Lung Segmentation with Lightweight Convolutional Attention Residual U-Net DOI Creative Commons

M Jannat,

Shaikh Afnan Birahim, Mohammad Asif Hasan

et al.

Diagnostics, Journal Year: 2025, Volume and Issue: 15(7), P. 854 - 854

Published: March 27, 2025

Background: Examining chest radiograph images (CXR) is an intricate and time-consuming process, sometimes requiring the identification of many anomalies at same time. Lung segmentation key to overcoming this challenge through different deep learning (DL) techniques. Many researchers are working improve performance efficiency lung models. This article presents a DL-based approach accurately identify mask region in CXR assist radiologists recognizing early signs high-risk diseases. Methods: paper proposes novel technique, Lightweight Residual U-Net, combining strengths convolutional block attention module (CBAM), Atrous Spatial Pyramid Pooling (ASPP) block, module, which consists only 3.24 million trainable parameters. Furthermore, proposed model has been trained using both RELU LeakyReLU activation functions, with yielding superior performance. The study indicates that Dice loss function more effective achieving better results. Results: evaluated on three benchmark datasets: JSRT, SZ, MC, score 98.72%, 97.49%, 99.08%, respectively, outperforming state-of-the-art Conclusions: Using capabilities DL cutting-edge processes, improves current efforts enhance for serious

Language: Английский

Lung Segmentation with Lightweight Convolutional Attention Residual U-Net DOI Creative Commons

M Jannat,

Shaikh Afnan Birahim, Mohammad Asif Hasan

et al.

Diagnostics, Journal Year: 2025, Volume and Issue: 15(7), P. 854 - 854

Published: March 27, 2025

Background: Examining chest radiograph images (CXR) is an intricate and time-consuming process, sometimes requiring the identification of many anomalies at same time. Lung segmentation key to overcoming this challenge through different deep learning (DL) techniques. Many researchers are working improve performance efficiency lung models. This article presents a DL-based approach accurately identify mask region in CXR assist radiologists recognizing early signs high-risk diseases. Methods: paper proposes novel technique, Lightweight Residual U-Net, combining strengths convolutional block attention module (CBAM), Atrous Spatial Pyramid Pooling (ASPP) block, module, which consists only 3.24 million trainable parameters. Furthermore, proposed model has been trained using both RELU LeakyReLU activation functions, with yielding superior performance. The study indicates that Dice loss function more effective achieving better results. Results: evaluated on three benchmark datasets: JSRT, SZ, MC, score 98.72%, 97.49%, 99.08%, respectively, outperforming state-of-the-art Conclusions: Using capabilities DL cutting-edge processes, improves current efforts enhance for serious

Language: Английский

Citations

0