🧬 biology

Impact of a phantomless calibration method on two finite element models predicting metastasised femoral strength

This study demonstrates that a phantomless air-fat-muscle calibration method is a reproducible and promising alternative to phantom-based calibration for predicting metastasised femoral strength using both linear and non-linear finite element models, with the linear Lyon model showing greater robustness to calibration differences.

Original authors: Aurélie Levillain, Marc Gardegaront, Marine Grandin, Thibault Gandet, Elise Jourdain, Mélanie Roche, Guillaume Larid, Jean-Christophe Faivre, Sylvain Grange, Mathilde Proriol, Jean-Baptiste Pialat, Da

Published 2026-07-07

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Aurélie Levillain, Marc Gardegaront, Marine Grandin, Thibault Gandet, Elise Jourdain, Mélanie Roche, Guillaume Larid, Jean-Christophe Faivre, Sylvain Grange, Mathilde Proriol, Jean-Baptiste Pialat, David Mitton, Cyrille Confavreux, Hélène Follet

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine your femur (thigh bone) is a bridge. If cancer spreads to that bone, it creates weak spots, like rust or holes in the bridge's steel. Doctors need to know if this "bridge" will collapse under the weight of a person walking. To predict this, they use a computer simulation called a Finite Element (FE) model. Think of this model as a digital twin of the bone that can be tested to see how much weight it can hold before breaking.

However, to build this digital twin accurately, the computer needs to know exactly how "dense" the bone is. In a CT scan, bone density shows up as a number called a Hounsfield Unit (HU). But these numbers can be tricky; they change depending on the machine used, much like how a thermometer might read slightly different temperatures depending on the brand.

The Problem: The "Calibration Ruler"

Traditionally, to make sure the computer reads the bone density correctly, patients have to lie next to a special plastic block (a phantom) filled with materials of known density during their scan. This block acts like a ruler to calibrate the machine.

The problem? This ruler is expensive, hard to find in every hospital, and a logistical nightmare to set up for every single patient. It's like trying to measure every car's speed by forcing them to drive next to a specific, expensive speedometer on the side of the road.

The Solution: The "Internal Ruler"

The researchers in this paper asked: Can we use the patient's own body as the ruler instead?

They tested a method called Air-Fat-Muscle (AFM) calibration.

The Idea: Every CT scan naturally contains air (in the lungs or around the body), fat, and muscle. The density of these tissues is generally known and consistent.
The Metaphor: Instead of bringing an external ruler to the factory, the factory uses the known size of the workers' hands (air, fat, muscle) to calibrate the measuring tape. The computer looks at the "peaks" of air, fat, and muscle in the scan and says, "Okay, if air is this low and muscle is that high, then the bone must be this dense."

The Experiment: Two Different "Bridge Builders"

The team tested this "Internal Ruler" method on two different computer models of the femur, which they named the Lyon Model and the Leuven Model.

The Lyon Model: This builder uses simple, straight-line logic. If the bone is a little denser, the strength goes up a little. It's like a basic calculator.
The Leuven Model: This builder uses complex, non-linear logic. It accounts for how bone behaves under extreme stress, bending and twisting before it breaks. It's like a sophisticated physics engine that simulates real-world chaos.

They took CT scans of 54 patients with bone metastases. For each patient, they ran the simulation twice: once using the traditional "External Ruler" (the phantom) and once using the new "Internal Ruler" (AFM).

The Findings: How Well Did It Work?

1. The "Internal Ruler" is Reliable
The study found that the AFM method (using the body's own tissues) produced results that were very similar to the gold-standard phantom method.

The Analogy: If the Phantom method says the bridge can hold 10,000 pounds, the AFM method usually says 9,900 or 10,100. They are in the same ballpark.
The Numbers: There was a very strong correlation (0.85) between the two methods. This means the "Internal Ruler" is a promising, reproducible way to measure bone strength without the expensive plastic block.

2. The "Simple Builder" (Lyon) vs. The "Complex Builder" (Leuven)
Here is where it gets interesting. The two models reacted differently to the new calibration method.

The Lyon Model (Simple): It was very stable. Even when the density numbers changed slightly between the two methods, the final strength prediction didn't wobble much. The "limits of agreement" (the margin of error) were narrow.
- Analogy: A simple wooden bridge is sturdy. If you measure the wood's thickness with a slightly different ruler, the bridge still feels just as strong.
The Leuven Model (Complex): It was more sensitive. Because it uses complex math to simulate how bone bends and breaks, small errors in the density measurement got amplified. The margin of error was wider.
- Analogy: A complex suspension bridge with many cables. If you measure the tension in one cable slightly wrong, the whole calculation of how much weight the bridge can hold might swing wildly.

The Conclusion

The paper concludes that using a patient's own tissues (Air, Fat, Muscle) to calibrate bone scans is a viable and promising alternative to using expensive plastic phantoms. It works well for predicting bone strength.

However, the study also suggests that if you are using a simpler computer model (like the Lyon model), you don't need to be as perfect with your measurements because the model is less sensitive to small errors. If you use a complex model (like Leuven), you need to be very precise, or the "Internal Ruler" might introduce more noise.

Important Note: The paper strictly tested the accuracy of the measurement method against the gold standard. It did not test whether these predictions actually saved lives or changed patient outcomes in a hospital setting. It simply proved that you can get a good measurement without the plastic block, and that simpler models might be more forgiving of measurement quirks.

Technical Summary: Impact of a Phantomless Calibration Method on Two Finite Element Models Predicting Metastasised Femoral Strength

Problem Statement
Patient-specific finite element (FE) models based on quantitative computed tomography (QCT) are established tools for predicting the fracture risk of femurs with bone metastases. These models require converting Hounsfield Units (HU) from CT scans into bone mineral density (BMD) to derive mechanical parameters. The current gold standard involves scanning a physical calibration phantom alongside the patient. However, this approach presents significant barriers to routine clinical adoption: the phantoms are expensive, and the logistical burden of simultaneous scanning limits scalability. While phantomless calibration methods using internal patient tissues (e.g., air, fat, muscle) have been proposed, their precision and reproducibility across different FE modeling frameworks—specifically those with varying constitutive laws (linear vs. non-linear)—remain insufficiently investigated.

Methodology
This study evaluated a phantomless calibration method based on internal tissues (Air-Fat-Muscle or AFM) against the gold-standard phantom-based calibration. The study utilized a dataset of 88 femurs from 45 patients (50 with metastases, 38 controls) acquired during the MEKANOS multicentric project.

Calibration Approaches:
- Phantom-based: HU values from a K2HPO4 calibration phantom were used to establish a linear regression slope and intercept for converting HU to density.
- AFM (Phantomless): HU peaks corresponding to air, fat, and muscle were extracted from global histograms of the CT scan volume. These peaks were linearly fitted to reference density values (derived from a subset of phantom scans) to generate a calibration equation.
Finite Element Models:
Two distinct FE models were constructed for each femur using both calibration methods, resulting in four simulations per femur:
- Lyon Model: Utilizes a linear constitutive law. It employs tetrahedral elements, a specific orientation (20°), and a failure criterion based on a critical volume of elements reaching ultimate stress.
- Leuven Model: Utilizes a non-linear constitutive law. It employs hexahedral elements, a different orientation (12°), and models non-linear isotropic material behavior with plasticity and softening.
Analysis:
The study compared the apparent density and predicted failure loads between the two calibration methods. Statistical analyses included linear regression, Bland-Altman plots (to assess bias and limits of agreement), and tests for intra- and inter-operator variability for the AFM method.

Key Results

Correlation and Bias: The AFM method demonstrated very good reproducibility and a strong correlation with the phantom-based method.
- Apparent Density: Significant correlation ( $R^2 = 0.78$ , $p < 0.01$ ) with a mean bias of approximately 0.4%.
- Failure Load: Significant correlations were found for both models ( $R^2 = 0.85$ for Lyon, $R^2 = 0.84$ for Leuven, $p < 0.01$ ). The mean bias in failure load was below 1% for both models.
Limits of Agreement: While correlations were strong, the limits of agreement were narrower for the Lyon (linear) model compared to the Leuven (non-linear) model.
- For failure load, the limits of agreement were $[-17.5\%; 18.3\%]$ for the Lyon model versus $[-26.3\%; 27.3\%]$ for the Leuven model.
- This suggests the linear model is less sensitive to variations in density calibration than the non-linear model.
Variability: Intra- and inter-operator variability for the AFM method was low. The relative difference in predicted failure loads was less than 5% for the Lyon model and less than 10% for the Leuven model across all femurs.
Outliers: Discrepancies were observed in specific cases (e.g., one patient showed a -40% difference in apparent density and up to -48% in failure load for the Leuven model), potentially attributed to fatty infiltration in muscles (sarcopenia) or variations in scanner effective energy not accounted for by fixed reference densities.

Significance and Claims
The authors conclude that the AFM phantomless calibration method is a promising, reproducible, and easily implemented alternative to phantom-based calibration for predicting femoral strength in patients with bone metastases.

Model Sensitivity: The study supports the hypothesis that models with linear constitutive laws (like the Lyon model) are less sensitive to density calibration errors than non-linear models. Consequently, a linear model may be more suitable for clinical applications where phantomless calibration is employed, as it yields narrower limits of agreement.
Clinical Utility: The AFM method allows for the processing of CT scans performed without a phantom, potentially broadening the routine use of bone strength numerical simulation in oncology.
Caveats: The authors note that the precision of the AFM method may be compromised by fatty infiltration in muscles (affecting the distinction between fat and muscle peaks) and changes in imaging protocols (e.g., peak voltage). Furthermore, while the linear model showed better agreement between calibration methods, the study emphasizes that the clinical ability of the Lyon model to accurately predict fracture risk compared to the non-linear Leuven model requires further in vivo validation.

The paper does not claim that the AFM method is perfect or that the linear model is definitively superior for all clinical scenarios; rather, it positions the linear model as a potentially more robust candidate for use with phantomless calibration due to its lower sensitivity to density variations.