Computing Infection Distributions and Longitudinal Evolution Patterns in Lung CT Images

Dongdong Gu, Liyun Chen, Fei Shan, Liming Xia, Jun Liu, Zhanhao Mo, Fuhua Yan, Bin Song, Yaozong Gao, Xiaohuan Cao, Yanbo Chen, Ying Shao, Miaofei Han, Bin Wang, Guocai Liu, Qian Wang, Feng Shi, Dinggang Shen, Xue ZhongResearch 2020 · DOI

Introduction

The global impact of COVID-19 necessitated rapid advancements in medical imaging analysis. While radiologists can qualitatively identify characteristic features like ground-glass opacities (GGO) and consolidation, performing large-scale, voxel-level quantitative analysis remains a significant challenge. Anatomical differences between patients and the dynamic nature of the disease make it difficult to compare infections across populations or track precise longitudinal changes. This study introduces an automated pipeline designed to segment lung infections and register them onto a common anatomical template, enabling the creation of statistical "probability maps" of the disease.

The Automated Pipeline

The researchers developed a three-stage framework to standardize the analysis of lung CT images, transforming raw clinical data into statistical maps.

1. Segmentation with VB-Net

At the core of the pipeline is VB-Net, a modified 3D convolutional neural network. To overcome the bottleneck of manual annotation, the team employed a Human-Involved-Model-Iterations (HIMI) strategy. In this approach, AI-generated masks were refined by radiologists and then used to retrain the model. This iterative process reduced the time required for lung infection annotation from 211 minutes to just 4.7 minutes per case, while maintaining a high Dice coefficient of 91.6%, which is close to human inter-rater performance.

2. Deformable Registration

To enable population-level comparisons, individual CT scans were warped onto a pre-defined "template" lung using deformable registration. By aligning the lung fields of different patients into a single coordinate system, the researchers ensured that specific spatial coordinates represented the same anatomical regions across the entire dataset.

3. Statistical Mapping

Once aligned, the system calculated the probability of infection at every voxel. This allowed for rigorous statistical comparisons between groups—such as COVID-19 versus community-acquired pneumonia (CAP)—and the tracking of lesion intensity shifts (measured in Hounsfield Units) over time.

Overview of the proposed pipeline for automated segmentation, registration, and statistical mapping of lung infections.

Clinical Insights and Results

The application of this pipeline to a dataset of thousands of images yielded several key findings regarding the spatial and temporal nature of COVID-19.

The Spatial Signature of COVID-19

The statistical maps revealed that COVID-19 infections are not randomly distributed. Instead, they show a strong preference for the peripheral, posterior, and middle-lower lobes of the lungs. In contrast, Community-Acquired Pneumonia (CAP) tends to be more localized near the diaphragm. COVID-19 lesions were found to be more widespread, with the highest probability of infection reaching 35.1% in specific peripheral subpleural regions.

Early Prediction Challenges

A critical finding of the study was the similarity between "severe" (hospitalized) and "critical" (ICU) patients at the onset of the disease. At the initial imaging stage, only 0.1% of the lung volume showed a statistically significant difference between these two groups. This suggests that early-stage imaging alone may be insufficient to predict clinical decline, highlighting the need for longitudinal monitoring.

Longitudinal Evolution Patterns

By tracking patients over the course of their illness, the researchers identified a distinct four-stage evolutionary pattern in critical cases:

  1. Progression: Rapid increase in lesion volume and density.
  2. Absorption: An initial phase of lesion reduction.
  3. Enlargement: A secondary peak or complication phase, typically occurring around days 14–17.
  4. Further Absorption: Final resolution of the infection.

During these stages, the researchers observed significant shifts in Hounsfield Units (HU). GGOs often transitioned into denser consolidations during the peak phases before fading back into GGOs or disappearing during recovery.

Conclusion

The developed pipeline provides a robust tool for visualizing and quantifying the spatial patterns of lung infections. By standardizing CT data through automated segmentation and registration, the study offers a population-level atlas of COVID-19. While the current registration relies primarily on lung surfaces, future work incorporating internal landmarks like airways and vessels could further refine these maps. Ultimately, these statistical features could serve as the foundation for predictive models to guide clinical decision-making and treatment strategies.

Summary updated

Abstract

Abstract Background: Spatial and temporal lung infection distributions of coronavirus disease 2019 (COVID-19) and their changes could reveal important patterns to better understand the disease and its time course. This paper presents a pipeline to analyze statistically these patterns by automatically segmenting the infection regions and registering them onto a common template. Methods: A VB-Net is designed to automatically segment infection regions in CT images. After training and validating the model, we segmented all the CT images in the study. The segmentation results are then warped onto a pre-defined template CT image using deformable registration based on registering CT images within the lung fields. Then, the spatial distributions of infection regions and those during the course of the disease are calculated at the voxel level. Visualization and quantitative comparison can be performed between different groups. As a result, we compared the distribution maps between COVID-19 and community acquired pneumonia (CAP), between severe and critical COVID-19, and across different course of the disease. Results: For the performance of infection segmentation, comparing the segmentation results with manually annotated ground truth, the average Dice is 91.6%±10.0%, which is close to the inter-rater difference between two radiologists (the Dice is 96.1%±3.5%). The distribution map of infection regions shows that high probability regions are in the peripheral subpleural (up to 35.1% in probability). COVID-19 GGO lesions are more widely spread than consolidations, and the latter are located more peripherally. Onset images of severe COVID-19 (inpatients) show similar lesion distributions but with smaller areas of significant difference in the right lower lobe compared to critical COVID-19 (intensive care unit patients). About the disease course, critical COVID-19 patients showed four distinct patterns (progression, absorption, enlargement, and further absorption) with remarkable concurrent HU patterns for GGO and consolidations. Conclusions: By segmenting the infection regions with a VB-Net and registering all the CT images and the segmentation results onto a template, spatial distribution patterns of infections can be computed automatically. The algorithm provides an effective tool to visualize and quantify the spatial patterns of lung infection diseases and their changes during the disease course. Our results demonstrate different patterns between COVID-19 and CAP, between severe and critical COVID-19, as well as four distinct disease course patterns of the severe COVID-19 patients studied, with remarkable concurrent HU patterns for GGO and consolidations.
BibTeX
@inproceedings{gu2020computing,
  author = {Dongdong Gu and Liyun Chen and Fei Shan and Liming Xia and Jun Liu and Zhanhao Mo and Fuhua Yan and Bin Song and Yaozong Gao and Xiaohuan Cao and Yanbo Chen and Ying Shao and Miaofei Han and Bin Wang and Guocai Liu and Qian Wang and Feng Shi and Dinggang Shen and Xue Zhong},
  title = {{Computing Infection Distributions and Longitudinal Evolution Patterns in Lung CT Images}},
  booktitle = {Research},
  year = {2020},
  doi = {10.21203/rs.3.rs-131598/v1},
}
Figure