Magnetic resonance imaging (MRI) is safe and versatile tool to acquire radiological data from the knee joint. While clinical imaging sequences are able to provide excellent soft tissue contrast and enable identification of trauma and gross degeneration, they are not sensitive enough for characterization of changes occurring before clinical onset of OA. Quantitative MRI (qMRI) parameters have proven promising in assessing cartilage quality¹ while gray level co-occurrence matrices (GLCM)² can be used to probe the spatial correspondences of the changes caused by OA, and thus, extract underlying information from T2 maps.

Implementations for calculating texture are available either free of charge, such as MaZda³, or in commercial software, such as the Image Processing toolbox of MATLAB (MathWorks, Natick, MA), but none of them is addressing the special requirements for cartilage, namely the curvature of the analyzed surfaces and relatively low resolution in relation to cartilage thickness. Therefore, customized optimization of texture analysis is necessary.

In this abstract, we present an angle dependent extrapolative method for improved GLCM analysis.

64 asymptomatic volunteers (38 female, mean age 54.9 years (standard deviation, SD 13.9), body mass index 24.9 kg/m2 (SD 3.2)) and 80 symptomatic patients (49 female, 59.9 years (SD 7.7), 29.0 kg/m2 (SD 4.29)) were recruited with the permission from the local ethics committee. A total of 80 asymptomatic volunteers were recruited via newspaper advertisement, while 16 had to be excluded due to MRI OA findings. Patients were selected from the hospital registry based on long-term non-specific knee pain or referring to knee replacement surgery. Exclusion criterion included trauma, rheumatoid diseases, previous knee surgery, or other underlying medical conditions that could affect the knee joint. Subjects were scanned on 3T Siemens Skyra (Siemens Healthcare, Erlangen, Germany). T2 relaxation time mapping was performed with a multi-slice multi-echo spin echo sequence (TR=1680ms, TE’s(5)=13.8…69.0ms, ETL=5, FOV=160x160mm, matrix=384x384, thickness=3 mm).

T2 maps were processed with an in-house developed segmentation and analysis tool (MATLAB, Mathworks Inc., Natick, MA). Gray level co-occurrence matrix (GLCM) texture features were calculated in parallel orientation (along the bone-cartilage interface direction) for each region of interest (ROI). Instead of determining the offset along x-y axis or flattening the cartilage⁴, we developed an approach where the offset for each pixel is changing along the curve of the cartilage: The offset is calculated from the perpendicular gradient of the cartilage-bone interface (Fig 1b). Next each pixel within the cartilage is assumed with the angle of the closest value along the gradient line. With the limited resolution of T2 maps, the orientation vectors tend to point to in-between pixels (Fig 1c). To overcome this limitation, we performed extrapolative soft-contour analysis for the offset endpoint (Fig 1d). This would allow realistic endpoint values for freely varying offset vectors.

Compared to preliminary results⁵, the improved algorithm shows higher sensitivity and smaller intraparameter variability (Fig 2). For the studied parameters, the improved algorithm provides smaller p-values than the conventional approach. Homogeneity, difference entropy and information measure of correlation display upper and lower quartiles closer to the median, which suggest that the parameters provide more consistent results between different subjects. With the conventional algorithm, energy did not show statistically significant differences between the subject groups, while with the improved algorithm both femoral regions show significant difference between control and patient group. Furthermore, when studied regions were compared for each parameter, the new algorithm maintains more stable level between the regions: With the conventional approach, the regions, especially the medial central femur, had a different baseline level that made comparison between ROIs impractical. With the improved algorithm, though, the parameter levels are consistent and results are more comparable and systematic.Improved texture analysis algorithm displays excellent performance in discerning symptomatic patients and asymptomatic volunteers. Point-wise offset angle with endpoint extrapolation allowed us to get realistic results from texture analysis even while the resolution of the T2 maps is limited and only eight-pixel neighborhood is originally available for use. The improved algorithm resulted in smaller p-values, smaller variability and more systematic baseline in overall.