More and more tests are executed in the fight against the Corona virus. And when it comes to testing, one big question is whether we should use computed tomography (CT) scans for frontline diagnosis. CT scans capture abnormalities on the lungs of COVID-19 patients, but many other infections look very similar, and doctors worry that this may lead to false-positive and false-negative diagnoses.
This problem inspired Lawrence Berkeley National Laboratory (Berkeley Lab) to explore whether image recognition, algorithms and a data analysis pipeline can help accurately distinguish COVID-19 abnormalities in CT scans and chest X-rays from other overlapping respiratory illnesses, such as influenza, H1N1, other SARS viruses, and MERS. Berkeley Lab scientist Dani Ushizima formed a research team and asked Daan Pelt, researcher at CWI’s Computational Imaging group, to join the team. Pelt is also a member of Berkeley Lab's Center for Advanced Mathematics for Energy Research Applications (CAMERA).
The role of CWI in the team is to use machine learning algorithms developed at the Computational Imaging group to enable automatic analysis of CT images of the lungs. Specifically, we apply Mixed-Scale Dense convolutional neural networks [1] to automatically detect abnormal lung regions. “As a mathematician, it is motivating to work on a problem that is close to society, important, and urgent,” said Daan Pelt. “I hope that our algorithms can contribute to a better understanding and diagnosis of COVID-19 cases.”
[1] Pelt, D. M., & Sethian, J. A. (2018). A mixed-scale dense convolutional neural network for image analysis. Proceedings of the National Academy of Sciences, 115(2), 254-259.
This article was published by Berkely Lab. Read the full text here.
