Latent Interpolation Learning Using Diffusion Models for Cardiac Volume Reconstruction

2026-07-27 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation, Medical Imaging & Reconstruction · Depth: Expert, extended

Summary

A novel Cardiac Latent Interpolation Diffusion (CaLID) framework is introduced for 3D and 3D+T cardiac volume reconstruction from sparse 2D Cardiac Magnetic Resonance (CMR) images. This data-driven approach, based on diffusion models, operates in the latent space to capture complex, non-linear relationships between sparse slices, improving reconstruction accuracy. CaLID achieves a 24x speedup in 3D whole-heart upsampling time, reducing computational overhead by requiring only 8 diffusion steps. It also attains state-of-the-art performance using only sparse 2D CMR images as input, eliminating the need for auxiliary data like segmentation labels. The framework extends to 2D+T data for effective spatiotemporal dynamics, demonstrating superior reconstruction quality and efficiency in volumetric evaluations and downstream segmentation tasks on the UK Biobank dataset.

Key takeaway

For AI Scientists and Machine Learning Engineers developing medical imaging solutions, CaLID offers a significantly faster and more accurate approach to 3D cardiac volume reconstruction from sparse MRI. You should consider integrating data-driven latent diffusion models to reduce computational overhead by up to 24x and simplify workflows by eliminating the need for auxiliary inputs, thereby enhancing clinical applicability.

Key insights

CaLID uses a data-driven latent diffusion model for efficient, accurate 3D cardiac reconstruction from sparse 2D MRI, without auxiliary inputs.

Principles

• Data-driven interpolation surpasses fixed schemes.
• Latent space operations enhance computational efficiency.
• Minimal input simplifies medical imaging workflows.

Method

CaLID employs a VAE for latent representation, then a U-Net denoiser with a two-stage conditioning module (τθ1, τθ2) to learn data-driven interpolation trajectories between adjacent slices.

In practice

• Reconstruct 3D cardiac volumes from sparse 2D MRI.
• Generate temporally coherent 2D+T sequences.
• Improve downstream segmentation accuracy.

Topics

• Cardiac MRI
• Diffusion Models
• 3D Reconstruction
• Latent Space
• Medical Imaging
• Spatiotemporal Modeling

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Anatomy-Guided Residual Motion Diffusion for Controllable 4D Cardiac MRI Synthesis — A cascaded latent diffusion model synthesizes controllable, anatomically consistent 4D cardiac MRI, improving AI model robustness with limited data.
• Diffusion Models in Medical Image Inpainting: Challenges, Solution Taxonomy, and Future Directions — Diffusion models are highly effective for medical image inpainting, generating anatomically consistent reconstructions despite validation challenges.
• Bi-PT: Bidirectional Cross-Attention Point Transformers for Four-Chamber Heart Reconstruction from Sparse Cardiac MRI Data — Bi-PT reconstructs 3D heart meshes from sparse CMR data using bidirectional cross-attention, semantic labels, and a Neural ODE for robust, diffeomorphic deformation.
• CONFLUX: A Latent Diffusion Model for 3D Chest-CT Synthesis with RL Post-Training — Latent diffusion with RL post-training significantly enhances controllable 3D medical image synthesis fidelity and faithfulness.
• Recover Semantics First, Generate Better: Improved Latent Modeling for 3D MRI Reconstruction and Cross-Contrast Synthesis — Prioritizing semantic recovery in 3D MRI latent space compression significantly improves reconstruction fidelity and cross-contrast synthesis.
• Hundreds of cardiac MRI traits derived using 3D diffusion autoencoders share a common genetic architecture — 3D diffusion autoencoders enable unsupervised discovery of heritable cardiac MRI phenotypes and their genetic basis.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.