Towards Patient-Specific Computational Modelling of Articular Cartilage on the Basis of Advanced Multiparametric MRI Techniques

Machine Learning Prediction of Collagen Fiber Orientation and Proteoglycan Content From Multiparametric Quantitative MRI in Articular Cartilage

BACKGROUND

Machine learning models trained with multiparametric quantitative MRIs (qMRIs) have the potential to provide valuable information about the structural composition of articular cartilage.

PURPOSE

To study the performance and feasibility of machine learning models combined with qMRIs for noninvasive assessment of collagen fiber orientation and proteoglycan content.

STUDY TYPE

Retrospective, animal model.

ANIMAL MODEL

An open-source single slice MRI dataset obtained from 20 samples of 10 Shetland ponies (seven with surgically induced cartilage lesions followed by treatment and three healthy controls) yielded to 1600 data points, including 10% for test and 90% for train validation.

FIELD STRENGTH/SEQUENCE

A 9.4 T MRI scanner/qMRI sequences: T₁ , T₂ , adiabatic T_1ρ and T_2ρ , continuous-wave T_1ρ and relaxation along a fictitious field (T_RAFF ) maps.

ASSESSMENT

Five machine learning regression models were developed: random forest (RF), support vector regression (SVR), gradient boosting (GB), multilayer perceptron (MLP), and Gaussian process regression (GPR). A nested cross-validation was used for performance evaluation. For reference, proteoglycan content and collagen fiber orientation were determined by quantitative histology from digital densitometry (DD) and polarized light microscopy (PLM), respectively.

STATISTICAL TESTS

Normality was tested using Shapiro-Wilk test, and association between predicted and measured values was evaluated using Spearman's Rho test. A P-value of 0.05 was considered as the limit of statistical significance.

RESULTS

Four out of the five models (RF, GB, MLP, and GPR) yielded high accuracy (R²  = 0.68-0.75 for PLM and 0.62-0.66 for DD), and strong significant correlations between the reference measurements and predicted cartilage matrix properties (Spearman's Rho = 0.72-0.88 for PLM and 0.61-0.83 for DD). GPR algorithm had the highest accuracy (R²  = 0.75 and 0.66) and lowest prediction-error (root mean squared [RMSE] = 1.34 and 2.55) for PLM and DD, respectively.

DATA CONCLUSION

Multiparametric qMRIs in combination with regression models can determine cartilage compositional and structural features, with higher accuracy for collagen fiber orientation than proteoglycan content.

EVIDENCE LEVEL

2 TECHNICAL EFFICACY: Stage 2.

Repeatability and Image Quality of IDEAL-IQ in Human Lumbar Vertebrae for Fat and Iron Quantification across Acquisition Parameters

Echo asymmetry and least square estimation-IQ (IDEAL-IQ) were used to quantify fat and iron to verify the effects of collection parameters on repeatability and image quality of water and fat in human vertebral body. Six IDEAL-IQ sequences were used to scan 48 healthy adult women. Reproducibility of fat and iron quantification and image quality were assessed for six IDEAL-IQ sequences. The results showed that the correlation index (0.987, 0.721) of FF and R2∗ between scans of sequence 2 was higher than that of other sequences, and the consistency of fat quantification was better than that of iron (0.860 vs. 0.579) (P < 0.001). Sequence 2 had the highest image quality score (4.9) and the lowest CV score (9.2%). In the FF figure, SNR (18.8) and CNR (17.8 ± 6.4) were the highest, while CV was the lowest (36.7%, 36.1%). In the R2∗ figure, sequence 3 had the highest SNR (21.8) and CNR (20.5), but its CV (51.8% and 56.1%) was significantly higher than that of sequence 2. The occurrence of fat-water exchange (FWS) was lowest in sequence 2 and sequence 4 (0, N = 96). In conclusion, the fat quantification of IDEAL-IQ was robust to the changes of collection parameters, and section thickness (ST) had a certain effect on maintaining good repeatability of R2∗ quantification. The higher the ST was, the better the image quality of FF and R2∗ was maintained and stable and the less the occurrence of FWS.

Magnetic Resonance Imaging-based biomechanical simulation of cartilage: A systematic review

MRI-based mathematical and computational modeling studies can contribute to a better understanding of the mechanisms governing cartilage's mechanical performance and cartilage disease. In addition, distinct modeling of cartilage is needed to optimize artificial cartilage production. These studies have opened up the prospect of further deepening our understanding of cartilage function. Furthermore, these studies reveal the initiation of an engineering-level approach to how cartilage disease affects material properties and cartilage function. Aimed at researchers in the field of MRI-based cartilage simulation, research articles pertinent to MRI-based cartilage modeling were identified, reviewed, and summarized systematically. Various MRI applications for cartilage modeling are highlighted, and the limitations of different constitutive models used are addressed. In addition, the clinical application of simulations and studied diseases are discussed. The paper's quality, based on the developed questionnaire, was assessed, and out of 79 reviewed papers, 34 papers were determined as high-quality. Due to the lack of the best constitutive models for various clinical conditions, researchers may consider the effect of constitutive material models on the cartilage disease simulation. In the future, research groups may incorporate various aspects of machine learning into constitutive models and MRI data extraction to further refine the study methodology. Moreover, researchers should strive for further reproducibility and rigorous model validation and verification, such as gait analysis.

No pressure, no diamonds? - Static vs. dynamic compressive in-situ loading to evaluate human articular cartilage functionality by functional MRI

Biomechanical Magnetic Resonance Imaging (MRI) of articular cartilage, i.e. its imaging under loading, is a promising diagnostic tool to assess the tissue's functionality in health and disease. This study aimed to assess the response to static and dynamic loading of histologically intact cartilage samples by functional MRI and pressure-controlled in-situ loading. To this end, 47 cartilage samples were obtained from the medial femoral condyles of total knee arthroplasties (from 24 patients), prepared to standard thickness, and placed in a standard knee joint in a pressure-controlled whole knee-joint compressive loading device. Cartilage samples' responses to static (i.e. constant), dynamic (i.e. alternating), and no loading, i.e. free-swelling conditions, were assessed before (δ₀), and after 30 min (δ₁) and 60 min (δ₂) of loading using serial T1ρ maps acquired on a 3.0T clinical MRI scanner (Achieva, Philips). Alongside texture features, relative changes in T1ρ (Δ₁, Δ₂) were determined for the upper and lower sample halves and the entire sample, analyzed using appropriate statistical tests, and referenced to histological (Mankin scoring) and biomechanical reference measures (tangent stiffness). Histological, biomechanical, and T1ρ sample characteristics at δ₀ were relatively homogenous in all samples. In response to loading, relative increases in T1ρ were strong and significant after dynamic loading (Δ₁ = 10.3 ± 17.0%, Δ₂ = 21.6 ± 21.8%, p = 0.002), while relative increases in T1ρ after static loading and in controls were moderate and not significant. Generally, texture features did not demonstrate clear loading-related associations underlying the spatial relationships of T1ρ. When realizing the clinical translation, this in-situ study suggests that serial T1ρ mapping is best combined with dynamic loading to assess cartilage functionality in humans based on advanced MRI techniques.

Complex mechanical behavior of human articular cartilage and hydrogels for cartilage repair

The mechanical behavior of cartilage tissue plays a crucial role in physiological mechanotransduction processes of chondrocytes and pathological changes like osteoarthritis. Therefore, intensive research activities focus on the identification of implant substitute materials that mechanically mimic the cartilage extracellular matrix. This, however, requires a thorough understanding of the complex mechanical behavior of both native cartilage and potential substitute materials to treat cartilage lesions. Here, we perform complex multi-modal mechanical analyses of human articular cartilage and two surrogate materials, commercially available ChondroFiller_liquid, and oxidized alginate-gelatin (ADA-GEL) hydrogels. We show that all materials exhibit nonlinearity and compression-tension asymmetry. However, while hyaline cartilage yields higher stresses in tension than in compression, ChondroFiller^liquid and ADA-GEL exhibit the opposite trend. These characteristics can be attributed to the materials' underlying microstructure: Both cartilage and ChondroFiller^liquid contain fibrillar components, but the latter constitutes a bi-phasic structure, where the 60% nonfibrillar hydrogel proportion dominates the mechanical response. Of all materials, ChondroFiller^liquid shows the most pronounced viscous effects. The present study provides important insights into the microstructure-property relationship of cartilage substitute materials, with vital implications for mechanically-driven material design in cartilage engineering. In addition, we provide a data set to create mechanical simulation models in the future.

Can sodium MRI be used as a method for mapping of cartilage stiffness?

Objective

Sodium concentration is responsible for (at least part of) the stiffness of articular cartilage due to the osmotic pressure it generates. Therefore, we hypothesized that we could use sodium MRI to approximate the stiffness of cartilage to assess early cartilage degeneration.

Methods

Four human tibial plateaus were retrieved from patients undergoing total knee replacement (TKR), and their cartilage stiffness mapped with indentation testing, after which samples were scanned in a 7 T MRI to determine sodium concentration. The relation of biomechanical parameters to MRI sodium and glycosaminoglycan (GAG) concentration was explored by a linear mixed model.

Results

Weak correlations of GAG concentration with apparent peak modulus (p = 0.0057) and apparent equilibrium modulus (p = 0.0181) were observed and lack of correlation of GAG concentration versus MRI sodium concentration was observed. MRI sodium concentration was not correlated with apparent peak modulus, though a moderate correlation of MRI sodium concentration with permeability was shown (p = 0.0014).

Discussion and conclusion

Although there was correlation between GAG concentration and cartilage stiffness, this was not similar with sodium concentration as measured by MRI. Thus, if the correlation between MRI sodium imaging and GAG concentration could be resolved, this strategy for assessing cartilage functional quality still holds promise.

Electronic supplementary material

The online version of this article (10.1007/s10334-020-00893-x) contains supplementary material, which is available to authorized users.