For decades, the standard metric for assessing heart health has been Ejection Fraction (EF)—the percentage of blood pumped out of the left ventricle with each beat. While intuitive, EF is a blunt, global volumetric measurement. It treats the heart as a simple squeezing balloon, completely overlooking the intricate, multi-directional mechanics of the cardiac muscle (myocardium). A patient can have a perfectly "normal" EF while suffering from severe, localized heart muscle damage, because healthy regions of the heart will temporarily overcompensate for injured ones.
To catch cardiac disease before irreversible damage occurs, researchers have shifted from measuring volume to tracking deformation. This paradigm is known as myocardial strain analysis. By tracking the physical displacement of heart tissue throughout the cardiac cycle, scientists can noninvasively quantify regional muscle function, mapping out exactly where and how the heart is failing to contract.
The Geometry of Strain: Decomposing 3D Motion
To understand why tissue tracking works, we must first look at how the heart moves. The left ventricle does not simply compress; it shortens from top to bottom, constricts radially, and twists like a wrung-out towel.
To capture this complex architecture, tissue tracking algorithms decompose myocardial motion into three distinct, orthogonal vectors:
- Longitudinal Strain: Measures the shortening and lengthening of muscle fibers from the base (top) to the apex (bottom) of the heart.
- Circumferential Strain: Measures the narrowing of the heart's circular perimeter during contraction.
- Radial Strain: Measures the thickening of the ventricular wall as muscle tissue compresses inward.
As Augustine et al. (2013) and Ye et al. (2023) demonstrate, analyzing these distinct vectors allows clinicians to pinpoint subclinical tissue dysfunction. For instance, because the innermost layer of the heart wall (the subendocardium) is primarily composed of longitudinal fibers, longitudinal strain is often the first metric to drop during early-stage ischemia or fibrosis—even when global EF remains unchanged (Wang et al., 2022).
▲ Base
│
│ ◄───► Circumferential (Perimeter shortening)
│ ▲
│ │ Radial (Wall thickening)
▼ ▼
Apex (Longitudinal shortening)
Turning Noise into Signal: Tracking Mechanisms
How do we track a moving piece of muscle inside a living body without invasive sensors? The field relies on two primary imaging modalities: Ultrasound (Echocardiography) and Cardiac Magnetic Resonance Imaging (CMR). Each uses a unique computational strategy to follow tissue over time.
Speckle Tracking Echocardiography (STE)
When ultrasound waves interact with myocardial tissue, they scatter off microscopic structures that are smaller than the ultrasound wavelength. This scattering creates a unique, random interference pattern of bright and dark spots known as "speckle."
While early engineers treated speckle as unwanted noise, modern tracking algorithms treat it as a natural fingerprint. As Kele Xu et al. (2014) note, these acoustic patterns move synchronously with the underlying tissue. By applying block-matching algorithms across sequential frames, Speckle Tracking Echocardiography (STE) follows these acoustic markers to calculate localized tissue deformation (Le et al., 2021; Chen et al., 2022).
Magnetic Resonance Feature Tracking and Tagging
In CMR imaging, tracking can be achieved either physically or algorithmically:
- Tissue Tagging: Specialized pulse sequences magnetically "bleach" a physical grid pattern onto the myocardium at the start of the heartbeat (Abd‐Elmoniem et al., 2012; Zheng et al., 2023). As the muscle contracts, the grid deforms, providing an explicit, high-resolution map of 3D tissue displacement.
- Feature Tracking (FT): Instead of using specialized pulse sequences, feature tracking applies computer vision algorithms directly to standard, widely available cine MRI videos (Augustine et al., 2013). It identifies and tracks anatomical boundaries—specifically the sharp contrast interfaces at the endocardial (inner) and epicardial (outer) borders—across the temporal cycle (Ye et al., 2023; Peters et al., 2016).
The Mathematics of Motion Estimation
The core algorithmic challenge is mapping a point $\mathbf{x}$ in frame $t$ to its corresponding point $\mathbf{x} + \mathbf{u}$ in frame $t+1$. Simple frame-to-frame tracking is highly susceptible to "drift" caused by image noise, respiratory motion, and signal dropouts. To solve this, researchers treat tracking as a global optimization problem.
Early foundational work by Ledesma‐Carbayo et al. (2001) introduced spline-based non-rigid registration. Instead of tracking pixels independently, the motion field is modeled using continuous spatio-temporal B-splines. The tracking is formulated as an optimization problem that minimizes an intensity difference objective function while enforcing spatial smoothness constraints.
To handle low-quality images or regional shadowing, Shanna Liu et al. (2023) proposed combining multiresolution optical flow with dynamic region-of-interest (ROI) tracking, allowing the algorithm to maintain its grip on the ventricular wall even when local image features fade.
Furthermore, because the heart's motion is inherently cyclical, tracking errors can be dramatically reduced by enforcing periodic constraints. Lu et al. (2018) developed a "Flow Network Tracking" framework that models point matching over a graph spanning the entire cardiac cycle. By enforcing a strict one-to-one correspondence and a cyclic closure constraint (ensuring the tracking path returns exactly to its origin at the end of the heartbeat), they effectively eliminated the frame-to-frame drift that plagues open-loop tracking systems.
Frame 0 ───► Frame 1 ───► Frame 2 ───► Frame 3
▲ │
└───────────────────────────────────────┘
Cyclic Closure Constraint
Clinical Utility: What Strain Reveals
By transforming raw video frames into quantitative deformation vectors, myocardial strain analysis exposes cardiac pathology long before it alters global pumping volumes.
Detecting Subclinical Damage and Fibrosis
When heart tissue undergoes remodeling or develops microscopic scarring (fibrosis), it loses its elasticity. Automated strain software can detect this localized stiffening immediately. For example, Cao et al. (2025) demonstrated that combining wall thickness measurements with global longitudinal strain (GLS) creates a highly sensitive diagnostic signal for cardiac amyloidosis. Similarly, Yim (2023) used 2D speckle-tracking to uncover hidden, subclinical tissue deformation in post-COVID-19 patients who otherwise presented with entirely normal standard echocardiograms.
Quantifying Mechanical Dyssynchrony
In a healthy heart, all segments of the ventricle contract in a tightly coordinated symphony. When electrical conduction is disrupted (such as in left bundle branch block), different regions contract at different times, severely reducing pumping efficiency.
By tracking the exact timing of peak strain across various segments of the heart wall, researchers can directly quantify mechanical dyssynchrony (Abd‐Elmoniem et al., 2012; Xie et al., 2021). This precise temporal tracking is critical for Cardiac Resynchronization Therapy (CRT). It allows clinicians to determine exactly where to place pacemaker leads to restore harmony to the heart's motion, and to predict which patients will respond favorably to the therapy (Liu et al., 2021; this study).
Key Foundational Work
Securing Network Input via a Trusted Input Proxy (2007) — contributed 1 concept elements to this direction.
Cardiac Motion Analysis from Ultrasound Sequences Using Non-rigid Registration (2001) — contributed 1 concept elements to this direction.