PETLFC: Parallel ensemble transfer learning based framework for COVID-19 differentiation and prediction using deep convolutional neural network models.

Despite a worldwide research involvement in the global COVID-19 pandemic, the research community is still struggling to develop reliable and faster prediction mechanisms for this infectious disease which is distinct from other respiratory diseases. The commonly used clinical RT-PCR test is not widely available in areas with limited testing facilities, and it performs and responds slowly. Using digital chest X-Ray images and CT scan images, recently a number of works are proposed using deep transfer learning and ensemble of these deep models as base classifiers. Though ensemble approaches exhibit better accuracy, they are computational intensive and have slower prediction time. Therefore, to handle computational-intensiveness and to accelerate prediction time without compromising accuracy, a Parallel Ensemble Transfer Learning based Framework for COVID (PETLFC) is proposed for the underlying multi-class classification problem. Three pre-trained convolutional neural network models (VGG16, ResNet18, and DenseNet121) were fine tuned to act as base models for the proposed parallelized bagging-based ensemble to predict COVID-19. The data parallel model is implemented on PARAM SHAVAK HPC system using MPI programming and a dataset of 21,165 chest X-Ray images (10,192 normal, 1345 pneumonia, 3616 COVID-19, and 6012 lung opacity). The results are compared with some state-of-the-art sequential ensemble approaches where the proposed PETLFC was observed to exhibit superior performance. Highlights: • This work proposes a parallel ensemble transfer learning based deep CNN framework for COVID-19 prediction from chest X-Rays. • The deep CNN models which are tuned using transfer learning are: DenseNet-121, VGG-16 and ResNet18. They act as base classifiers for the proposed parallelized bagging-based ensemble. • The underlying problem is formulated using data parallel model and as a multi-class classification problem to differentiate COVID-19 from other similar pulmonary diseases. This enables the model to predict COVID-19 in lesser time due to parallelism, with equal accuracy with state-of-the-art sequential bagging-based ensemble methods. • An in-depth evaluation of the system is carried out considering standard performance metrics like accuracy, precision, recall and F1-score. The parallel system is tested using standard parallel performance metrics like speedup and efficiency. [ABSTRACT FROM AUTHOR]