Abstract:
COVID-19 had infected 127 million people and killed 2.5 million by March 31, 2021. COVID-19 patients and the highly contagious disease must be diagnosed quickly. Since non-contrast chest computed tomography (CT) is clinically useful for COVID-19 diagnosis, deep learning (DL)-based automated methods have been proposed to help radiologists read the massive amounts of CT exams due to the pandemic. We address the data source bias problem in deep convolutional neural networks for COVID-19 classification using real-world multi-source data. The data source bias problem occurs when certain data sources contain only one class of data, which may cause DL models to distinguish data sources instead of COVID-19. We propose MIx-aNd-Interpolate (MINI), a simple, efficient, and effective training strategy. The MINI approach generates volumes of the absent class by combining hospital samples, expanding the sample space of the source-biased dataset. Experimental results on a large collection of real patient data (1,221 COVID-19 and 1,520 negative CT images, the latter consisting of 786 community acquired pneumonia and 734 non-pneumonia) from eight hospitals and health institutions show that: 1) MINI can improve COVID-19 classification performance over the baseline (which does not address source bias), and 2) MINI outperforms competing methods in terms of improvement.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
