Comments on “COVIDX-Net: A Framework of Deep Learning Classifiers to Diagnose COVID-19 in X-Ray Images”

Abstract

In an early study by (Pan et al., 2020), the authors proposed a deep learning framework; called COVIDX-Net to support radiologists in the process of automatic diagnosis of COVID-19 infection from X-ray images. In this comment, we argue that Pan et al. 's paper include three failings that may impact the performance of models available in COVIDX-Net leading to erroneous results and incorrect conclusions. First, the study lacks a clear and distinct framework, instead employing a conventional approach of training and testing a set of pre-existing deep learning models. Second, the study lies in the utilization of a very small dataset for training complex deep learning models, with no information regarding the source and annotation process. Finally, the training deep learning models is inconsistent and suffer from overfitting. The study under inspection represents a troubling example of (un/intended) exploitation of the urgency surrounding the COVID-19 pandemic for self-serving purposes, particularly to accrue citations without obeying the ethical publishing practices or demonstrating due diligence.