CovidSafe: A Deep Learning Framework for Covid Detection Using Multi-Modal Approach

Abstract

This work introduces CovidSafe, a framework designed to classify COVID patients as positive or negative based on CXR images. With the global impact of COVID-19, there is a need for effective tools to aid in patient diagnosis. CovidSafe utilizes a multi-model deep feature extractor, combining features from the ResNet family and an autoencoder. The ResNet 18, ResNet 34, and ResNet 50 architectures are employed, and model weights are fine-tuned for optimal performance. The framework leverages the ResNet family models and theautoencoder to extract relevant features. The features from intermediate layers in Resnet family are extracted to capture increasingly complex and high-level information. Similarly, the bottleneck features from Autoencoder are extracted to capture compressed, abstract representation and important features of the image for the classification. These features are then utilized for analysis and classification within the CovidSafe framework. To ensure accessibility for online users, the proposed model is integrated into a web application. Through experimental analysis, the performance of the proposed model is evaluated with various metrics. Results indicate that CovidSafe outperforms baseline models, achieving a precision of 98.67%, recall of 98.62%, and F1-score of 98.64%. These promising results highlight the efficacy of CovidSafe in accurately classifying COVID patients based on CXR images, offering a valuable tool in the fight against the ongoing pandemic.