CIMIM is off to MICCAI 2026!
We are delighted to announce that three CIMIM papers have been accepted for presentation at MICCAI 2026 in Strasbourg, France. Congratulations to Shaokun Lan, a second-year PhD student supervised by Professor Alex Frangi, who has been invited to give an oral presentation. Two further papers, led by Dr Fengming Lin, a postdoctoral fellow, and Jiayang Lu, an MSc student, will be presented as posters. Further details of CIMIM’s contributions to MICCAI 2026 are provided below.
Conditional Latent Diffusion Model with Fourier-based Motion Modelling for Virtual Population Synthesis (Oral Presentation)
Shaokun Lan#, Haoran Dou#, Jinghan Huang, Arezoo Zakeri, Fengming Lin, Zherui Zhou, Jinming Duan, Alejandro F. Frangi
This paper introduces 4D F-LDM, designed to generate highly realistic, beating 3D virtual hearts. These virtual hearts are primarily used for “in-silico trials”, which are computer simulations that allow scientists to safely test new medical devices before using them on real patients. Traditionally, AI-generated virtual hearts struggle to create a perfect, continuous heartbeat cycle. When a heartbeat finishes, the 3D model often fails to return to its exact starting position, creating a glitch known as “drift” that makes medical simulations inaccurate. To solve this, the authors applied a mathematical concept called a “Fourier series” to the motion tracking. This method strictly guarantees that the generated cardiac cycle without any drift. Furthermore, customize these virtual hearts can be generated based on specific patient traits, like age and gender. Tested on 5,000 subjects, this model significantly outperformed previous methods, offering a highly reliable solution for future medical testing.
HeartVolMesh: Cardiac Volumetric Mesh Reconstruction via Covariance-Guided Graph Deformation
Fengming Lin, Arezoo Zakeri, Haoran Dou, Zherui Zhou, Shaokun Lan, Jinming Duan, Alejandro Frangi
A paper led by Fengming Lin published at MICCAI 2026 introduces HeartVolMesh, a framework for reconstructing patient-specific tetrahedral cardiac meshes from volumetric medical images. Accurate cardiac meshes are essential for digital twins and in-silico trials, but conventional segmentation-based pipelines may lose thin anatomical structures and provide limited correspondence across patients. HeartVolMesh combines 3D image analysis with graph-based deformation to transform a fixed cardiac template into an anatomically accurate volumetric mesh while preserving its connectivity and cross-patient correspondence. Experiments demonstrate improved surface accuracy and volumetric mesh fidelity over existing deformation-based approaches. By producing reliable, resolution-controllable cardiac models, this work supports more robust computational simulations and the future development of personalised cardiovascular medicine.
DINO-3DRA: Leveraging 2D Foundation Model Semantics for 3D Cerebral Aneurysm Segmentation
Jiayang Lu, Fengming Lin∗, Alejandro F. Frangi∗, and Ali Sarrami-Foroushani∗†
A paper led by MSc student Jiayang Lu published at MICCAI 2026 introduces DINO-3DRA, a framework for cerebral aneurysm segmentation in 3D rotational angiography. Aneurysms are difficult to identify because they are small and often resemble surrounding vessels. DINO-3DRA transfers structural knowledge from DINOv3, a foundation model trained on approximately 1.7 billion 2D images, into an efficient 3D segmentation network. It achieves state-of-the-art performance on multi-centre data with only 5.72 million trainable parameters and generalises to external datasets without fine-tuning. This work demonstrates how large 2D foundation models can improve reliable, data-efficient 3D medical image analysis.