COVID-19 lung infection segmentation from chest CT images based on CAPA-ResUNet

Abstract

Coronavirus disease 2019 (COVID-19) epidemic has devastating effects on personal health around the world. It is significant to achieve accurate segmentation of pulmonary infection regions, which is an early indicator of disease. To solve this problem, a deep learning model, namely, the content-aware pre-activated residual UNet (CAPA-ResUNet), was proposed for segmenting COVID-19 lesions from CT slices. In this network, the pre-activated residual block was used for down-sampling to solve the problems of complex foreground and large fluctuations of distribution in datasets during training and to avoid gradient disappearance. The area loss function based on the false segmentation regions was proposed to solve the problem of fuzzy boundary of the lesion area. This model was evaluated by the public dataset (COVID-19 Lung CT Lesion Segmentation Challenge-2020) and compared its performance with those of classical models. Our method gains an advantage over other models in multiple metrics. Such as the Dice coefficient, specificity (Spe), and intersection over union (IoU), our CAPA-ResUNet obtained 0.775 points, 0.972 points, and 0.646 points, respectively. The Dice coefficient of our model was 2.51% higher than Content-aware residual UNet (CARes-UNet). The code is available at https://github.com/malu108/LungInfectionSeg.

Keywords: COVID‐19; area loss function; computed tomography (CT) image segmentation; deep learning; pre‐activated residual block.

FIGURE 1

The network architecture of the proposed CAPA‐ResUNet.

FIGURE 2

Modules illustration. (A) The structure of the residual block. (B) The structure of the proposed pre‐activated residual block. The dotted line on the right is residual connection.

FIGURE 3

The proposed loss function schematic diagram. ${\Omega }_1$ and ${\Omega }_2$ are masks, ${\Omega }_2$ and ${\Omega }_3$ are predictions. The area of ${\Omega }_1$ and ${\Omega }_3$ are calculated by using the concept of mathematical calculus. Our purpose is to continuously reduce the area of ${\Omega }_1$ and ${\Omega }_3$.

FIGURE 4

CT image segmentation results of all models. The first column is the original CT images, and the second column is their corresponding masks. Columns from the third to the last are predicted results of the different models. Orange outlines indicate ground truth contours.

FIGURE 5

Experimental results of ablation studies. “Baseline,” namely CARes‐UNet, is trained with only the combo loss, and “AL” represents the area loss based on error segmentation regions. “PA” represents the pre‐activated residual block module. “Baseline + AL + PA” is our CAPA‐ResUNet.