automated mri based 3d joint space analysis

1
José G. Tamez-Peña, Saara Totterman and Patricia González 3D METHOD FOR THE AUTOMATED ANALYSIS OF THE KNEE JOINT SPACE: MRI DATA FROM THE OSTEOARTHRITIS INITIATIVE Rochester NY Introduction: •The quantitative image analysis of high- resolution structural 3D MRI data from the OAI public data sets is challenging due the large number of slices to be analyzed. Robust and automated methods are required to enable the effective evaluation and quantification of the OAI data sets. Scope: •To evaluate a fully automated method for the segmentation of the femur, tibia and patella of the human knee joint and evaluate the three dimensional (3D) joint space width as a candidate for the quantitative evaluation of OA related changes. Materials and Methods: •Publicly available registration algorithms from the ITK (http://www.itk.org) were build in a windows vista system and tested for an atlas based segmentation of the right knee of 82 “progression cohort” subjects from the OAI data public use data sets release 0.C.1, 1.C.1 and 2.D.1. •First, the femur, tibia and patella, were manually segmented from sagittal 3D MRI DESS images acquired using Siemens 3T scanner at high sagittal resolution (0.3646 X 0.456 and 0.7mm slice thickness) of a normally looking right knee. •Then the weight bearing areas, medial tibia-femoral, lateral tibia-femoral and patella-femoral, were defined. The combination of an affine registration algorithm and a spline deformable registration algorithm were used to independently register the femur, the tibia and the patella. •The independent registration parameters of the femur, tibia and patella were used to map automatically the segmentation of the normal subject knee into the sagittal MRI DESS series of the three time points (Baseline, 12 months and 18 months). After the mapping, each one of the segmentations was relaxed to maximize the probability of Statistical Analysis: •The statistical analysis included the correlation of the baseline 3D medial joint space width (3D MJSW) to the total WOMAC scores of the right knee and the correlation of the change in 3D joint space to the baseline total WOMAC right knee scores. Results: •Figure 1 shows the Atlas used to segment the 82 subjects. In a period of two weeks 246 time points were analyzed. From the 246 analyzed time points, the registration algorithm successfully registered 245 femurs, 244 tibias and 222 patellas. Figure 2 shows a typical example of an atlas based segmentation. The success rate of the femur and tibia allowed for the complete longitudinal analysis of the right knee’s medial joint space of 80 subjects. The right knee 3D joint space width of the medial tibia-femoral joint (3DMJSW) showed a week correlation to the baseline WOMAC total right knee scores (r=-0.27, p=0.01). A positive correlation was observed between BL WOMAC total score at the right knee and the 3DMJSW change between baseline (BL) and 12 Months (r=0.30,p<0.01). On the other hand, a stronger negative correlation was observed between the BL WOMAC total right knee score and the 3DMJSW change between 12 months and 18 months (r=-0.38,p<0.01). Figure 3 shows the scatter plots of the association between total WOMAC Scores and the medial joint space width. Conclusion: •This pilot study showed that a fully automated analysis of the bones in 3D MR DESS images is possible; further that the analysis of their segmentation provides valuable and measurable OA related information. The preliminary analysis of the 3D joints space shows that 3D MJSW is correlated to clinical symptoms of OA and that the apparent change in the 3D MJSW is Figure 3:Top, scatter plot of Baseline Total WOMAC and 3D Medial Joint Space Width (3DMJSW). Middle, Observed Annual Change in 3DMJSW and Total WOMAC. Bottom, Observed change in 3DMJSW between the 12 Month observation and the 18 Month Observation vs. Total WOMAC. Figure 1: OAI 3D DESS series was used to create a knee atlas. The bones: Femur, tibia and patella were traced and then regions of interest were defined for quantitative analysis of the joint space. Figure 2: The atlas was tracked to all the subjects. Once the atlas was mapped into each time point. The segmentation was inspected by a train observed.

Upload: josetamepena

Post on 26-Jul-2015

868 views

Category:

Health & Medicine


2 download

TRANSCRIPT

Page 1: Automated MRI based 3D Joint Space Analysis

José G. Tamez-Peña, Saara Totterman and Patricia González

3D METHOD FOR THE AUTOMATED ANALYSIS OF THE KNEE JOINT SPACE: MRI DATA FROM THE OSTEOARTHRITIS INITIATIVE

Rochester NY

Introduction:

• The quantitative image analysis of high-resolution structural 3D MRI data from the OAI public data sets is challenging due the large number of slices to be analyzed. Robust and automated methods are required to enable the effective evaluation and quantification of the OAI data sets.

Scope:• To evaluate a fully automated method for the segmentation of

the femur, tibia and patella of the human knee joint and evaluate the three dimensional (3D) joint space width as a candidate for the quantitative evaluation of OA related changes.

Materials and Methods:

• Publicly available registration algorithms from the ITK (http://www.itk.org) were build in a windows vista system and tested for an atlas based segmentation of the right knee of 82 “progression cohort” subjects from the OAI data public use data sets release 0.C.1, 1.C.1 and 2.D.1.

• First, the femur, tibia and patella, were manually segmented from sagittal 3D MRI DESS images acquired using Siemens 3T scanner at high sagittal resolution (0.3646 X 0.456 and 0.7mm slice thickness) of a normally looking right knee.

• Then the weight bearing areas, medial tibia-femoral, lateral tibia-femoral and patella-femoral, were defined. The combination of an affine registration algorithm and a spline deformable registration algorithm were used to independently register the femur, the tibia and the patella.

• The independent registration parameters of the femur, tibia and patella were used to map automatically the segmentation of the normal subject knee into the sagittal MRI DESS series of the three time points (Baseline, 12 months and 18 months). After the mapping, each one of the segmentations was relaxed to maximize the probability of voxel classification into bone or other soft tissues.

• Once the automated segmentation of the bones was completed, the average distance between bones at joint spaces were computed for each MRI imaging time point and it was statistically analyzed.

Statistical Analysis:

• The statistical analysis included the correlation of the baseline 3D medial joint space width (3D MJSW) to the total WOMAC scores of the right knee and the correlation of the change in 3D joint space to the baseline total WOMAC right knee scores.

Results:

• Figure 1 shows the Atlas used to segment the 82 subjects. In a period of two weeks 246 time points were analyzed. From the 246 analyzed time points, the registration algorithm successfully registered 245 femurs, 244 tibias and 222 patellas. Figure 2 shows a typical example of an atlas based segmentation. The success rate of the femur and tibia allowed for the complete longitudinal analysis of the right knee’s medial joint space of 80 subjects. The right knee 3D joint space width of the medial tibia-femoral joint (3DMJSW) showed a week correlation to the baseline WOMAC total right knee scores (r=-0.27, p=0.01). A positive correlation was observed between BL WOMAC total score at the right knee and the 3DMJSW change between baseline (BL) and 12 Months (r=0.30,p<0.01). On the other hand, a stronger negative correlation was observed between the BL WOMAC total right knee score and the 3DMJSW change between 12 months and 18 months (r=-0.38,p<0.01). Figure 3 shows the scatter plots of the association between total WOMAC Scores and the medial joint space width.

Conclusion:

• This pilot study showed that a fully automated analysis of the bones in 3D MR DESS images is possible; further that the analysis of their segmentation provides valuable and measurable OA related information. The preliminary analysis of the 3D joints space shows that 3D MJSW is correlated to clinical symptoms of OA and that the apparent change in the 3D MJSW is also correlated to the WOMAC total scores. Future work will be done in the analysis of the JSW on the full progression cohort and the exploration of more sensitive image based biomarkers and the validation of the fully automated analysis via the direct comparison of the unsupervised segmentations to segmentations done by expert observers.

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

Figure 3:Top, scatter plot of Baseline Total WOMAC and 3D Medial Joint Space Width (3DMJSW). Middle, Observed Annual Change in 3DMJSW and Total WOMAC. Bottom, Observed change in 3DMJSW between the 12 Month observation and the 18 Month Observation vs. Total WOMAC.

Figure 1: OAI 3D DESS series was used to create a knee atlas. The bones: Femur, tibia and patella were traced and then regions of interest were defined for quantitative analysis of the joint space.

Figure 2: The atlas was tracked to all the subjects. Once the atlas was mapped into each time point.The segmentation was inspected by a train observed.