ResearchBib Share Your Research, Maximize Your Social Impacts
Sign for Notice Everyday Sign up >> Login

Lung Disease Detection and Classification Using Single Shot Multi-Box Detector Network: A Comprehensive Study

Journal: Journal of Medicinal and Chemical Sciences (Vol.6, No. 11)

Publication Date:

Authors : ; ; ;

Page : 2849-2866

Keywords : Detection Hyper; parameters Lung diseases Optimizer Single shot detector;

Source : Downloadexternal Find it from : Google Scholarexternal

Abstract

Lung diseases significantly impact the world regarding health, economic cost, and social and psychological well-being. X-ray images are a primary method for diagnosing lung diseases, but the manual analysis of these images can be time-consuming, subjective, and prone to inaccuracies. However, it is essential to diagnose lung diseases in a timely manner and with high accuracy to ensure effective treatment and management. This study introduces an innovative deep-learning version termed the "ESSDN-LN model" to overcome these challenges. It is a variant of the single shot detector (SSD) network. This model aims to rapidly and accurately detect and classify six types of lung disease: aortic enlargement, cardiomegaly, pleural thickening, pulmonary fibrosis, COVID-19, and pneumonia. The ESSDN-LD model was introduced in three versions: ESSDN-LDV1, ESSDN-LDV2, and ESSDN-LDV3. ESSDN-LDV1 incorporates the SSD with batch normalization, dropout regularization, and data augmentation techniques. ESSDN-LDV2 builds upon the advancements of ESSDN-LDV1 by incorporating the random search algorithm for adjusting model hyper-parameters and introducing the skip connections technique to enhance the detection performance. Furthermore, ESSDN-LDV3 further enhances the capabilities of ESSDN-LDV1 using the genetic algorithm for hyper-parameter tuning and incorporating feature fusion and skip connections techniques, thereby significantly improving the detection performance. The ESSDN-LDV3 model demonstrated exceptional performance compared to other versions, achieving a remarkable accuracy of 96.5% and a prediction time of 0.018 seconds in the seven-class classification. Furthermore, it achieved a total accuracy of 98.4% and a prediction time of 0.013 seconds in the three-class classification, encompassing Covid-19, pneumonia, and no-finding cases. These impressive results highlight the effectiveness and efficiency of the proposed method in accurately classifying lung diseases and can contribute to improved patient outcomes and treatment decisions.

Last modified: 2023-07-31 18:50:53