In clinical medicine, we can easily observe the consequences of biology—a heart beating, a knee bending, or a lung expanding—using tools like MRI and ultrasound. However, the underlying causes of these motions, such as the stiffness of a heart valve or the contractility of a muscle, remain hidden. These intrinsic material properties are often the earliest indicators of disease, yet they cannot be measured directly without invasive biopsies or catheters.
To bridge this gap, researchers are developing a framework of inverse modeling. By combining observable motion data with the laws of physics, they can "back-calculate" the hidden mechanical properties of living tissue, effectively performing a "virtual palpation" of the body's internal structures.
The Inverse Problem
Most engineering is "forward" modeling: if you know the stiffness of a beam and the weight of a car, you can predict how much the bridge will sag. In biomechanics, we face the opposite challenge. We see the "sag" (the deformation of the heart or ligament) and must infer the "stiffness."
As Federica Caforio and Matthias Höfler note in On Parameter Identification in Three-Dimensional Elasticity and Discretisation with Physics-Informed Neural Networks, this is a mathematically "ill-posed" problem. Because biological tissues are nonlinear and heterogeneous, many different combinations of internal properties could theoretically result in the same observed motion. Furthermore, clinical data is often sparse and noisy, making it difficult to pinpoint a single, accurate set of parameters.
The Optimization Loop
The classic solution to this problem is the Inverse Finite Element Method (iFEM). This approach treats the estimation as an iterative game of "hot or cold":
- Create a Model: Build a digital twin of the organ (e.g., a heart or knee).
- Make a Guess: Assign initial material properties (like elasticity) to the model.
- Simulate: Run a physics simulation to see how that "guessed" organ would move under stress.
- Compare: Measure the difference between the simulation and the actual clinical imaging.
- Adjust: Tweak the material properties and repeat until the simulation matches reality.
Early work by Lianghao Han, J. Alison Noble, and Michael Burcher in Non-invasive Measurement of Biomechanical Properties of in vivo Soft Tissues demonstrated that this loop could accurately estimate stiffness ratios in heterogeneous materials. By matching simulated deformations to observed internal boundaries, researchers can reconstruct a "stiffness map" of a patient's anatomy.
From Physics Simulations to Neural Networks
While powerful, iFEM is computationally expensive. A single high-fidelity heart simulation can take hours, making it impractical for real-time surgical planning. To solve this, researchers are moving toward two faster alternatives:
Statistical Emulators
Dirk Husmeier and colleagues have pioneered the use of statistical emulators (or surrogate models). In Statistical Emulation of Cardiac Mechanics: An Important Step towards a Clinical Decision Support System, they show how Gaussian Process models can replace slow physics simulations. By training on a library of pre-computed simulations, these emulators can estimate tissue stiffness in seconds rather than weeks, bringing MRI-based diagnosis into the clinical timeframe.
Physics-Informed Neural Networks (PINNs)
A more recent breakthrough involves embedding the laws of physics directly into the architecture of a neural network. In Physics-informed Neural Network Estimation of Material Properties in Soft Tissue Nonlinear Biomechanical Models, Federica Caforio and Francesco Regazzoni describe a framework where the network's loss function includes the partial differential equations (PDEs) governing biomechanics.
Unlike a standard "black-box" AI, a PINN is constrained by the conservation of mass and momentum. This allows it to infer hidden properties—like the location of "scar tissue" in a 3D heart model—even when the input displacement data is sparse or noisy.
Applications Across the Body
The ability to "see" mechanical properties has profound implications for various medical fields:
- Cardiology: Researchers like Philippe Moireau use these models to estimate local muscle contractility, identifying specific areas of heart failure that global metrics miss (Progress Toward using MRI and a Heart Model to Estimate Patient-Specific Indices of Cardiac Function).
- Orthopedics: By observing joint movement during a squat, Laura Bartsoen and Ilse Jonkers attempt to reverse-engineer the stiffness of knee ligaments to improve surgical outcomes (Bayesian parameter estimation of ligament properties based on tibio-femoral kinematics during squatting).
- Neurology: Brain tissue is "ultrasoft" and notoriously difficult to characterize. Jixin Hou and Wei Zhang highlight how inverse finite element modeling is essential for capturing the complex, non-linear geometries of the human brain (Mechanical Characterization of Brain Tissue: Experimental Techniques, Human Testing Considerations, and Perspectives).
By treating the body's own natural movements as a "mechanical test," this research direction turns standard medical imaging into a powerful diagnostic tool for quantifying the invisible physics of life.