Related Experiment Video
Updated: Dec 17, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.2K
Automated CT and MRI Liver Segmentation and Biometry Using a Generalized Convolutional Neural Network.
Kang Wang1,2, Adrija Mamidipalli2, Tara Retson1,2
1Artificial Intelligence and Data Analytic Laboratory (AiDA lab), Department of Radiology, University of California, San Diego. La Jolla, CA 92092.
Radiology. Artificial Intelligence
|June 26, 2020
Summary
A novel convolutional neural network (CNN) can accurately segment livers across various imaging types, enabling automated liver biometry. This AI tool shows promise for liver volumetry and tissue characterization in clinical practice.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiology
Background:
- Accurate liver segmentation is crucial for quantitative analysis in clinical practice.
- Current segmentation methods can be time-consuming and operator-dependent.
- Automating liver segmentation across diverse imaging modalities remains a challenge.
Purpose of the Study:
- To assess the feasibility of training a convolutional neural network (CNN) for automated liver segmentation.
- To apply automated segmentation for liver biometry, including volumetry and fat fraction estimation.
- To evaluate the CNN's performance across different imaging modalities and techniques.
Main Methods:
- A 2D U-Net CNN was trained in two stages using 330 abdominal MRI and CT exams.
- Transfer learning was employed to generalize the CNN using contrast-enhanced MRI and CT data.
- Performance was validated on a multi-institutional dataset (n=498) using Dice scores, Pearson correlation, and Bland-Altman analysis.
Main Results:
- High Dice scores achieved: 0.94 ± 0.06 for CT, 0.95 ± 0.03 for T1w MRI, and 0.92 ± 0.05 for T2*w MRI.
- Automated and manual liver volume measurements showed close agreement for CT and T1w MRI.
- Automated and manual hepatic proton density fat fraction (PDFF) estimates also demonstrated strong agreement.
Conclusions:
- A transfer-learning strategy enables a CNN to generalize liver segmentation across various imaging modalities.
- The developed CNN is feasible for automating liver volumetry and hepatic tissue characterization.
- Further validation may lead to broad clinical applicability of CNNs in liver imaging analysis.
