Related Experiment Video
Updated: Nov 20, 2025
07:53
Author Spotlight: Advancing 3D Modeling for Enhanced Diagnosis and Treatment of Pulmonary Nodules in Early-Stage Lung Cancer
Published on: October 13, 2023
1.8K
Building Large-Scale Quantitative Imaging Databases with Multi-Scale Deep Reinforcement Learning: Initial Experience
David J Winkel1, Hanns-Christian Breit2, Thomas J Weikert2
1Department of Radiology, University Hospital of Basel, Basel, Basel-Stadt, Switzerland. davidjean.winkel@usb.ch.
Journal of Digital Imaging
|January 20, 2021
Summary
Deep reinforcement learning (DRL) automates whole-body organ volume analysis from CT scans. This fast, robust method is unaffected by contrast phase or slice thickness, aiding quantitative imaging follow-up.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiology
Background:
- Accurate organ volumetry is crucial for quantitative imaging and disease monitoring.
- Traditional methods for organ volume calculation can be time-consuming and labor-intensive.
- Deep reinforcement learning (DRL) offers potential for automating complex image analysis tasks.
Purpose of the Study:
- To assess the feasibility of a fully automated workflow for whole-body organ volumetric analysis using DRL.
- To investigate the impact of contrast phase (CP) and slice thickness (ST) on DRL-derived organ volumes.
- To establish the speed and robustness of the DRL approach for large-scale volumetric database creation.
Main Methods:
- Retrospective analysis of 431 multiphasic CT datasets, encompassing 10,508 organ volumes.
- Utilized multi-scale DRL for 3D anatomical landmark detection and 3D organ segmentation.
- Performed repeated measures analyses of variance (ANOVA) to evaluate the influence of CP and ST.
Main Results:
- The DRL algorithm successfully calculated organ volumes for the liver, spleen, kidneys, and lungs.
- No statistically significant effects of variable CP or ST were observed on the calculated organ volumes.
- Mean computational time per case was a rapid 10 seconds.
Conclusions:
- A fully automated DRL workflow is feasible for whole-body organ volumetric analysis.
- The DRL approach is robust and efficient, unaffected by variations in CP and ST.
- This method enables rapid processing of large datasets, facilitating the creation of organ-specific volumetric databases for quantitative imaging follow-up.
