COVID-19 Detection from Lung Ultrasound Images using a Fuzzy Ensemble-based Transfer Learning Technique

Abstract

Due to the rapid spread of COVID-19 as a global pandemic, it has become increasingly critical to have fast, cheap, and reliable tools to assist physicians in diagnosing COVID19. Several automated systems using deep learning techniques have demonstrated promising results by analyzing Computed Tomography (CT-scan) or X-ray data to complement conventional diagnostic tools. In this paper, we aim to emphasize the role of point-of-care ultrasound imaging using deep learning as a tool to detect COVID-19 more prominently. Ultrasound imaging is non-invasive and widely available in medical facilities all over the world. This paper presents an ensemble technique based on Sugeno Fuzzy Integrals with convolutional neural networks (CNNs) as the base model. It classifies lung ultrasound (LUS) images of patients into COVID-19 and Non-COVID-19 categories. The lack of COVID-19 data makes it challenging to train a traditional CNN from scratch, so we have adapted a transfer learning approach instead of training the base classifiers VGG16, ResNet-50, and GoogLeNet. We apply the gained knowledge in the target domain of small lung ultrasound frames, considering the ImageNet dataset as the source domain. We have also adapted image pre-processing techniques to remove noises so that the model can only focus on specific features. Our proposed framework is evaluated on a publicly available dataset, achieving 96.7% accuracy. The proposed architecture outperforms the state-of-the-art method on the same dataset and proves to be a reliable COVID-19 detector.