Severity scoring of COVID-19

Customer: Beth Israel Deaconess Medical Center (Boston, United States)

Summary: In this study, we propose a two-step workflow for segmentation and scoring of lung diseases. The workflow inherits the quantification, qualification, and visual assessment of lung diseases on X-ray images estimated by radiologists and clinicians. It requires the completion of two core stages dedicated to lung and disease segmentation, as well as an additional post-processing stage dedicated to scoring. The latter integrated block is utilized mainly for estimating segment scores and computing the overall severity score of a patient. The models of the proposed workflow were trained and tested on four publicly available X-ray datasets of COVID-19 patients and two X-ray datasets of patients without lung pathology. Based on a combined dataset of 580 COVID-19 patients and 784 patients without disease, our best-performing algorithm is based on a combination of DeepLabV3+ for lung segmentation and MA-Net for disease segmentation. The proposed algorithm's mean absolute error (MAE) of 0.3 out of 6.0 is significantly reduced compared to established COVID-19 algorithms; BS-net and COVID-Net-S, which have MAEs of 2.5 and 1.8, respectively. Moreover, the proposed two-step workflow was not only more accurate but also computationally efficient, being approximately 11 times faster than the mentioned methods. In conclusion, we have proposed an accurate, time-efficient, and versatile approach for segmentation and scoring of lung diseases illustrated for COVID-19 and with broader future applications for pneumonia, tuberculosis, pneumothorax, and others.

Collaborators: Alexander Proutski (Quantori, Hague, Netherlands), David Nefaridze (Quantori, Tbilisi, Georgia), Diana Litmanovich (Beth Israel Deaconess Medical Center, Boston, United States), Yuriy Gankin (Quantori, Cambridge, United States)

Project type: Commercial / Research

Media: Journal paper, GitHub repo

Figure 1. Schematic illustration of the proposed workflow