A lightweight CNN-based network on COVID-19 detection using X-ray and CT images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Traditional COVID-19 detection relies on manual interpretation of medical images, which can lead to errors.
• Convolutional Neural Networks (CNNs) offer a potential solution for automated and accurate disease identification.

Purpose of the Study:

• To develop and evaluate a lightweight CNN model for efficient and accurate COVID-19 detection.
• To compare the performance of the proposed model against established CNN architectures.

Main Methods:

• Fine-tuning seven CNNs (InceptionV3, ResNet50V2, Xception, DenseNet121, MobileNetV2, EfficientNet-B0, EfficientNetV2) for COVID-19 detection.
• Proposing and evaluating a novel lightweight CNN, LightEfficientNetV2, on chest X-ray and CT images.
• Utilizing five-fold cross-validation across three datasets (NIH Chest X-rays, SARS-CoV-2, COVID-CT).

Main Results:

• LightEfficientNetV2 achieved 98.33% accuracy on chest X-ray images and 97.48% on CT images.
• After fine-tuning, EfficientNetV2 reached 97.73% accuracy on X-rays, and Xception achieved 96.78% on CT scans.
• The proposed LightEfficientNetV2 outperformed existing models on the tested datasets.

Conclusions:

• LightEfficientNetV2 demonstrates significant potential for COVID-19 detection using medical imaging.
• The model offers a promising, accurate, and efficient solution compared to state-of-the-art methods.