Abdul Kareem, Razia Sulthana ORCID: https://orcid.org/0000-0001-5331-1310 and Kumari, Priyanshi (2026) A lightweight CNN for pleural effusion localisation and multi-class chest X-ray classification in resource-constrained settings. In: 2026 Intelligent Methods, Systems, and Applications (IMSA). IEEE Xplore . Institute of Electrical and Electronics Engineers (IEEE), Piscataway, New Jersey, pp. 386-392. ISBN 979-8331584887 (doi:10.1109/IMSA70415.2026.11700259)

Preview

PDF (Author's Accepted Manuscript) 54539 SULTHANA ABDUL KAREEM_A_Lightweight_CNN_For_Pleural_Effusion_Localisation_(IEEE AAM)_2026.pdf - Accepted Version Available under License Creative Commons Attribution. Download (4MB) | Preview

Abstract

Chest X-ray is a fundamental and rapidly acquired diagnostic for accurate and initial interpretation for diagnosing respiratory infection such as pneumonia, COVID-19, pulmonary diseases like tuberculosis and cancer, pleural effusion etc. The aim of this study is to apply and evaluate the effectiveness of deep learning techniques on two Chest X-ray analysis tasks: a) pleural effusion detection using MobileNet and DenseNet-101, and b) multi-class classification of Chest X-ray using DenseNet-101 into four classes: normal lungs, pneumonia, tuberculosis, and COVID-19. Experiments were conducted on two medical imaging datasets namely: NIH Chest X-ray Dataset and LungXrays Grayscale Dataset. For the pleural effusion detection both the models are trained to generate bounding boxes to localise affected regions, thereby providing spatial interpretability of the affected areas. For a representative test image, the model predicted a fluidrelated finding (probability 0.593) despite the ground-truth label of 'no effusion', highlighting the challenge of detecting subtle pleural abnormalities, while Grad-CAM visualisation were used to enhance interpretability in a hospital-based environment. For the classification task, DenseNet-101 achieved strong discriminatory performance with Area Under the Curve values of 0.96 for tuberculosis, 0.94 for normal cases, 0.87 for pneumonia and 0.79 for COVID-19. The light weight MobileNet architecture is evaluated and optimised for CPU-based inference, making it suitable for deployment in densely populated countries and particular in countries operating with resource-constrained clinical environments. The impact of this research lies in the potential to assist clinicians esp. those involved in the screening process, by acting as decision support tool and prioritise cases that require urgent attention, especially in low-resource settings.

Item Type:	Conference Proceedings
Title of Proceedings:	2026 Intelligent Methods, Systems, and Applications (IMSA)
Additional Information:	Conference held 11th -12th July 2026 in Giza, Egypt.
Uncontrolled Keywords:	chest X-ray, pleural effusion localisation, multi-class deep learning, resource-constrained clinical environments, grad-CAM interpretability
Subjects:	Q Science > Q Science (General) Q Science > QA Mathematics > QA75 Electronic computers. Computer science R Medicine > R Medicine (General) T Technology > T Technology (General)
Faculty / School / Research Centre / Research Group:	Faculty of Engineering & Science Faculty of Engineering & Science > School of Computing & Mathematical Sciences (CMS)
Last Modified:	02 Oct 2026 14:59
URI:	https://gala.gre.ac.uk/id/eprint/54539

Actions (login required)

View Item

Downloads