A Deep Batch Normalized Convolution Approach for Improving COVID-19 Detection from Chest X-ray Images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Pre-trained machine learning models are prevalent for COVID-19 detection from X-rays.
• These models often suffer from bias due to extensive pre-trained weights and parameters.
• A need exists for more efficient and accurate detection methods.

Purpose of the Study:

• To propose a novel Batch Normalized Convolutional Neural Network (BNCNN) model.
• To enhance COVID-19 detection from chest X-ray images.
• To improve model performance and reduce computational complexity compared to pre-trained networks.

Main Methods:

• The BNCNN model involves data pre-processing, feature extraction, and classification.
• Feature extraction utilizes convolutional, batch normalization, and max-pooling layers.
• Classification employs dense, batch normalization, and dropout layers to prevent overfitting.

Main Results:

• The proposed BNCNN model demonstrated superior performance over four comparative pre-trained models on a public dataset.
• The model achieved better results in both three-way and two-way classification frameworks.
• BNCNN requires fewer parameters, indicating suitability for low-resource devices.

Conclusions:

• The BNCNN model offers an effective and computationally efficient approach for COVID-19 detection using X-ray images.
• It outperforms existing pre-trained models while maintaining a smaller parameter footprint.
• This suggests potential for wider deployment in resource-constrained environments.