Automatic Evaluation of Chest Severity using Lung Volumetric Parenchymal Infections Semantic Segmented Images from High Resolution Computed Tomography

Chest infections must be diagnosed at an early stage for preventive measurement. Delay in interventions due to pathological findings processing time or low contrast scanning images may lead to death. Computed tomography (CT) scans provide early and clear information of the lungs ground glass opacities (GGO) which can lowers the global mortality rate. However, assessments from imaging modalities are subjective; whose precision totally depends on radiologists’ experience and knowledge. Therefore, this paper presents the automatic estimation technique for volumetric parenchymal infection of the lungs from CT slices. The process is based on segmenting the parenchymal anomalies due to pneumonia classes and their severity. Based on the percentage of the lungs volume being affected, computed tomography severity score (CTSS) can be evaluated and standardized. A deep-learning method is proposed for standardizing CTSS score predictor algorithm and its automatic evaluation. The proposed method uses high resolution computed tomography (HRCT) scans to provide effective clinical classification of patients based on the amount of lung parenchymal involvement by the presence of COVID-19 or other pneumonia infections. Our modified U-Net architecture performs semantic segmentation based classifications of the GGO; which is used to provide the predicted volumetric infections (PVI) values automatically. PVI is correlated with the radiologists labelled CTSS and a relationship is standardized between both to propose an automatic CT severity predictor equation. Percentage root mean square difference (PRD) between actual CTSS and predicted CTSS comes out to be less than 7%, which can help in consistent clinical decisions making.