Kirscher/psat-mixed-continual-learning

PSAT – Mixed Continual Learning

This repository contains the pretrained PSAT model trained with mixed continual learning.

📄 Paper

https://arxiv.org/abs/2507.05764

PSAT: Pediatric Segmentation Approaches via Adult Augmentations and Transfer Learning

🔗 Project

GitHub: https://github.com/ICANS-Strasbourg/PSAT

📦 Model

File: mixed_model_continual_learning.zip

⚙️ Usage

nnUNetv2_install_pretrained_model_from_zip mixed_model_continual_learning.zip

🧪 Training Data

This model was trained using a continual learning strategy combining:

• Adult CT scans from the TotalSegmentator dataset
• Pediatric data from the TCIA Pediatric-CT-Seg dataset

TotalSegmentator dataset: https://zenodo.org/record/6802614

TCIA Pediatric-CT-Seg: https://www.cancerimagingarchive.net/collection/pediatric-ct-seg/

⚠️ Data Usage Notice

The datasets used for training have their own licenses and access conditions. Users must comply with the original dataset terms when using this model.