Synthetic Multi-timepoint AI-enhanced Breast MRI

Objetivo

Smart-MRI (Synthetic Multi-timepoint AI-enhanced Breast MRI) aims to eliminate the need for gadolinium-based contrast agents while preserving the diagnostic accuracy of Dynamic Contrast-Enhanced MRI (DCE-MRI).
Smart-MRI consists of three key phases. The first phase focuses on AI model optimization and external validation. Our deep learning model is trained on a large dataset of paired contrast-enhanced and non-contrast breast MRI scans, designed to capture essential temporal and morphological features for lesion characterization. By reducing imaging complexity and relying only on pre-contrast T1-weighted and DWI sequences, Smart-MRI streamlines imaging protocols, supporting feasibility for widespread clinical adoption.
The second phase involves a multi-center external validation in a screening setting to assess the generalizability and clinical applicability of Smart-MRI across different imaging systems and diverse patient populations. Structured reader studies are conducted with expert radiologists to compare Smart-MRI-generated synthetic contrast-enhanced images with standard DCE-MRI for this purpose. These studies evaluate lesion detectability, image fidelity, and diagnostic confidence.
In the third, and overlapping phase, we assess market potential, and analyse barriers for integration into routine clinical practice, aiming to create an initial prospective integration into existing imaging platforms and workflows and translate the innovation into a usable product in collaboration with our technology transfer office and potential industrial partners.
If successful, Smart-MRI is expected to transform breast cancer imaging by offering a contrast-free alternative to DCE-MRI without compromising diagnostic performance. By eliminating Smart-MRI is an addition to the ERC Consolidator project SAFE-MRI, focussing on the optimisation and validation of a market ready synthetic breast MRI solution, enabling widely accessible contrast-free breast MRI as a screening solution.