COVID-19 infection prediction from CT scan images of lungs using Iterative Convolution Neural Network model

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• COVID-19, caused by SARS-CoV-2, is a contagious respiratory illness.
• Chest CT scans offer a reliable and rapid method for detecting COVID-19 infections.

Purpose of the Study:

• To evaluate the performance of Iterative Convolutional Neural Networks (CNNs) for COVID-19 detection using chest CT images.
• To compare two CNN architectures (CNN1 with two hidden layers, CNN2 with three hidden layers) and determine optimal training parameters.

Main Methods:

• Trained supervised classifiers, specifically Iterative CNNs, on chest CT images categorized as COVID-19 positive or negative.
• Utilized six different training data sizes and five iterations for model training and testing.
• Evaluated two CNN architectures (CNN1 and CNN2) with a fixed testing dataset of 20 images.

Main Results:

• The performance of the models degraded after the 6th iteration, leading to the selection of 5 iterations.
• A total of 60 different models were generated across two CNN architectures and varying training set sizes.
• The CNN2 model, with 100 training sets and in its 5th iteration, achieved the highest classification accuracy of 89%.

Conclusions:

• Iterative Convolutional Neural Networks, particularly the CNN2 architecture, demonstrate high accuracy in classifying COVID-19 from chest CT scans.
• The study highlights the potential of AI-driven image analysis for efficient and reliable COVID-19 diagnosis.