What question did this study set out to answer?

This research aims to evaluate the effectiveness of various CNN models in diagnosing respiratory diseases using chest X-ray images.

June 18, 2026

AI-Powered Diagnosis of Respiratory Diseases: A CNN-Based Study on Chest X-ray Image Classification

Key Points

This research aims to evaluate the effectiveness of various CNN models in diagnosing respiratory diseases using chest X-ray images.
Utilized established CNN models: AlexNet, VGG16, InceptionV3, ResNet50, DenseNet121, MobileNetV2, EfficientNetB0.
Conducted tests on public CXR datasets, specifically ChestX-ray14 and COVID.
Measured performance using accuracy and F1 score metrics.
The customized CNN model achieved the highest accuracy of 91.05% and an F1 score of 95%.
The EfficientNetB0 model demonstrated a good balance between accuracy and computational complexity.
Models ResNet50 and DenseNet121 showed robust generalization across different respiratory diseases.

Abstract

Introduction:: Precise and timely diagnosis of pulmonary diseases plays a significant role in successful medical treatment, especially during global health emergencies like the COVID-19 pandemic. Chest X-ray analysis has previously remained an available, contactless method for the diagnosis of pulmonary diseases, including COVID-19, Pneumonia, and tuberculosis. Recent developments in deep learning techniques make the process of automatic diagnosis possible through the use of convolutional neural networks (CNNs). Materials and Methods:: In this study, the performance of established CNN models such as AlexNet, VGG16, InceptionV3, ResNet50, DenseNet121, MobileNetV2, and EfficientNetB0 was tested with the public CXR datasets (ChestX-ray14) and COVID. The experiments were conducted with an equal preprocessing environment to ensure fairness of the investigation. The metrics employed to test the models include accuracy and F1 measure. Results:: The customized CNN model performed better than all other benchmark models by achieving the highest accuracy of 91.05% and an F1 score of 95%. This result outshines VGG16, ResNet50, MobileNetV2, and EfficientNetB0 models. The model EfficientNetB0 showed good accuracy and computational complexity trade-off. The models ResNet50 and DenseNet121 proved their robust generalization performances on many respiratory diseases. Discussion:: The results emphasize the differences in the performance of the various CNN models and the need to select the appropriate architecture according to certain deployment requirements. Although the simple models, such as the MobileNetV2, perform well and require less computation, other deeper models, including the DenseNet121, perform better in terms of overall generalization and applicability to different diagnostic tasks. Conclusion:: The proposed model performed better than all benchmark models with an accuracy of 91.05% and an F1 score of 95%, which clearly indicates its efficacy in being utilized in real-world scenarios for automatic pediatric as well as adult respiratory disorders through chest X-ray imaging.

Mark Helpful

Bookmark

Relay