Generative Slice Reconstruction for Accelerated, Patient-Centred MRI

Overview

This project aims to significantly reduce MRI scan times by developing a generative AI framework capable of reconstructing full 3D MRI volumes from sparsely acquired slices. By leveraging advanced diffusion models, transformer-based architectures, and physics-informed reconstruction, the system generates high-fidelity, anatomically consistent images while preserving diagnostic accuracy.

The approach incorporates uncertainty-aware reconstruction to support clinical decision-making and integrates k-space data consistency to ensure safety and reliability. Validation will be conducted using large, diverse MRI datasets and include both quantitative image quality metrics and radiologist-led clinical evaluation. The proposed solution targets improved patient comfort, reduced motion artifacts, and increased scanner throughput, supporting the delivery of faster, safer, and more patient-centered MRI examinations.