Compact Transformer Neural Network for Pulmonary Disease Classification from Radiological Imaging

Global catastrophe has resulted from the unique COVID-19 outbreak in the year 2020. Early detection of this illness can nearly entirely prevent its lethal and dangerous spread. Artificial intelligence and machine learning have been efficiently proven to accelerate value-based healthcare provided to the patients by the medical professionals by providing automated support for faster and timely diagnosis, healthcare needs and treatment profiling. With this initiative, we have taken up the challenge of proposing a methodology for detection of pulmonary diseases (corona and bacterial pneumonia) from radiological images of chest. The objective of this paper is to investigate compact vision transformer-based process for pulmonary diseases (corona and bacterial pneumonia) detection from chest radiological images. The proposed CVT-7/16 methodology has been comparatively evaluated with ResNet deep learning model. The CVT-7/16 model demonstrated an increased performance in terms of accuracy of 2–3%, over the ResNet model. A reduced performance has been observed of CVT-7/16 against ResNet-F1-score of 1–2% and precision of 2–3%. Overall, the compact convolutional transformer learning models have been a stimulating experimentation towards the benefits of using transformer-based approaches for the medical vision tasks. After analyzing the results, it has been reasoned that more fine-tuning of the compact vision transformer is to be performed for achieving better results.