[Paper Review] Lung Infection Quantification of COVID-19 in CT Images with Deep Learning

Fei Shan, Yaozong Gao+7 more|arXiv (Cornell University)|Mar 10, 2020

COVID-19 diagnosis using AI35 references537 citations

TL;DR

A deep learning segmentation system automatically quantifies COVID-19 lung infection on CT scans, achieving high agreement with manual annotations and low infection percentage error, aided by a human-in-the-loop strategy to accelerate training and correction.

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.

Motivation & Objective

• Develop a DL-based system to automatically quantify infection ROIs and their volumetric ratios in the lung from CT images.
• Provide automatic segmentation with minimal manual intervention.
• Enable rapid training sample delineation and corrective updates via HITL to accelerate workflow.

Proposed method

• Train a deep learning segmentation model to identify infection regions in chest CT scans.
• Compare automatic segmentations to manual delineations on 300 scans from 300 patients.
• Incorporate a human-in-the-loop (HITL) strategy to assist radiologists for faster delineation and model updates (3 iterations).
• Evaluate performance with Dice similarity coefficient between automatic and manual segmentations.
• Report mean error in percentage of infection (POI) for the whole lung.

Experimental results

Research questions

• RQ1 Can a DL-based segmentation system accurately quantify infection regions in COVID-19 CT images?
• RQ2 What is the agreement between automatic and manual infection segmentations on a large patient set?
• RQ3 Does a HITL approach reduce manual delineation time and improve segmentation updates over iterations?
• RQ4 What is the accuracy of infection fraction estimation (POI) for the whole lung?

Key findings

• Average Dice similarity coefficient between automatic and manual infection segmentations is 91.6%.
• Mean estimation error of percentage of infection (POI) for the whole lung is 0.3%.
• HITL approach substantially reduces total segmentation time to 4 minutes after 3 iterations of model updating.