Please note that the ITEA Office will be closed from 25 December 2024 to 1 January 2025 inclusive.
Quantib deep-learning framework
- Project
- 16016 STARLIT
- Type
- New standard
- Description
- The Quantib deep-learning framework facilitates the straightforward training of neural networks by 1) Pre-processing all input images to a consistent image spacing necessary for training, 2) Augmenting the input images to increase variability, 3) Allowing the user to set arbitrary hyperparameters for training, including the used optimizers, learning rate and number of epochs for training
- The trained neural networks, created in the previous step, can be applied to input images to predict the corresponding labels
- Contact
- Jorrit Glastra
- j.glastra@quantib.com
- Technical features
Input(s):
- Medical images
- Associated label images
Main feature(s):
- The Quantib deep-learning framework facilitates straightforward training of neural networks
- Perform model inference on provided medical images
- The architecture of the framework allows easy extension of functionality as well as adapting new network topologies
Output(s):
- A trained neural network
- Resulting label image after applying model inference
- Integration constraints
- Various Python libraries are required to use the deep-learning framework, including NumPy, SimpleITK, Keras and TensorFlow
- Because model training is computationally expensive, high quality Graphics Processing Units (GPU’s) are necessary
- The training of neural networks requires large amounts of (consistent) medical image data
- Targeted customer(s)
Researchers / developers that have access to medical images seeking to develop AI solutions.
- Conditions for reuse
Licensing
- Confidentiality
- Public
- Publication date
- 04-09-2020
- Involved partners
- Quantib BV (NLD)