Diagnosing COVID-19 of Lung CT Scan by Using Convolution Neural Network

EasyChair Preprint 6112, version 2

Abstract

The rapid spread of the Coronavirus and its threat to humanity prompted us to search for the best ways to diagnose this disease since the traditional PCR method is limited in number and its sensitivity to the disease reaches 70% and takes time. One of the proposed methods for diagnosing Covid is through CT images, and since CT images require specialized and experienced radiologists and their number is small compared to the number of infected people. So we proposed an automatic COVID diagnosis through CT images using deep learning. An effective network for classifying images is CNN.

Two datasets were obtained from the Kaggle website. Two groups were formed by merging the first images consisting of (2924) CT images and the second consisting of (4153) images.

Where a model was proposed consisting of three CNN models, one trained from scratch (MuNet), which was selected through which it was compared with two other networks using dataset1 and the other two were pre-trained (ResNet50, and InceptionV3) were selected after training several models (DenseNet121, MobileNetV2, NASNetMobile, VGG16, EfficientNetB0, ResNet50, and InceptionV3) on dataset2 and they obtained the best accuracy within 20 epochs. Before training the networks we did a pre-processing step to standardize and augment the data. In the third step after training the three models, the ensemble method is used to obtain the best accuracy among them.

The proposed model after training the proposed networks over 40 epochs obtained the accuracy (1.00), Sensitivity (1.0000), and Specificity (0.9950) for training, and accuracy (0.91) and Sensitivity (0.9529) and Specificity (0.8525) for testing.

Keyphrases: CNN, CT scan, Diagnosing COVID-19

@booklet{EasyChair:6112,
  author    = {Mustafa Khalaf Hashem and Ahmed Saleem Abbas},
  title     = {Diagnosing COVID-19 of Lung CT Scan by Using Convolution Neural Network},
  howpublished = {EasyChair Preprint 6112},
  year      = {EasyChair, 2021}}