[Paper Review] Coronavirus Detection and Analysis on Chest CT with Deep Learning

Ophir Gozes, Maayan Frid-Adar+4 more|arXiv (Cornell University)|Apr 6, 2020

COVID-19 diagnosis using AIMedicine7 references104 citations

TL;DR

This paper presents a deep learning pipeline that detects, localizes, and quantifies COVID-19 manifestations on chest CTs using 2D slice classification, Grad-CAM based localization, and a corona score for case-wise severity estimation, evaluated on Chinese datasets.

ABSTRACT

The outbreak of the novel coronavirus, officially declared a global pandemic, has a severe impact on our daily lives. As of this writing there are approximately 197,188 confirmed cases of which 80,881 are in "Mainland China" with 7,949 deaths, a mortality rate of 3.4%. In order to support radiologists in this overwhelming challenge, we develop a deep learning based algorithm that can detect, localize and quantify severity of COVID-19 manifestation from chest CT scans. The algorithm is comprised of a pipeline of image processing algorithms which includes lung segmentation, 2D slice classification and fine grain localization. In order to further understand the manifestations of the disease, we perform unsupervised clustering of abnormal slices. We present our results on a dataset comprised of 110 confirmed COVID-19 patients from Zhejiang province, China.

Motivation & Objective

• Develop a lightweight, weakly supervised deep learning system to detect COVID-19 features in chest CT scans.
• Localize abnormal lung tissue and provide a quantitative disease severity score.
• Enable case-wise analysis by aggregating slice-level results into a 3D localization map and corona score.
• Explore disease patterns via unsupervised clustering of abnormal and normal slices.
• Utilize available datasets to validate slice-level, case-level detection and severity estimation.

Proposed method

• Lung region localization and ROI extraction using a U-Net with a pre-trained ImageNet encoder.
• 2D ResNet-50 classifier (pre-trained on ImageNet) to label lung ROIs as normal vs abnormal (COVID-19).
• Multi-scale Grad-CAM to produce fine-grained localization maps, fused to create a 3D corona score.
• Creation of a volumetric disease extent measure by summing localized activations above a threshold across slices.
• Unsupervised clustering (k-means) in a Grad-CAM weighted feature space to explore disease manifestations.
• Evaluation on datasets including development and testing cohorts with and without COVID-19 confirmed cases.

Experimental results

Research questions

• RQ1 Can a 2D slice-based classifier detect COVID-19 manifestations on chest CT with high AUC, sensitivity, and specificity?
• RQ2 Can multi-scale Grad-CAM localization provide accurate fine-grained localization to support radiologists?
• RQ3 Does a volumetric corona score correlate with case-level disease presence and radiologist severity grades?
• RQ4 What unsupervised structure emerges from Grad-CAM weighted features regarding disease manifestations (normal, focal, diffuse)?

Key findings

• Slice-level COVID-19 detection achieved AUC of 0.994 with 94% sensitivity and 98% specificity (threshold 0.5).
• Corona score yielded AUC 0.948 (95% CI: 0.912-0.985) in distinguishing COVID-19 vs non-COVID-19 cases.
• Corona score distributions differ between severe and non-severe cases (median scores: 61.5 cm3 vs 227.5 cm3; p = 0.0064).
• 3D localization maps provide a case-wise visualization of disease extent to assist radiologists.
• Unsupervised clustering revealed distinct groups corresponding to Normal, Focal, and Diffuse manifestations.
• The approach uses weak slice-level annotations and Grad-CAM guidance to learn disease patterns.