Detection of COVID 19 from CT Image by The Novel LeNet-5 CNN Architecture

Islam, M. R.; Matin, A.; Ieee,

The COVID-19 is an infectious disease that primarily affects the lungs and leads to death in the severe stage. It also changes the lung CT scans of affected patients. For introducing a more convenient COVID-19 identification technique during this pandemic, we have implemented a simple convolution neural network (CNN) based model by using lung CT images. And finally, we have used LeNet-5 CNN architecture for this purpose. For training and testing purposes, we have obtained a dataset that contained 349 COVID-19 lung CT frames and 397 number of NON COVID-19 CT frames. We have introduced the data augmentation technique and got 1744 CT frames of COVID-19 and 1588 CT frames of NON COVID-19 patients. Among them, we have used 80% of lung CT frames for training purposes and 20% frames for testing purposes. The total number of trainable parameters of our LeNet-5 CNN architecture was 82,146. After completing the whole process, we got the accuracy of 86.06%, f1 score of 87%, the precision of 85%, and recall of 89%, and area under the ROC curve of 0.86 for COVID-19 detection.