Lung Infection Quantification of COVID-19 in CT Images with Deep Learning (2003.04655v3)

Abstract: CT imaging is crucial for diagnosis, assessment and staging COVID-19 infection. Follow-up scans every 3-5 days are often recommended for disease progression. It has been reported that bilateral and peripheral ground glass opacification (GGO) with or without consolidation are predominant CT findings in COVID-19 patients. However, due to lack of computerized quantification tools, only qualitative impression and rough description of infected areas are currently used in radiological reports. In this paper, a deep learning (DL)-based segmentation system is developed to automatically quantify infection regions of interest (ROIs) and their volumetric ratios w.r.t. the lung. The performance of the system was evaluated by comparing the automatically segmented infection regions with the manually-delineated ones on 300 chest CT scans of 300 COVID-19 patients. For fast manual delineation of training samples and possible manual intervention of automatic results, a human-in-the-loop (HITL) strategy has been adopted to assist radiologists for infection region segmentation, which dramatically reduced the total segmentation time to 4 minutes after 3 iterations of model updating. The average Dice simiarility coefficient showed 91.6% agreement between automatic and manual infaction segmentations, and the mean estimation error of percentage of infection (POI) was 0.3% for the whole lung. Finally, possible applications, including but not limited to analysis of follow-up CT scans and infection distributions in the lobes and segments correlated with clinical findings, were discussed.

Lung Infection Quantification of COVID-19 in CT Images with Deep Learning

In the context of the COVID-19 pandemic, the ability to accurately quantify lung infection from CT images is crucial for efficient patient management. The paper "Lung Infection Quantification of COVID-19 in CT Images with Deep Learning" presents a novel deep learning (DL)-based system leveraging the VB-Net architecture to automatically segment and quantify infection regions in chest CT scans of COVID-19 patients.

Methodology

The paper employs a retrospective dataset comprising 249 COVID-19 patient CT images for training and 300 for validation. The VB-Net neural network, a modification of the V-Net with bottleneck structures, is utilized for the segmentation tasks due to its efficacy in handling large 3D volumetric data. A human-in-the-loop (HITL) strategy is integrated to refine the accuracy of automatic annotations, significantly accelerating the training cycle by iteratively incorporating radiologists' corrections.

Results

The results demonstrate a high Dice similarity coefficient of 91.6%±10.0% between automatic and manual segmentations, indicating substantial accuracy in delineating infection regions. Volume estimation errors and percentage of infection (POI) estimation errors are minimal (0.3% for the whole lung), affirming the reliability of the system. Moreover, the iterative HITL strategy efficiently reduces manual delineation time from hours to under five minutes per scan after sufficient iterations.

Discussion

The implications of this research are pertinent both clinically and theoretically. Practically, the DL-based system facilitates rapid and consistent quantification of lung infections, thereby enhancing radiological assessment and aiding in the longitudinal tracking of disease progression. The potential application of the system extends to evaluating therapeutic responses due to its capability to quantify longitudinal changes in follow-up CT scans accurately.

From a theoretical standpoint, the integration of HITL strategies exemplifies an advancement in interactive AI systems, prioritizing model transparency and radiologist involvement in the training process. The HITL approach enhances model accuracy iteratively, thus providing a practical solution to the challenge of obtaining extensive annotated datasets.

Future prospects could include generalizing this segmentation approach to other types of pneumonia and exploring transfer learning methodologies to adapt the system for broader clinical applications. However, limitations such as the geographic diversity of the dataset and its current specificity to COVID-19 must be addressed through future multi-center collaborations.

Conclusion

This investigation into deep learning for CT-based COVID-19 infection quantification represents a significant step in advancing radiological tools with enhanced precision and efficiency. The system not only complements RT-PCR testing by providing rapid imaging-based assessments but also contributes valuable quantitative data for clinical decision-making and research into disease progression and treatment effects.