Related Experiment Video
Updated: Sep 27, 2025
Lung CT Segmentation to Identify Consolidations and Ground Glass Areas for Quantitative Assesment of SARS-CoV Pneumonia
Published on: December 19, 2020
Deep Learning-based Automatic Lung Segmentation on Multiresolution CT Scans from Healthy and Fibrotic Lungs in Mice
Francesco Sforazzini1, Patrick Salome1, Mahmoud Moustafa1
1Clinical Cooperation Unit Radiation Oncology, German Cancer Research Center (DKFZ), Im Neuenheimer Feld 280, 69120 Heidelberg, Germany (F.S., P.S., M.M., C.Z., C.S., K.R., N.B., A.K., J.D., A.A., M.K.); Department of Radiation Oncology (F.S., P.S., M.M., C.Z., C.S., K.R., N.B., A.K., J.D., A.A., M.K.) and National Center for Tumor Diseases (NCT) (F.S., P.S., M.M., C.Z., C.S., K.R., N.B., J.D., A.A., M.K.), Heidelberg University Hospital (UKHD), Heidelberg, Germany; German Cancer Consortium (DKTK) Core Center Heidelberg, Heidelberg, Germany (F.S., P.S., M.M., C.Z., C.S., K.R., J.D., A.A., M.K.); National Center for Radiation Oncology (NCRO), Heidelberg Institute for Radiation Oncology (HIRO), Heidelberg, Germany (F.S., P.S., M.M., C.Z., C.S., K.R., N.B., A.K., J.D., A.A., M.K.); Heidelberg Ion-Beam Therapy Center (HIT), Heidelberg, Germany (F.S., P.S., M.M., C.Z., C.S., K.R., N.B., J.D., A.A., M.K.); Department of Clinical Pathology, Suez Canal University, Ismailia, Egypt (M.M.); Department of Radiation Oncology, Nanfang Hospital, Southern Medical University, Guangzhou, China (C.Z.); and The D-Laboratory and the M-Laboratory, Department of Precision Medicine, GROW-School for Oncology, Maastricht University, Maastricht, the Netherlands (H.W., L.D., P.L.).
A deep learning model accurately segments mouse lungs in CT images, regardless of fibrosis or resolution. This method shows high accuracy for both standard and high-resolution scans, aiding animal studies.
Area of Science:
- Medical Imaging
- Computational Biology
- Radiology
Background:
- Accurate segmentation of mouse lungs in CT images is crucial for preclinical research, especially in studies involving lung fibrosis.
- Existing methods may struggle with varying disease states and image resolutions.
Purpose of the Study:
- To develop and validate a deep learning model for precise mouse lung segmentation.
- To assess the model's performance across different fibrosis levels and CT resolutions.
Main Methods:
- A U-Net deep learning model was trained and validated on a large dataset of mouse CT images.
- The model was tested on independent datasets, including high-spatial-resolution micro-CT images, using Dice score coefficient (DSC) and Hausdorff distance (HD).
- Transfer learning was employed to adapt the model for micro-CT data.
Main Results:
- The model achieved high median DSC scores (0.984 in group A, 0.966 in group B) and low median HD (0.47 mm in group A, 0.31 mm in group B) on standard CT images.
- Segmentation accuracy remained robust for high-resolution micro-CT images, with a median DSC of 0.905.
- Analysis identified specific regions near air-tissue interfaces as prone to segmentation deviation.
Conclusions:
- The developed deep learning method provides accurate and reliable mouse lung segmentation.
- The model is effective across various disease states (healthy, fibrotic) and CT resolutions, demonstrating its utility in animal studies.
