Small size CNN-Based COVID-19 Disease Prediction System using CT scan images on PaaS cloud

COVID-19 is a highly contagious disease that can quickly spread and overwhelm healthcare systems if not controlled in time. Reverse Transcription Polymerase Chain Reaction (RT-PCR) is commonly used to diagnose COVID-19 but has low sensitivity and can be time-consuming. Computed Tomography (CT) scans can identify specific lung patterns or abnormalities associated with COVID-19 infection, which can help diagnose the disease. This paper presents an efficient forecasting framework for COVID-19 based on Convolutional Neural Networks (CNNs) to aid medical professionals in diagnosing COVID-19. The proposed framework was trained on the online COVID-19 dataset from Kaggle, which was split into train, validation, and test sets. The CNN achieved an accuracy of 99.11% on the test set. K-fold cross-validation was applied to the CNN, resulting in an average accuracy of 97.2%. The research explores alternative Machine Learning (ML) models, including Logistic Regression, Support Vector Machine, Decision Tree, K-Nearest Neighbour, and Random Forest, alongside Deep CNNs like ResNet50, VGG16, and InceptionV3 for COVID-19 prediction. The CNN model underwent analysis using the Local Interpretable Model-Agnostic Explanations (LIME) method and bootstrap resampling for Confidence Interval (CI) estimation to enhance interpretability. This can help to understand the model's predictions and assess their uncertainty. The developed CNN model, optimized for reduced memory usage, was seamlessly deployed on the Platform-as-a-Service (PaaS) cloud. Post-deployment, an accessible Hypertext Transfer Protocol Secure (HTTPS) link facilitates mobile phone accessibility, offering a user-friendly interface for widespread utilization. The proposed CNN-based forecasting framework is a promising tool for improving the accuracy and accessibility of COVID-19 diagnosis. The deployment of the CNN model to the PaaS cloud makes it accessible to a broader range of users, including those in remote or underserved areas. The HTTPS link generated after deployment allows users to access the model from their mobile phones, making it a convenient and portable tool for COVID-19 diagnosis.