Continuous tracking of aortic aneurysm diameter with peripheral pulse waves: a computational framework combining sequential Markov chain Monte Carlo with Kalman filtering

This paper proposes and validates a computational framework that combines sequential Markov chain Monte Carlo with Kalman filtering to continuously track abdominal aortic aneurysm diameter using peripheral pulse waves, demonstrating that aggregating data from wearable sensors can achieve clinically relevant accuracy even when patient-specific hemodynamic parameters are uncertain.

The Gist

The Big Problem: The Silent Ticking Time Bomb

Imagine your body's main highway (the aorta) has a weak spot that is slowly bulging out, like a tire with a slow leak. This is an Aortic Aneurysm. If it gets too big (usually over 55mm), it can burst, which is often fatal.

Right now, doctors check this "tire" by taking you to a hospital every 6 to 24 months for an ultrasound or MRI. It's like checking your car's tire pressure only twice a year. The problem? If the tire starts losing air rapidly in month 7, you won't know until your next appointment, which could be too late.

The New Idea: The "Smart Watch" for Your Arteries

The researchers asked: Can we check this "tire" every single day, right from home, without a hospital visit?

They propose using PPG (Photoplethysmography). You've probably seen this in smartwatches or fitness trackers; it's the little green light on the back that shines into your skin to measure your pulse.

The theory is this: When blood rushes through your body, it creates a wave (a pulse wave). If there is a bulge (aneurysm) in your aorta, it changes how that wave travels, kind of like how a pothole changes the sound of a car driving over it. The researchers wanted to see if they could listen to that "pothole" sound from your wrist or ankle and calculate exactly how big the bulge is.

The Challenge: The "Noisy Room" Problem

Here is the catch: The signal from the aneurysm is tiny. It's like trying to hear a whisper in a rock concert.

• The Noise: Your heart rate changes when you run. Your blood pressure changes when you eat or get stressed. These changes make the pulse wave look different, often masking the tiny signal from the aneurysm.
• The Result: If you take just one measurement (one snapshot), it is impossible to tell the difference between "the aneurysm got bigger" and "you just drank a cup of coffee." The math shows that a single guess is basically a shot in the dark.

The Solution: The "Crowdsourcing" Strategy

So, how do they solve this? They use aggregation.

Imagine you are trying to guess the weight of a very light feather. If you weigh it once on a shaky scale, you get a bad number. But if you weigh it 1,600 times over the course of a day while the wind blows, the wind pushes the scale up and down randomly. If you average all those numbers, the wind cancels itself out, and you get the exact weight of the feather.

The researchers used a similar trick:

1. Wear the device all day: Collect thousands of pulse readings while your heart rate and blood pressure naturally go up and down.
2. Use a Super-Computer Brain: They built a complex mathematical model (a "Digital Twin" of your circulatory system) that knows how blood waves behave.
3. The Detective Work: They used a method called Bayesian Inference (think of it as a detective updating their theory with every new clue). As the computer sees thousands of different pulse patterns, it starts to realize: "Okay, when the heart rate goes up, the wave changes like this. But the aneurysm size stays the same. Therefore, the aneurysm must be X size."

The Two Scenarios Tested

The researchers ran two types of simulations to see if this works:

Scenario A: We Know Your Body (The "Ideal" World)

• Setup: They knew your specific body measurements (artery length, stiffness, etc.) perfectly, except for the aneurysm size.
• Result: It worked beautifully! By averaging thousands of readings, they could track the aneurysm growth with an error of less than 0.3 millimeters. That is incredibly precise.

Scenario B: We Don't Know Your Body (The "Real World")

• Setup: They didn't know your specific body measurements. They only knew the "average" human range. This is much harder because the computer has to guess your body type and the aneurysm size at the same time.
• Result: It was harder, but still very promising. Even without knowing your specific body details, the system could track the aneurysm with an average error of about 0.65 to 1.4 millimeters. This is still good enough to catch dangerous growth before it's too late.

The "Magic" of the Math

To make this fast enough to run on a computer, they used a Neural Surrogate.

• Analogy: Imagine a physics professor who can calculate the exact path of a falling leaf, but it takes them 10 minutes to do the math. Now, imagine you train a super-smart student (the Neural Network) by showing them 40,000 examples of falling leaves. Afterward, the student can guess the path in a split second with 99% accuracy.
• The researchers trained this "student" to predict how pulse waves behave, allowing them to run the complex math millions of times quickly.

The Verdict: What Does This Mean for You?

This paper is a proof of concept. It hasn't been tested on real humans yet (that's the next step), but the math says it should work.

The Takeaway:
Instead of waiting 6 months to see if your "tire" is getting worse, a smartwatch could potentially monitor it every day. If the aneurysm suddenly starts growing faster (accelerating), the watch could alert your doctor immediately, allowing for surgery before a rupture happens.

It turns a "snapshot" check-up into a "live video feed" of your vascular health, using the power of math to filter out the noise and find the signal.

Technical Summary

1. Problem Statement

Abdominal Aortic Aneurysms (AAAs) pose a significant mortality risk, with rupture often occurring before symptoms appear. Current clinical management relies on intermittent imaging (Ultrasound, CT, MRI) at 6–24 month intervals. This approach has critical limitations:

• Missed Acceleration: Rapid growth acceleration between visits may go undetected until the aneurysm reaches the surgical threshold (55 mm).
• Measurement Gaps: Infrequent sampling fails to capture the continuous dynamics of aneurysm progression.
• Feasibility Gap: Continuous monitoring via wearable devices is desirable but technically challenging because single peripheral pulse wave measurements (e.g., from Photoplethysmography or PPG) are heavily confounded by physiological variables (heart rate, blood pressure) and noise, making single-observation diameter estimation ill-posed.

The paper investigates whether continuous PPG monitoring combined with Bayesian sequential estimation can theoretically track AAA diameter with sufficient accuracy to complement traditional imaging.

2. Methodology

The authors propose a physics-informed computational framework that integrates hemodynamic modeling with advanced statistical inference.

A. Hemodynamic Modeling

• Model: A simplified 1D compliant tube model extending from the aortic root to the pedal digital artery.
• Governing Equations: Solved using the nonlinear 1D Navier-Stokes equations (mass and momentum conservation) via a Lax-Friedrichs finite difference scheme.
• Constitutive Relations: Uses a Kelvin–Voigt tube law for wall mechanics and a 3-element Windkessel model for terminal microcirculation.
• Aneurysm Representation: Modeled as a Gaussian dilation envelope at the abdominal region ($x=0.35$ m) with a 40% local reduction in wall stiffness ($\beta$) to simulate increased compliance.
• Neural Surrogate: To make inverse problems computationally tractable, a Multi-Layer Perceptron (MLP) neural network was trained to approximate the 1D solver. It takes 11 inputs (diameter, HR, MAP, and 8 systemic circulation parameters) and outputs 7 PPG features, achieving a 1000x speedup over the numerical solver.

B. Feature Extraction

Seven hemodynamic features are extracted from the simulated waveforms:

1. Pulse Transit Time (PTT)
2. Effective Pulse Wave Velocity (PWV)
3. Augmentation Index (AI)
4. Reflection Index (RI)
5. Stiffness Index (SI)
6. Systolic Time Ratio (STR)
7. Dicrotic Notch Timing (DNT)
   Note: For inverse problems, only AI, RI, STR, and DNT (features derivable from PPG alone) are used.

C. Inference Frameworks

The study evaluates two scenarios:

1. Scenario 1: Known Systemic Parameters ($\theta_{circ}$)

   • Assumes patient-specific systemic parameters (vessel length, resistance, compliance, stiffness) are known within narrow physiological bounds.
   • Method: A Bayesian Kalman-style filter. It fuses a biomechanical growth model prior with PPG-based observations.
   • Process: Aggregates thousands of noisy observations to invert the model for diameter ($D$), updating the growth rate estimate adaptively.

2. Scenario 2: Unknown Systemic Parameters ($\theta_{circ}$)

   • Assumes systemic parameters are only known to lie within broad population-level bounds (no patient-specific calibration).
   • Method: Sequential MCMC with Kalman Fusion.
     • MCMC: An ensemble sampler (emcee) jointly estimates the unknown systemic parameters ($\theta_{circ}$) and the current diameter ($D_t$) by marginalizing over the parameter space.
     • Warm-Starting: The MCMC ensemble is initialized from the posterior of the previous time step to ensure continuity.
     • Kalman Fusion: The marginal posterior mean of $D$ from MCMC is fed into a Kalman filter step to fuse with the growth model prior, handling the temporal evolution of the aneurysm.

3. Key Contributions

1. Quantification of Identifiability Limits: Demonstrated that single-observation diameter estimation is fundamentally ill-conditioned (Condition Number $\kappa \approx 740$) due to the coupling between diameter, heart rate, and blood pressure.
2. Theoretical Feasibility via Aggregation: Proved via Cramér-Rao Bound (CRB) analysis that aggregating thousands of measurements (e.g., 1,600 observations) under natural physiological variation reduces diameter uncertainty to sub-millimeter levels (< 1 mm).
3. Novel Hybrid Tracking Framework: Developed a sequential MCMC-Kalman framework that successfully tracks aneurysm diameter even when systemic circulation parameters are unknown and only bounded by population statistics.
4. Robustness to Growth Dynamics: Showed the system can detect and adapt to sudden growth acceleration (a critical clinical event) without prior knowledge of the change.
5. Diagnostic Metrics: Identified that MCMC acceptance fractions and Kalman innovation weights can serve as real-time indicators of tracking failure, allowing for corrective clinical actions.

4. Key Results

Sensitivity and Identifiability

• Single Observation: Poor identifiability. A 1 mm change in diameter produces feature changes of only 0.01–0.05 standard deviations relative to noise.
• Multi-Observation: Uncertainty scales roughly with $1/\sqrt{N}$. With 1,600 measurements, diameter uncertainty drops to 0.8 mm (assuming known parameters).

Tracking Performance (12-Month Simulations)

• Scenario 1 (Known $\theta_{circ}$):
  • Achieved an average Root Mean Square Error (RMSE) of ~0.3 mm across 8 virtual patients.
  • Successfully tracked both constant growth and accelerating growth patterns.
• Scenario 2 (Unknown $\theta_{circ}$):
  • Across 50 virtual patients with diverse systemic parameters, the MCMC-Kalman tracker achieved a median RMSE of 0.65 mm and a mean RMSE of 1.4 ± 0.3 mm.
  • While 3 patients experienced catastrophic failure (>6 mm error) due to extreme parameter coupling, the majority maintained clinically relevant accuracy (<1 mm).
  • The system successfully detected growth acceleration events within 1–2 measurement cycles.

Parameter Identifiability

• While individual systemic parameters (like distal resistance) are difficult to identify in isolation, the joint estimation of parameters and diameter allows the system to converge on a functional diameter estimate even if individual parameters remain partially unidentifiable.
• MCMC diagnostics (acceptance fraction < 0.25) correlated strongly with tracking failure, providing a mechanism to flag unreliable estimates.

5. Significance and Conclusion

This work provides a theoretical proof-of-concept that wearable PPG sensors could revolutionize AAA surveillance.

• Clinical Impact: It suggests a pathway to move from intermittent, reactive imaging to continuous, proactive monitoring. This could enable earlier detection of rapid growth acceleration, potentially preventing ruptures.
• Methodological Advance: The combination of physics-based modeling, neural surrogates, and sequential Bayesian inference (MCMC + Kalman) offers a robust solution to high-dimensional, non-linear inverse problems in hemodynamics.
• Future Work: The authors emphasize that while the computational results are promising, clinical translation requires addressing real-world challenges such as motion artifacts, signal quality variability, and prospective validation against imaging data.

In summary, the paper argues that while a single PPG reading cannot measure an aneurysm, continuous aggregation of thousands of readings, processed through a sophisticated Bayesian framework, can achieve the precision necessary for clinical utility.