OS26-11 Addressing Noise Challenges in CNN-based Pneumonia Detection: A Study using Primary Indonesian Thoracic Imagery

Title:	OS26-11 Addressing Noise Challenges in CNN-based Pneumonia Detection: A Study using Primary Indonesian Thoracic Imagery
Publication:	ICAROB2025
Volume:	30
Pages:	772-777
ISSN:	2188-7829
DOI:	10.5954/ICAROB.2025.OS26-11
Author(s):	Wahyu Andi Saputra, Andi Prademon Yunus
Publication Date:	February 13, 2025
Keywords:	pneumonia classification, CNN, salt-and-pepper noise, gaussian noise
Abstract:	Accurate pneumonia diagnosis is vital, especially in resource-limited areas like Indonesia. While CNNs show promise for automated detection using chest X-rays, real-world image quality affects their performance. This study addresses this challenge by using a primary dataset—images directly from Indonesian patients—to avoid the biases of preprocessed secondary data. This ensures our findings are relevant to the Indonesian context. We tested how different noise types (salt-and-pepper and Gaussian) impact the accuracy of several common CNN architectures. These noise types mimic common image imperfections. Our analysis reveals that noise degrades the CNN's ability for 3% to 5% performance. This highlights the need for better pre-processing methods and potentially specialized CNN designs to handle noisy images. Ultimately, our work improves our understanding of deploying CNNs for pneumonia diagnosis in real-world settings, leading to more reliable and helpful diagnostic tools. Using primary data from diverse populations is crucial for building trustworthy AI in healthcare.
PDF File:	https://alife-robotics.co.jp/members2025/icarob/data/html/data/OS/OS26/OS26-11.pdf
Copyright:	© The authors. This article is distributed under the terms of the Creative Commons Attribution License 4.0, which permits non-commercial use, distribution and reproduction in any medium, provided the original work is properly cited. See for details: https://creativecommons.org/licenses/by-nc/4.0/

HOME