IMPROVED DCE-MRI OF OA SYNOVITIS IN THE PRESENCE OF EFFUSION

Abstract

INTRODUCTION

Synovitis is increasingly important in OA, both for disease understanding and as a therapeutic target. Dynamic contrast-enhanced (DCE) MRI is a powerful tool providing regional pharmacodynamic biomarkers. Investigators commonly map synovitis using compartmental models, such as the Extended Tofts model (ETM) originally developed for neuroscience and oncology (1). ETM assumes the extravascular extracellular space (v_e) is a well-mixed compartment, an assumption commonly violated in the presence of effusion. Use of an unsuitable compartmental model sometimes produces physiologically implausible imaging biomarkers which lack face validity and damage confidence in the interpretation of any changes.

OBJECTIVE

1) to develop a 3-compartment model (3CM) suitable for DCE-MRI in OA in the presence of effusion; 2) to characterize the model by simulation; 3) to compare performance of new 3CM and conventional ETM in an OA study with between- and within-subject comparison.

METHODS

The model (2) (figure 1A), includes v_e in exchange with a well-mixed vascular plasma compartment v_p, and also with a third effusion-like compartment receiving contrast from, but not returning it to, v_e. It has previously been characterized in an RA setting (2). A previously-reported (3) knee OA DCE-MRI study includes 61 datasets from 21 subjects (6 healthy, 11 KL2, 4 KL3) imaged on multiple occasions, all segmented by a musculoskeletal radiologist (JWM). Data were fit voxelwise in VoxelFlow (Bioxydyn) using 3CM, ETM, and a Patlak-type uptake-only model (UOM). Akaike Information Criterion (AIC) was used to determine which model each voxel preferred. Between-subject means±SD and between-scan repeatability coefficients of variation (CoV) were determined for each biomarker, and also for the AIC-imposed parcellations.

RESULTS

In simulations when the generative model was 3CM, ETM performed poorly (except at low k₁), but when the generative model was ETM, 3CM performed almost as well as the generative model across the whole parameter space. In OA subjects (Figure 1B, Table 1), extreme unphysiologic values of v_e (red in Figure 1C) were seen with ETM but not 3CM, while repeatability CoV did not deteriorate for the new 3CM k₁ in comparison to conventional ETM (Table 1). Differences between healthy and OA subjects were preserved.

CONCLUSION

The new 3CM model provides plausible biomarker values and informative maps, avoiding unphysiologic parameter estimates. This offers drug developers greater confidence in interpreting drug-induced pharmacodynamic responses.