Advanced Capsule Networks with Feature Thresholding for Enhanced COVID-19 Detection in X-ray Images
Year: 2025
Type: article
Abstract: Objectives: The primary goal involves creating an advanced detection system which utilizes capsule networks with autoencoders alongside dynamic thresholding to improve COVID-19 X-ray image recognition. Methods: The dataset gathered from Kaggle contained 341 publicly accessible X-ray images, which split between 111 COVID-19 infected cases and 230 negative cases. The integrated system combined autoencoders with capsule networks to obtain spatially conscious compressed features. The method implemented dynamic thresholds to optimize classification processes by adjusting threshold values according to input variability rates. The model completed its training process using various clinical examples while standard evaluation metrics measured accuracy, precision, sensitivity, and specificity. The proposed system underwent comparative testing with conventional CNNs and SVM-based classifiers alongside the TVFx approach to demonstrate its better diagnostic outcomes. Findings: The proposed model achieved superior results to traditional diagnostic methods with accuracy reaching 97.82% and precision at 97.15% and sensitivity at 97.10%. The model shows a dynamic response to new clinical data points which improves its reliability and robustness for detecting COVID-19 in medical imaging application. Dynamic thresholding features let the model continuously modify its detection parameters when processing different X-ray images to maximize accuracy. The research demonstrates that capsule networks linked to autoencoders together with a dynamically adapted thresholding framework result in substantial diagnostic advancement thus creating an essential instrument for combating COVID-19. Novelty: The integration of capsule networks and dynamic thresholding led to a novel system which automatically adjusted classification thresholds during real-time operations to improve X-ray image assessment for COVID-19 detection. Keywords: Covid19 Detection, CNN, Capsule Networks, Image Recognition, Data Augmentation
Authors: R. Sivarasan, Dr. M. S. Mythili
Institution: Bharathidasan University
Language: English
Topic: COVID-19 diagnosis using AI
Subfield: Radiology, Nuclear Medicine and Imaging
Field: Medicine
Domain: Health Sciences
Open Access status: diamond
OpenAlex ID: W4410518827