Related Experiment Video
Updated: Aug 5, 2026
08:05
Lung CT Segmentation to Identify Consolidations and Ground Glass Areas for Quantitative Assesment of SARS-CoV Pneumonia
Published on: December 19, 2020
14.2K
External validation, radiological evaluation, and development of deep learning automatic lung segmentation in
Krit Dwivedi1,2, Michael Sharkey3, Samer Alabed4
1Department of Infection, Immunity & Cardiovascular Disease, Medical School, University of Sheffield, Sheffield, UK. k.dwivedi@sheffield.ac.uk.
European Radiology
|September 29, 2023
Summary
This study developed an accurate CT pulmonary angiography (CTPA) lung segmentation model. The model demonstrated minimal clinical errors in diverse patient cohorts with pulmonary hypertension and interstitial lung disease, aiding clinical translation.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Radiology
- Pulmonary Disease Diagnosis
Background:
- Accurate lung segmentation in CT pulmonary angiography (CTPA) is crucial for clinical applications.
- Evaluating AI model performance, limitations, and failure points is essential for clinical translation.
- This study addresses the need for robust CTPA lung segmentation models in diverse patient populations.
Purpose of the Study:
- To develop and evaluate an accurate CTPA lung segmentation model using nnU-Net.
- To assess the model's performance in two distinct patient cohorts: pulmonary hypertension (PH) and interstitial lung disease (ILD).
- To perform radiological evaluation of the model's outputs to understand clinical significance and limitations.
Main Methods:
- Retrospective development of an nnU-Net based segmentation model using multi-center data.
- Training, testing, and clinical evaluation on a combined cohort of 225 PH patients and 28 ILD patients.
- Quantitative assessment using Dice Score Coefficient (DSC) and Normalized Surface Distance (NSD), followed by radiologist review for clinical significance.
Main Results:
- The model achieved high accuracy with mean DSC of 0.990 and NSD of 0.983.
- No segmentation failures were observed across all cases.
- Radiological review indicated clinically insignificant errors in 18% (internal) and 25% (external) of cases, with only one (0.5%) clinically significant error.
Conclusions:
- The developed CTPA lung segmentation model is accurate and externally validated.
- The model shows minimal clinical errors, supporting its use in clinical practice for lung volume and disease quantification.
- Robust radiological review is vital for the clinical translation of AI models in medical imaging.
