Related Experiment Video
Updated: Sep 1, 2025
06:27
Transcutaneous Microcirculatory Imaging in Preterm Neonates
Published on: December 31, 2015
8.3K
Generating depth images of preterm infants in given poses using GANs
Giuseppe Pio Cannata1, Lucia Migliorelli1, Adriano Mancini1
1Department of Information Engineering, Università Politecnica delle Marche, Italy.
Computer Methods and Programs in Biomedicine
|August 11, 2022
Summary
This study introduces a deep learning framework using Generative Adversarial Networks to create synthetic depth images of preterm infants. This approach addresses the lack of data for developing advanced infant movement monitoring systems.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Developmental Pediatrics
Background:
- Deep learning for preterm infant monitoring can aid early detection of motor and behavioral disorders.
- Development of these algorithms is hindered by a scarcity of annotated datasets.
Purpose of the Study:
- To present a Generative Adversarial Network (GAN)-based framework for generating synthetic depth images of preterm infants in specific poses.
- To overcome the limitations posed by the lack of publicly available annotated datasets for deep learning model training.
Main Methods:
- A novel framework utilizing a bibranch encoder and a conditional Generative Adversarial Network was developed.
- The framework generates both rough and refined depth images of preterm infants.
Main Results:
- The framework achieved a low Fréchet inception distance (142.9) on the Moving INfants In RGB-D dataset.
- Generated images exhibited an inception score (2.8) comparable to real-image distributions (2.6).
Conclusions:
- The developed framework demonstrates potential in generating realistic synthetic depth images of preterm infants.
- Further research in synthetic data generation can lead to more advanced deep learning-based infant monitoring systems.
