNNs) with a focus on transformer architectures, aim is improve both accuracy and efficiency image analysis. A central aspect this approach involves utilizing pre-trained such as VGG16, ResNet50, MobileNetV2 create feature ensemble. notable innovation adoption stacked ensemble technique, which combines outputs from multiple models generate comprehensive representation. In approach, each undergoes individual processing through three networks, pooled images are extracted just before flatten layer model. Consequently, in 2D grayscale format obtained for original image. These serve samples creating 3D resembling RGB stacking, intended classifier input subsequent analysis stages. By incorporating pooling layers facilitate ensemble, broader range features utilized while effectively managing complexities associated augmented pool. Moreover, study incorporates Swin Transformer architecture, known capturing local global features. The architecture further optimized using artificial hummingbird algorithm (AHA). fine-tuning hyperparameters patch size, multi-layer perceptron (MLP) ratio, channel numbers, AHA optimization technique aims maximize accuracy. proposed framework, featuring AHA-optimized features, evaluated diverse datasets—VinDr-CXR, PediCXR, MIMIC-CXR. observed accuracies 98.874%, 98.528%, 98.958% respectively, underscore robustness generalizability developed model across various clinical scenarios imaging conditions.

Language: Английский

Citations

0