Resolving Through-Plane Motion in Volumetric Analysis

You are reading version 1 of this direction, frozen 2026-10-04. The latest version is version 2.

Version 1 of 2107 papers129 conceptsfrozen 2026-10-04

When we capture a two-dimensional slice of a three-dimensional process—whether it is a beating heart, a shifting glacier, or a migrating cell—we encounter a fundamental "dimensionality gap." A 2D image acts like a window: we can easily track movement that stays within the frame, but if an object moves toward or away from the camera, it appears to shrink, expand, or vanish entirely. This phenomenon, known as through-plane motion, is the central challenge in modern motion estimation research.

The Dimensionality Gap

In medical imaging, particularly MRI and ultrasound, data is often acquired in thin slices. As Sébastien Roujol notes, traditional 2D registration frameworks are fundamentally limited because they cannot recover anatomical information that moves out of the imaging plane. This is not just a technical nuisance; it leads to significant errors in clinical metrics.

For example, the human heart does not just contract; it twists and shortens along its longitudinal axis. Pengcheng Shi and Jan Kybic highlight that 2D tracking fails to capture this torsional twisting, resulting in an incomplete and often misleading representation of cardiac health. Similarly, in cellular biology, Christopher Holmes observes that retinal clumps moving vertically through tissue layers can be mistaken for objects changing size, simply because they are moving through the focal plane.

Bridging the Gap with Multi-View Fusion

The most direct way to resolve through-plane ambiguity is to look at the subject from multiple angles simultaneously. Researchers are moving away from single-slice analysis toward frameworks that fuse disparate 2D views into a coherent 3D model.

A prominent example is the MulViMotion framework developed by Qingjie Meng and Wenjia Bai. By integrating short-axis and long-axis MRI views, their model can estimate the "true" 3D deformation field of the heart, capturing the longitudinal contraction that single-view methods miss. Similarly, Ismail Ben Ayed uses "track-to-track fusion" to combine data from multiple 2D views, allowing standard clinical scans to produce high-fidelity 3D motion maps without requiring specialized 3D imaging hardware.

Physics-Informed Reconstruction

When multiple views aren't available, researchers rely on the "rules of the world"—physics—to fill in the missing dimension. If we know how a tissue is supposed to deform or how a fluid is supposed to flow, we can infer the 3D movement from 2D observations.

  • Biomechanical Constraints: Meng Ye introduced Volumetric Neural Deformable Models ($\upsilon$NDMs), which use deep learning to bridge the gap between "apparent" 2D motion and true 3D trajectories. By constraining the model with anatomical priors, it becomes possible to resolve the ambiguity of sparse 2D data.
  • Fluid and Continuum Dynamics: Jong-Hoon Ahn proposed a "transport-teleport duality" that allows models to learn complex dynamics from as few as two frames, effectively accounting for mass that moves out of the plane of view.
  • Structural Priors: In electrophysiology, Carlos Ruiz Herrera uses Physics-Informed Neural Networks (PINNs) to estimate 3D heart fiber orientation from 2D surface maps, using the physics of electrical propagation to "see" through the thickness of the heart wall.

From Pixels to Latent Dynamics

Recent work has shifted from tracking pixels directly to modeling the underlying "state" of the system. Niklas Gunnarsson and Thomas B. Schön demonstrate that by mapping image sequences into a low-dimensional latent space, models can learn the dynamics of anatomical motion in a way that is more robust to the noise and signal loss caused by through-plane movement.

This shift is critical for real-time applications. For instance, Thibault Lagache developed the EMC² algorithm to track neurons in moving animals. By using visible neurons as "anchors" to map local tissue deformation, the system can predict where "silent" or out-of-focus neurons have moved, maintaining a consistent 3D track even when the imaging plane shifts.

The Path to Native 3D

The consensus across these 92 papers is that 2D-based motion correction is a temporary patch. As Veronika Spieker and Hannah Eichhorn argue in their review of deep learning for MRI, the next frontier is the development of native 3D architectures. Whether it is measuring the mass balance of a glacier Christoph Vogel or diagnosing Parkinson's disease through 3D kinematics Kyungdo Kim, the goal is to move beyond the "window" of the 2D slice and capture the full volumetric reality of a world in motion.

Papers in this version

Showing 60 of 107 papers.

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 74 papers
  2. Version 1 · 2026-10-04 · 107 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_out_of_plane_motion_artifacts_ed62d12b,
  title = {out-of-plane motion artifacts},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/out-of-plane-motion-artifacts/rdr_ed62d12bc0e648499f5ae835e647986b},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_resolving_through_plane_motion_in_volume_ed62d12b_v1,
  title = {Resolving Through-Plane Motion in Volumetric Analysis},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/resolving-through-plane-motion-in-volumetric-analysis/rdr_ed62d12bc0e648499f5ae835e647986b/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}