Severity Assessment and Progression Prediction of COVID-19 Patients Based on the LesionEncoder Framework and Chest CT

Abstract

Automatic severity assessment and progression prediction can facilitate admission, triage, and referral of COVID-19 patients. This study aims to explore the potential use of lung lesion features in the management of COVID-19, based on the assumption that lesion features may carry important diagnostic and prognostic information for quantifying infection severity and forecasting disease progression. A novel LesionEncoder framework is proposed to detect lesions in chest CT scans and to encode lesion features for automatic severity assessment and progression prediction. The LesionEncoder framework consists of a U-Net module for detecting lesions and extracting features from individual CT slices, and a recurrent neural network (RNN) module for learning the relationship between feature vectors and collectively classifying the sequence of feature vectors. Chest CT scans of two cohorts of COVID-19 patients from two hospitals in China were used for training and testing the proposed framework. When applied to assessing severity, this framework outperformed baseline methods achieving a sensitivity of 0.818, specificity of 0.952, accuracy of 0.940, and AUC of 0.903. It also outperformed the other tested methods in disease progression prediction with a sensitivity of 0.667, specificity of 0.838, accuracy of 0.829, and AUC of 0.736. The LesionEncoder framework demonstrates a strong potential for clinical application in current COVID-19 management, particularly in automatic severity assessment of COVID-19 patients. This framework also has a potential for other lesion-focused medical image analyses.

HFCF‐Net: A hybrid‐feature cross fusion network for COVID‐19 lesion segmentation from CT volumetric images

“…To validate the efficacy of the proposed network in COVID‐19 lesion segmentation, we considered nine state‐of‐the‐art and classical networks for comparison, including: (1) COPLE‐Net 11 that employs the squeeze‐and‐excitation block and Atrous Spatial Pyramid Pooling (ASPP) module to extract features and integrates a self‐ensembling training framework to promote the robustness against noise; (2) Inf‐Net 17 that uses reverse attention module to explore discriminative infection regions and adopts a parallel partial decoder to generate the global map; (3) a weakly supervised segmentation network (Weakly‐Net) 18 based on spatial transformation consistency; (4) a modified lightweight U‐Net with EfficientNetB7 backbone (Eff‐Net) 20 ; (5) 2.5D segmentation network (2.5D‐Net) 37 that decomposes the 3D segmentation problem into three independent 2D segmentation problems; (6) two‐level nested U‐structure network (U 2 ‐Net) 38 ; (7) ConResNet 14 that designs the context residual module to explicitly perceive 3D context to boost the network's ability; (8) 2D U‐Net 26 ; (9) 3D U‐Net 39 ; (10) V‐Net. 40 Note that 2D networks adopted the same training manner as the original papers did, that is, using only slices with lesions for training.…”

“…The 2D‐based segmentation methods explore 2D convolutional neural networks (CNNs) to predict the lesion region of each slice in CT volume data. 16 , 17 , 18 , 19 , 20 , 21 , 22 For example, Wang et al. 11 proposed a novel noise‐robust learning framework based on self‐ensembling of 2D CNN for slice‐by‐slice segmentation.…”

“…The current methods for COVID‐19 lesion segmentation mainly include two‐dimensional (2D)‐based and 3D‐based segmentation methods. The 2D‐based segmentation methods explore 2D convolutional neural networks (CNNs) to predict the lesion region of each slice in CT volume data 16–22 . For example, Wang et al 11 .…”

“…The 2D-based segmentation methods explore 2D convolutional neural networks (CNNs) to predict the lesion region of each slice in CT volume data. [16][17][18][19][20][21][22] For example, Wang et al 11 proposed a novel noise-robust learning framework based on selfensembling of 2D CNN for slice-by-slice segmentation. Fan et al 17 designed a semi-supervised lung infection segmentation deep network (Semi Inf -Net) for CT slices.…”

Diagnosis/Prognosis of COVID-19 Images: Challenges, Opportunities, and Applications

“…Deep Learning: To identify discrete severity scores of the COVID-19 patients, CNN-based models can alternatively developed[82]. A two stage DL framework is proposed in Reference[84] for COVID-19 severity classification. In the first stage, CT scans are individually fed to a U-Net model, whose extracted features are stored for the second stage.…”