2023/09/29 | Research | Artificial Intelligence
To improve the initial patient assessment and disease characterization from lung CT scans, the Medical Image Analysis lab of the ARTORG Center, Uni Bern, and the Department of Radiology of the Inselspital, Bern University Hospital have developed a 2-stage multiclass lung lesion model trained to classify disease severity based on the WHO Clinical Progression Scale. The models performance exceeded that of a single-class model as well as the radiologists’ assessment.
The proposed method, AssessNet-19, was developed in 2 stages in this retrospective study. Four COVID-19–induced tissue lesions were manually segmented to train a 2D-U-Net network for a multiclass segmentation task followed by extensive extraction of radiomic features from the lung lesions. The model was evaluated using two multicenter cohorts: a development cohort of 145 COVID-19–positive patients from three centers to train and test and an evaluation set of 90 COVID-19–positive patients from two centers to evaluate AssessNet-19 in a fully automated fashion.
In the evaluation set, the model’s severity assessment was more accurate than that of three expert thoracic radiologists. The model showed good performance in identifying four different lesion types and its verdict on disease extent was in high agreement with the radiologists.
Link to the study
Medical Image Analysis