|
Transfer Learning-Based Segmentation of Pneumonia from Chest X-Rays Images |
|---|---|
| รหัสดีโอไอ | |
| Creator | Kyi Pyar |
| Title | Transfer Learning-Based Segmentation of Pneumonia from Chest X-Rays Images |
| Publisher | Faculty of Informatics, Mahasarakham University |
| Publication Year | 2569 |
| Journal Title | Journal of Applied Informatics and Technology |
| Journal Vol. | 8 |
| Journal No. | 1 |
| Page no. | 255530 |
| Keyword | Deep learning, DeepLabV3, Pneumonia Segmentation, SegNet |
| URL Website | https://ph01.tci-thaijo.org/index.php/jait |
| Website title | Journal of Applied Informatics and Technology |
| ISSN | 3088-1803 |
| Abstract | Pneumonia remains a significant global health concern, warranting pre-cise and efficient diagnostic tools. This study introduces a comprehensiveapproach to pneumonia segmentation leveraging advanced deep learningtechniques. The primary goal is to enhance the precision of pneumonialocalization within medical images, specifically chest X-rays, through theutilization of state-of-the-art deep learning models. This study exploresthe application of advanced segmentation models, namely DeepLabV3 andSegNet, for the automated identification and delineation of pneumonia-affected regions within chest X-ray images. DeepLabV3, renowned for itssemantic segmentation capabilities that partitions an image into multiplesegments or regions, each of which is associated with a specific seman-tic label, and SegNet, featuring an encoder-decoder that consists of twomain components: an encoder and a decoder, are selected as the segmen-tation models. The training process of the system leverages the widelyacknowledged Kermany dataset, specifically composed of chest X-ray im-ages depicting cases of pneumonia. This dataset is well-established andholds a prominent status within the field, recognized for its relevance andsignificance in the context of pneumonia detection and classification tasks.As per the evaluation findings, it is evident that the system attains en-hanced accuracy by 0.844 and Intersection over Union score of 0.81 whenemploying the DeepLabV3 architecture compared to the SegNet architec-ture.1. IntroductionPneumonia, characterized by inflammation of the lungtissue, imposes a significant burden on healthcare sys-tems globally. This medical condition involves the in-flammation of lung tissue, specifically affecting the airsacs, or alveoli. Typically, this inflammation arises frominfections, including viruses, bacteria, or fungi. Pneu-monia manifests in diverse forms, varying in severityand presenting symptoms such as cough, chest pain,difficult breathing, and fever.There are three main types of pneumonia: vi-ral pneumonia, bacterial pneumonia, and fungal pneu-monia.Viral pneumonia, exemplified by illnesseslike COVID-19, primarily spreads through respiratorydroplets, contact with contaminated surfaces, or closeinteraction with an infected individual. Bacterial pneu-monia can spread through respiratory droplets, trans-mitted when an infected person coughs or sneezes.On the other hand, fungal pneumonia is often asso-ciated with environmental exposure and may not di-rectly transmit between individuals. Diagnostic toolscommonly employed in pneumonia cases include chestX-rays and computed tomography (CT) scans, pro-viding visualizations of lung abnormalities. Addition-ally, blood tests play a crucial role in identifying thecausative agent, be it bacteria or viruses. Sputum testsare conducted to analyze the material produced duringcoughing, aiding in the accurate diagnosis and targetedtreatment of pneumonia.The understanding of pneumonia’s diverse formsand transmission methods is pivotal for effective health-care management and preventive measures, especially inthe context of widespread respiratory infections. Earlyand precise diagnosis, facilitated by a combination ofimaging techniques and laboratory tests, is essential forISSN 3088-1803 | Copyright© 2026. Published by the Faculty of Informatics, Mahasarakham University. All rights reserved.This is an open access article distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License (CC BY-NC-ND 4.0). |
