In silico study demonstrates high diagnostic accuracy using a lightweight network on chest X-rays, suggesting feasible COVID-19 screening on edge devices.
Key Points
To design a lightweight convolutional neural network capable of detecting COVID-19 from chest X-ray images with minimal computational complexity for deployment on edge devices.
Developed a three-stage CNN architecture incorporating image pre-processing (black padding removal), convolution with filter banks, and deep feature extraction utilizing skip connections.
Partitioned chest X-ray datasets into training (70%), validation (10%), and testing (20%) sets.
Benchmarked the proposed architecture against four existing deep learning models: LMNet, CoroNet, CVDNet, and Deep GRU-CNN.
The proposed model achieved 99.47% accuracy on training data and 98.91% accuracy on testing data.
Diagnostic evaluation demonstrated a precision of 97.54%, recall of 98.19%, specificity of 99.49%, and an F1-score of 97.86%.
The model maintained performance on par with complex models while exhibiting reduced architectural complexity and lower susceptibility to overfitting.