OBJECTIVE IMAGE-BASED MULTIVARIABLE OA STAGE BIOMARKER: DEVELOPMENT AND CHARACTERIZATION

DATA FROM THE OAIJosé Tamez-Peña*, Edward Schreyer^, Saara Totterman^, Patricia González#

and Víctor Treviño*, *Instituto Tecnológico y de Estudios Superiores de Monterrey, Monterrey, Nuevo León, México, #IMITEK, Monterrey, Nuevo León, México

^4QImaging, Rochester NY

INTRODUCTION: Accurately determining disease state in OA is a very important task in understanding the disease and in the evaluation of OA treatments. Current methodologies for assessing disease state are based on human expertise in the case of KL and WORMS scores or are based on patient-reported evaluation of OA related symptoms, i.e. WOMAC. Therefore there is a need for a human-independent method for determining OA state based either on molecular biomarkers or imaging biomarkers. Molecular biomarkers have not been very successful in OA due to the lack of clear association to OA disease state or to an anatomic location. On the other hand, imaging biomarkers have been hindered by the lack of an efficient and objective method for evaluating images. To tackle the imaging-biomarker issues we developed an automated system that objectively evaluates human knee MR images; therefore it has the potential to provide useful image-based biomarkers. This work is aimed in the development and the characterization of image-based multivariate biomarkers for assessing OA disease state, based on the a set of measurements provided by the automated image-analysis system.

OBJECTIVE: Objectively assessing OA state would help in the development and evaluation of effective treatments for OA. The purpose of this work was to develop and evaluate a multivariate quantitative image-based biomarker as a surrogate measurement of OA disease state using the publicly available OAI data sets.

METHODS: OAI Progression Cohort image data releases 0.D.1, 1.D.1, 2.D.1 and 3.C.1 were used in this study. OAI Biomarker0x, Joint0x, physicalexam0x datasets were also used. All the right knee MRI 3D WE DESS images of these data sets were automatically segmented and quantified five times by 4Qimaging Knee Atlas using CiPAS (Jose Tamez-Pena). The automated atlas-based segmentation extracted the bones as well as the cartilage present in the knee joint. The extracted tissue included definitions of the trochlea, and the medial and lateral weight bearing areas of the knee. Once the tissues were automatically segmented the five separate measurements of volume, thickness, bone curvature, average boundary signal and tibio-femoral inter-bone distances were computed for all regions and for all time points. Figure 1 shows sample 3D reconstructions of the measurements. All five repeated measurements were then exported to Excel (Microsoft) with the RExcel add-in (Erich Neuwirth). Excel was then used to identify all the subjects with a full set of four time points and full successful measurements. Each one of the five repeated measurement were then summarized with the trimmed average (Average of the three middle measurements). The trimmed average measurements were then used to compute the standardized femur curvature, the standardized minimum bone-cartilage contrast, and the standardized minimum of medial and lateral tibia-to-femur bone distances. Those measurements were used as surrogate measurements for bone shape, changes in bone and cartilage composition and unloaded tibia-femoral joint distance. Those measurements then were used to compute a tree variable linear composite metric associated to the OA grade (biomarker00 data set) and to the total WOMAC score of the right knee (Joint0x data set). The weights of the linear combination of three variables were selected in such a way to provide a strong association to the total WOMAC score and the OA Grade score.

The qMRI composite index then was evaluated for responsiveness to time using a linear model adjusted with a generalized least square method that included a general correlation structure and the baseline measurement as a covariate. RESULTS: From 196 subjects included in data release 3.C.1, only 179 subjects had the complete set of four time points with full MRI quantification. The computed qMRI composite index was associated to total WOMAC (r=0.33, t=4.6, p<0.001) and to the ROA grade scores (Spearman rho=0.51, p<0.001).

The response to time coefficients of the linear model are shown in a table above. The linear model of the responsiveness of the qMRI composite index reduced the longitudinal variability of the index in 13%. (r2=0.13, p<0.001). The index had an average annualized SRM of 0.14 for the 179 subjects. The lower portion of the Index (MRI Index< 1.84) showed an annual time effect of 0.75, while the upper portion of the Index showed an annual time effect of -0.67. The bottom plot of Figure 2 show the evolution of the MRI index across time and across discrete values of the index.

CONCLUSIONS: The image-analysis method and biomarker discovery approach presented in this work was able to provide a fully automated index for assessing OA state; that index is associated to the WOMAC scores and to the OA grade scores. This index has the potential to be used for automated and objective screening for OA progression. The method has still to be validated using more time points. It effectiveness and usefulness has still to be tested in different OA study cohorts

Acknowledgment: Funded in part by NIAMS (contracts N01-AR-2-2261, N01-AR-2-2262 and N01-AR-2-2258).

Figure 1. Image-Based Biomarkers used for the computation of the MRI Index. Bone cartilage contrast, bone surface curvature, and tibio-femoral inter-bone distance were used for the computation of the objective MRI based multivariate biomarker.

0.0 mm 10.0 mm

Figure 2. Box-plots of the MRI index. Top left, box-plot of the index vs the ROA grade. Top-right, MRI index vs the WOMAC scores. Bottom, evolution of the index over time. The index starts to get bigger at early OA stages and at later stages it starts to decrease with time

Femur curvature map Femur bone-cartilage contrast map

Femur-Tibia Inter-Bone Distance map

<-0 >+0

MedialLateral