SP1: Deep Learning for Advanced X-Ray Imaging
The SP1 sub-project of the IMPACT project leverages deep learning to advance breast and chest X-ray imaging through four key strategic objectives:
Key Objectives
- Image Quality Assessment & Biomarkers (SO1): Development of deep learning models to automate image quality evaluation and extract patient-specific data, such as breast density and biomarkers. These models also support the creation of realistic 3D anthropomorphic phantoms for quality assurance in X-ray imaging.
- Optimization of Emerging Imaging Modalities (SO2): Applying deep learning to enhance Phase-Contrast Imaging (PCI) by optimizing acquisition parameters, improving phase reconstruction, and mitigating artifacts (e.g., Moiré patterns). The focus includes denoising and reconstruction tasks to facilitate the clinical translation of PCI.
- Foundational Models & Limited Data Learning (SO3): Research into foundation models, knowledge transfer, and semi-supervised learning to overcome the scarcity of labeled medical data. Strategies include knowledge distillation, few-shot learning, and data synthesis using state-of-the-art diffusion models and neural cellular automata.
- Computer-Aided Detection (SO4): Development of advanced algorithms for lesion detection, diagnosis, and treatment response prediction. These models incorporate Explainable AI (XAI) techniques—such as feature importance analysis—to improve transparency for radiologists. It also explores dense prediction tasks like segmentation and object detection, integrating real and synthetic data from SO3.
Our Impact
The IMPACT project’s focus on deep learning drives innovation in image analysis, optimizes detection and diagnosis, and promotes personalized healthcare outcomes.
Expert Contribution (UdG)
The SP1 experts from the University of Girona (UdG) provide multidisciplinary expertise in:
- Medical image analysis using Artificial Intelligence, Deep Learning, and Computer Vision.
- Clinical Radiology for the validation and evaluation of developed tools.
- Technology and Software Development for the implementation and accessibility of medical imaging tools, with specialized experience in breast and chest X-ray modalities.
