Osteoarthritis (OA) is a degenerative disease characterized by cartilage thinning and compositional changes¹. Initial signs of cartilage degeneration are molecular and biochemical changes within the extracellular matrix. MR quantitative T1ρ mapping has been extensively used to probe early biochemical changes². While several studies are still limited to analyzing average T1ρ values within specific compartments, there is growing interest in the analysis of spatial distribution and local patterns of T1ρ maps. A novel fully-automatic and unbiased algorithm for studying knee relaxation times by creating an atlas knee and using Voxel-Based Relaxometry (VBR) was recently proposed³. This technique could potentially allow investigation of local cartilage composition differences between two cohorts through voxel-based Statistical Parametric Mapping (SPM)³. Moreover, it permits each patient to be considered as a data-point in a multi-dimensional cloud. Machine learning techniques, such as Principal Component Analysis (PCA) can be adopted to extract latent pattern in the data. In this study we propose to couple VBR and PCA to compare biochemical patterns changes in knees with osteoarthritis and ACL injury.

180 subjects from two different cohorts were considered in this study. OA cohort includes 93 osteoarthritic patients (age=54.88±9.1 years, BMI=25.04±3.51 kg/m2, 62 female, KL=1.96) and 25 matched controls (age=51.2±7.72 years, BMI=24.12±3.76 kg/m2, 18 female, KL=0). ACL cohort includes 52 patients with unilateral ACL tears imaged 6 months after surgical reconstruction (age=28.05±12.1 years, BMI=24.31±2.8 kg/m2, 21 female) and 10 controls (age= 32.01±4 years, BMI=22.8±3.12 kg/m2, 5 female). All imaging was done using a 3T MRI scanner (GE Healthcare, Milwaukee, WI, USA) with an 8-channel phased array knee coil (Invivo Inc, Orlando, FL, USA). Sagittal 3D T1ρ imaging sequences were obtained with the following parameters: TR/TE=9ms/min full, FOV=14 cm, matrix=256x128, slice thickness=4 mm, Views Per Segment=64, time of recovery=1.2 s, spin-lock frequency=500 Hz, ARC phase AF=2, time of spin lock (TSL)=0/10/40/80ms for the ACL cohort and 0/2/4/8/12/20/40/80 ms for the OA cohort. All images were morphed to the space of a reference using a previously proposed technique³. All the morphed T1ρ maps were converted into voxel-by-voxel Z-score maps considering the matched age control groups (Figure 1). The values of Z-score maps were then decomposed using PCA, obtaining twenty orthogonal bases (Principal Components, PCs) that describe patterns of Z-score spatial distribution. Each patient’s Z-score map was then projected on those bases, obtaining similarity scores with each PC. Differences between OA, ACL and control groups in the PC scores were analyzed using unpaired t-tests. Significance level was considered as α<0.05.

Figure 1 shows an example of Z-score conversion from one subject of the OA cohort. Four PCs showed significant differences between OA and controls (Table 1). OA and ACL subjects showed a similar pattern in PC1 that is related to the global average Z-score values. As expected, OA and ACL subjects demonstrated higher Z-scores then controls (Figure 2). OA subjects also showed significantly higher PC2 values, while ACL subjects showed significantly lower PC2 values when compared to controls. The modeling of PC2 proves that higher PC’s values (OA) are related to lower values in the superficial layer and higher values in the deep layer when compared to the average subject, with decreasing difference between the two layers. The relationship for the ACL subjects is inverted (Figure 3).

In this study, OA and ACL subjects are analyzed using PCA-VBR technique. The usage of voxel-based Z-score mapping was shown to normalize T1ρ maps at the voxel level, respective to the spatial distribution of control subjects. The PCA analysis reveals that subjects in both cohorts show higher values of T1ρ Z-score. Interestingly, PC2 shows how the cartilage layer effect is more emphasized in ACL patients, consistent with a previous study that reported the superficial layer as more involved in the early stages of the degeneration⁴. In OA subjects, the integrity of the collagen matrix and permeability of fluid of the layers is compromised, and thus the two layers may show more similar values. The results of this study suggest that PCA-VBR is a technique that can be used to analyze biochemical patterns and could, potentially, allow for a more accurate analysis, disease phenotyping and early stratification of patients after injury. Further analyses are necessary to better understand the meaning of the remaining features and other possible applications.