Detecting COVID-19 in Inter-Patient Ultrasound Using EfficientNet

In order to stop the coronavirus from spreading, an early diagnosis is essential. In order to help with this, we suggest in this work a deep learning-based method for coronavirus patient detection utilizing ultrasound images. We suggest using the family of EfficientNet models for the classification of ultrasound images of potential patients that were trained on the ImageNet dataset. In particular, we consider both ordinary networks trained in a supervised setting and their noisy student counterpart pre-trained in a semi-supervised setting. By categorizing images as either positive or negative, we approach the detection problem from a binary classification standpoint. On the POCOVID-Net ultrasound dataset from the experiments, we assessed the models on inter-patient scenarios. This dataset includes 59 pictures and 202 lung ultrasound videos from 216 different people. This dataset contains samples from COVID-19 patients, patients with viral pneumonia, patients with bacterial pneumonia, and healthy controls. 96.97% accuracy was attained overall using EfficientNet-B2.