Related Experiment Video
Updated: Oct 4, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
Polycystic liver: automatic segmentation using deep learning on CT is faster and as accurate compared to manual
Bénédicte Cayot1,2, Laurent Milot3,4, Olivier Nempont3,5
1Department of Medical Imaging, Hospices Civils de Lyon, University of Lyon, Lyon, France. benedicte.cayot@chu-lyon.fr.
European Radiology
|February 10, 2022
Summary
This study developed an artificial intelligence (AI) model for faster, automated segmentation of polycystic livers in CT scans. The AI achieved comparable performance to manual methods, significantly reducing segmentation time.
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Medicine
- Computational Pathology
Background:
- Polycystic liver disease (PLD) requires accurate liver volumetry for patient management.
- Manual segmentation of polycystic livers on CT images is time-consuming and prone to inter-observer variability.
- Deep learning offers potential for automating complex medical image segmentation tasks.
Purpose of the Study:
- To develop and evaluate a deep learning model, specifically a convolutional neural network (CNN), for automated segmentation of polycystic livers in CT imaging.
- To compare the performance and efficiency of the CNN-based segmentation against manual segmentation by expert radiologists.
Main Methods:
- A retrospective study utilizing CT images from 88 consecutive patients with polycystic livers.
- Supervised training and validation of the CNN model using 190 CT series, with performance assessment on 41 additional CT series.
- Comparison of CNN segmentation with manual segmentations by two radiologists (Rad1a, Rad2) and assessment of intra-observer variability (Rad1b).
- Evaluation metrics included Dice Similarity Coefficient (DSC) for overlap and Concordance Correlation Coefficient (CCC) for volume agreement. Segmentation time was recorded.
Main Results:
- The CNN model achieved a high Dice Similarity Coefficient (DSC) of 0.95 ± 0.03.
- Volume analysis showed excellent agreement between CNN segmentation and manual reference, with a Concordance Correlation Coefficient (CCC) of 0.995.
- No statistically significant difference was found in CCC between CNN automatic segmentation and manual segmentations (inter-observer and intra-observer).
- Automated segmentation using CNNs took significantly less time (2.0–5.0 seconds) compared to manual segmentation (22.4 ± 10.4 minutes).
Conclusions:
- Automated segmentation of polycystic livers using a deep learning (CNN) approach is significantly faster than manual segmentation.
- The deep learning method demonstrates comparable performance to expert manual segmentation, with no statistical difference in accuracy.
- This AI-driven approach offers a promising tool for efficient and reliable liver volumetry in polycystic liver disease.
