Artificial Intelligence in Medical Imaging (T5)

Objectives
• Development of semi-automatic segmentation and annotation methods based on Foundation Models
• Domain adaptation learning
• Semi-supervised learning, weakly supervised learning, and few-shot learning
• Development of advanced deep learning models for medical image segmentation and classification based on transfer learning, pre-trained Large Vision Models (LVM), and optimization

Research challenges / Novelty / Innovation
Challenges: data quality and volume, annotation complexity, data multimodality, data heterogeneity, imbalanced classes, generalization capability, clinical impact of results

Research results:
• Training datasets
• New models for imaging modalities
• Demonstrator for 2D image segmentation/annotation exploiting Large Vision Model (LVM) approaches (TRL4)

Inovation:
• Original models for handling imaging modalities