COVID-19 Detection Using Multimodal and Multi-model Ensemble Based Deep Learning Technique

Abstract

COVID-19 is a fatal disease that threatens the people’s health worldwide in the last few years. Although the testing techniques for COVID-19 had become more widespread, they still lack the speed and accuracy of disease pattern detection. Thanks to Artificial Intelligence (AI) as it can accelerate the detection process by deep learning techniques that can be used to achieve high performance in COVID-19 identification. Many types of Convolutional Neural Networks (CNN) as the most image classification deep learning techniques are used for automatically diagnosing this disease using X-ray or Computerized Tomography (CT-scan) medical images. The individual CNN types can obtain good results with a specific type of images like X-ray or CT-scan images in a certain dataset but, it could not give the same quality for other types of images or datasets. Through this paper, multiple standards model and custom CNN model have been merged using ensemble method to enhance the overall performance, while the accuracy of each model is a parameter in majority voting. Consequently, the proposed method will started with an initial simple classifier to classify between X-ray image and CT-image then followed by the ensemble model, and lasted by the decision making algorithm. Using different image types like X-ray and CT-scan images from different dataset sources enhance the overall performance as will be cleared in our results. The proposed model has three main parts: Multimodal imaging data, Multi-model based CNN structure, and decision-making diffusion based on the Multi-model output part. The main objective of using multiple models or multiple algorithms in detecting COVID-19 is to decrease the error percentage and increase the validation accuracy. Testing and validation results assure that the performance of the proposed method for COVID-19 chest X-rays and CT-scan images outperforms the individual and classical CNN learners’ design.