Poster R13 - CARTILAGE QUANTITATIVE CONDUCTIVITY MAPPING: TOWARDS AN ELECTROMECHANICAL BIOMARKER OF KNEE JOINT HEALTH

Introduction: Knee osteoarthritis (OA) stands as one of the largest causes of disability, affecting over 50 million Americans [1]. This irreversible degradation of cartilage often leads to painful surgical interventions, imposing large socioeconomic burdens on patients and medical systems. Earlier detection means faster intervention—improving the quality of life of millions. Quantitative Magnetic Resonance Imaging (qMRI) aids in OA detection by measuring tissue material properties including electrical conductivity [2-7]. Variations in conductivity can indicate early OA due to its proportionality with the streaming potential (SP), which arises when charged fluid moves through a charged medium under pressure, as in loaded cartilage. Proteoglycans in the cartilage extracellular matrix create an electronegative environment that drives conductivity and aids in tissue hydration, ultimately contributing to the tissue’s mechanical response. The measurement of conductivity could enable the measurement of a SP, which has been observed via benchtop, arthroscopic and electroarthrography methods [8-11], but never with MRI, where its noninvasiveness and high-resolution increases sensitivity and creates a more sensitive diagnostic biomarker. This study employed Magnetic Resonance Electrical Properties Tomography with two aims: 1) validating a quantitative conductivity mapping method using ground truth conductivity phantoms and forward simulations and 2) obtaining a quantitative conductivity map (QCM) of healthy knee articular cartilage in vivo. This study aims to develop a robust metric for knee health using cartilage conductivity as a biomarker that could enable precise comparisons between healthy and diseased states, and significantly enhance clinical decision-making, improve patient outcomes and pave way for new therapeutic strategies.

Materials and

Methods: Phantoms, Human Subjects, MR Imaging: Conductivity phantoms were prepared with a 1.28 S/m conductivity standard (HI7030L, Hanna Instruments, Woonsocket, RI) and 0.5 S/m saline (0.32 w/v% NaCl) solution. Healthy human subjects (N = 3 Males, age = 28 ± 6.26) were recruited with institutional review board approval. A 3T Siemens Magnetom Prismafit MRI system with a Tx/Rx Knee 15 Flare Coil was used to get phase images from a 3D balanced steady state free precession (bSSFP) sequence (Table 1).
Quantitative Conductivity Mapping, Forward Model: Analysis was conducted on MATLAB (MathWorks Inc., Natick, MA, United States). Regions of interest (ROIs) were manually segmented to outline multiple phantoms and medial contact regions of the femoral condyle and tibial plateau. No phase wraps were detected. Conductivity calculations followed the 3D Helmholtz phase-based (Eq. 1) [3-6] method with an edge-aware Laplacian algorithm estimating the 2nd order spatial derivative by convolving a tri-variate 2nd order polynomial. Phase voxels deviating more than the threshold from the central phase voxel were excluded from the fits (Eq. 2). Voxels with negative conductivities were zeroed. All QCMs were smoothed using 3D Gaussian smoothing, and the transceiver phase assumption was applied. Parameter optimization was conducted on phantoms with pad sizes: 5, 10, 20 voxels, and deviation thresholds: Rthresh = 10, 20, 40%. Additionally, an in silico forward model was made from MRI phase noise measurements (0.3% on average) to further validate the parameter sensitivity with lower and higher levels of noise (0.1% and 0.9%).

Results, Conclusions, and Discussions: The forward model demonstrated how sensitive the QCM parameters were in recreating a ground truth conductivity. The smaller the pad size ( < 10) the more noise-prone the results, and too high of a pad size (>20) also increased the percent error (>5%) although with precision under 0.2 S/m. Excessive Gaussian 3D smoothing was not favorable after 5 rounds and the optimal parameters entailed a variance of 0.5-1 with a kernel size of 5-11 pixels for all noise levels (Fig. 1). The optimization of the post-processing was found to be sensitive to the pad size and the deviation threshold of the 3D edge-aware Laplacian. This sensitivity is primarily due to noise amplification by the Laplacian operator. Additionally, larger ROIs increased computational expense, limiting rapid turnaround capability. The optimal parameters were determined to be a pad size of 10, a deviation threshold of 20%, and three rounds of 3D Gaussian smoothing.
The expected phantom conductivities of 0.5 S/m and 1.28 S/m for the multi-phantom and 1.28 S/m for the large phantom were successfully validated (Fig. 2C). The phantom conductivities were within the expected range, resulting in errors of 0.7%-20% from the actual conductivities. When applying these parameters in vivo, the medial femoral condyles and tibial plateaus exhibited mean conductivities of 0.68 ± 0.412 S/m and 1.09 ± 0.489 S/m, respectively (Fig. 3B). These results are consistent with literature values for cartilage conductivity, which range from 1.12-2.98 S/m [6], 0.51-1.15 S/m [12], and 0.48 S/m at 128 MHz [7]. The observed variation in bulk conductivities may indicate physiological phenomena related to cartilage health. This work provides a proof-of-concept and validation of conductivity mapping in computational models, ground truth phantoms and in vivo articular cartilage. We hypothesize based on previous electromechanical studies on cartilage that QCMs serve as potential biomarkers for early osteoarthritis detection; however, further experimentation is required to understand how these variations correlate with disease states and the phenomena of articular cartilage under load.

Acknowledgements (Optional):