Accuracy Evaluation of Automatic Quantification of the Articular Cartilage Surface Curvature from MRI

doi:10.1016/j.acra.2007.07.001

Academic Radiology

Volume 14, Issue 10, October 2007, Pages 1221-1228

https://doi.org/10.1016/j.acra.2007.07.001 Get rights and content

Rationale and Objectives

To study the articular cartilage surface curvature determined automatically from magnetic resonance (MR) knee scans, evaluate accuracy of the curvature estimates on digital phantoms, and an evaluation of their potential as disease markers for different stages of osteoarthritis (OA).

Materials and Methods

Knee MR data were acquired using a low-field 0.18T scanner, along with posteroanterior x-rays for evaluation of radiographic signs of OA according to the Kellgren-Lawrence index (KL). Scans from a total of 114 knees from test subjects with KL 0–3, 59% females, ages 21–79 years were evaluated. The surface curvature for the medial tibial compartment was estimated automatically on a range of scales by two different methods: Euclidean shortening flow and boundary normal comparison on a cartilage shape model. The curvature estimates were normalized for joint size for intersubject comparisons. Digital phantoms were created to establish the accuracy of the curvature estimation methods.

Results

A comparison of the two curvature estimation methods to ground truth yielded absolute pairwise differences of 1.1%, and 4.8%, respectively. The interscan reproducibility for the two methods were 2.3% and 6.4% (mean coefficient of variation), respectively. The surface curvature was significantly higher in the OA population (KL > 0) compared with the healthy population (KLi = 0) for both curvature estimates, with P values of .000004 and .000006, respectively. The shape model based curvature estimate could also separate healthy from borderline OA (KL = 1) populations (P = .005).

Conclusion

The phantom study showed that the shape model method was more accurate for a coarse-scale analysis, whereas the shortening flow estimated fine scales better. Both the fine- and the coarse-scale curvature estimates distinguished between healthy and OA populations, and the coarse-scale curvature could even distinguish between healthy and borderline OA populations. The highly significant differences between populations demonstrate the potential of cartilage curvature as a disease marker for OA.

Section snippets

Population and Image Acquisition

The test subjects were between 22 and 79 years of age (average 56 years), with 59% females, and there were both healthy and osteoarthritic subjects according to the KL index. The dataset consisted of MRI scans and x-rays from 114 knees that were used for the evaluation of the method. In addition, 25 knee scans were used for the training of the automatic method. For reproducibility evaluation, 31 knees were rescanned after approximately 1 week, making the total number of knee datasets 170.

Evaluation on Phantoms

The results of estimating the curvature with both the shortening flow and the shape model methods on the same five phantoms with increasing known curvature values can be seen in Fig 5. In the shortening flow method, the mean curvature was calculated using Gaussian derivatives with scale 0.9, because this generated curvature values closest to ground truth. The results show that the shape-model curvature is more accurate when it comes to estimating low curvature values. This is a result of the

Discussion

In this article, we have presented two methods for automatically estimating the curvature on the articular cartilage surface. One method is based on Euclidean shortening flow, in which an object is evolved according to its surface mean curvature, which leads to the shrinking into one or several spheres and eventually disappearance of the object. It is therefore essential that the curvature is measured early in the evolution while the object still has an anatomically meaningful representation.

References (23)

E. Pessis et al.
Assessment of progression in knee osteoarthritis: results of a 1 year study comparing arthroscopy and MRI
Osteoarthr Cartilage
(2003)
C. Peterfy et al.
MRI protocols for whole-organ assessment of the knee in osteoarthritis
Osteoarthr Cartilage
(2006)
S. Gandy et al.
No loss of cartilage volume over three years in patients with knee osteoarthritis as assessed by magnetic resonance imaging
Osteoarthr Cartilage
(2002)
M. Terukina et al.
Analysis of the thickness and curvature of articular cartilage of the femoral condyle
Arthroscopy
(2003)
D. Jackson et al.
Symptomatic articular cartilage degeneration: the impact in the new millennium
Clin Orthop Related Res
(2001)
J. Kellgren et al.
Radiological assessment of osteo-arthrosis
Ann Rheum Dis
(1957)
H. Graichen et al.
Quantitative assessment of cartilage status in osteoarthritis by quantitative magnetic resonance imaging
Arthr Rheum
(2004)
T. Stammberger et al.
Determination of 3D cartilage thickness data from MR imaging: computational method and reproducibility in the living
Magnet Res Med
(1999)
E.B. Dam et al.
Automatic cartilage thickness quantification using a statistical shape model
MICCAI Joint Disease Workshop, Copenhagen, Denmark
(2006)
S. Solloway et al.
The use of active shape models for making thickness measurements of articular cartilage from MR images
Magn Res Med
(1997)

J.A. Lynch et al.

Cartilage segmentation of 3d MRI scans of the osteoarthritic knee combining user knowledge and active contours

SPIE

(2000)

Cited by (19)

Automating the design of resection guides specific to patient anatomy in knee replacement surgery by enhanced 3D curvature and surface modeling of distal femur shape models
2014, Computerized Medical Imaging and Graphics
Citation Excerpt :
Recent approaches in computer assisted orthopedic surgery aimed at overcoming operator dependability focusing on automatic bone shape processing methods. Among different techniques, surface curvature was considered useful for 3D image segmentation [24,25], surface registration [26] and automatic extraction of anatomical features [27–32]. Operatively, surface curvature is an invariant measure for shape description, independent on the position and the orientation that can be exploited for reliable detection of concavities and convexities.
Personalized resection guides (PRG) have been recently proposed in the domain of knee replacement, demonstrating clinical outcome similar or even superior to both manual and navigated interventions. Among the mandatory pre-surgical steps for PRG prototyping, the measurement of clinical landmarks (CL) on the bony surfaces is recognized as a key issue due to lack of standardized methodologies, operator-dependent variability and time expenditure. In this paper, we focus on the reliability and repeatability of an anterior–posterior axis, also known as Whiteside line (WL), of the distal femur proposing automatic surface processing and modeling methods aimed at overcoming some of the major concerns related to the manual identification of such CL on 2D images and 3D models. We show that the measurement of WL, exploiting the principle of mean-shifting surface curvature, is highly repeatable and coherent with clinical knowledge.
Correlations between radiographic assessments and MRI features of knee osteoarthritis - a cross-sectional study
2013, Osteoarthritis and Cartilage
To assess correlations between Kellgren & Lawrence (KL) gradings, minimum joint space width (mJSW) measurements and the Boston Leeds Osteoarthritis Knee Score (BLOKS) within a cohort of obese patients with knee osteoarthritis (KOA).
192 Participants were recruited from an outpatient clinic (ClinicalTrials.gov: NCT00655941). Inclusion criteria were age ≥50 years, body mass index (BMI) ≥30 kg/m² plus symptomatic and verified KOA. 1.5 T magnetic resonance imaging (MRI) scans were assessed using BLOKS and bi-plane radiography by mJSW and KL. Statistics used were Spearman rank correlation coefficients.
The average patient was 63 years of age, female and had a BMI of 37. KL gradings correlated to cartilage damage, bone marrow lesions and meniscus pathology (r = 0.15–0.76) and similar results were found for the relationship between BLOKS and mJSW. BLOKS assessed knee joint pathology co-segregated with compartment and grade specific KL (P < 0.0001). BLOKS variables were statistically significant correlated, particularly in the medial tibiofemoral compartment (r = 0.42–0.80). Adjusting for age, gender and BMI did not alter these associations.
Extensive pathological damage is present even in mild radiographic KOA and BLOKS gradings and KL scores increase together. Analyses of compartment specific KL scores revealed differences in their relationship to the assessed MRI variables. Our study displays the segregation of MRI gradings with respect to location and level of radiographic scores, reveals a high inter-dependency of MRI-assessed structures, and describes some redundancy of specific BLOKS variables.
Responsiveness and reliability of MRI in knee osteoarthritis: A meta-analysis of published evidence
2011, Osteoarthritis and Cartilage
To summarize literature on the responsiveness and reliability of MRI-based measures of knee osteoarthritis (OA) structural change.
A literature search was conducted using articles published up to the time of the search, April 2009. 1338 abstracts obtained with this search were preliminarily screened for relevance and of these, 243 were selected for data extraction. For this analysis we extracted data on reliability and responsiveness for every reported synovial joint tissue as it relates to MRI measurement in knee OA. Reliability was defined by inter- and intra-reader intra-class correlation (ICC), or coefficient of variation, or kappa statistics. Responsiveness was defined as standardized response mean (SRM) - ratio of mean of change over time divided by standard deviation of change. Random-effects models were used to pool data from multiple studies.
The reliability analysis included data from 84 manuscripts. The inter-reader and intra-reader ICC were excellent (range 0.8–0.94) and the inter-reader and intra-reader kappa values for quantitative and semi-quantitative measures were all moderate to excellent (range 0.52–0.88). The lowest value (kappa = 0.52) corresponded to semi-quantitative synovial scoring intra-reader reliability and the highest value (ICC = 0.94) for semi-quantitative cartilage morphology.
The responsiveness analysis included data from 42 manuscripts. The pooled SRM for quantitative measures of cartilage morphometry for the medial tibiofemoral joint was −0.86 (95% confidence intervals (CI) −1.26 to −0.46). The pooled SRM for the semi-quantitative measurement of cartilage morphology for the medial tibiofemoral joint was 0.55 (95% CI 0.47–0.64). For the quantitative analysis, SRMs are negative because the quantitative value, indicating a loss of cartilage, goes down. For the semi-quantitative analysis, SRMs indicating a loss in cartilage are positive (increase in score).
MRI has evolved substantially over the last decade and its strengths include the ability to visualize individual tissue pathologies, which can be measured reliably and with good responsiveness using both quantitative and semi-quantitative techniques.
Systematic review of the concurrent and predictive validity of MRI biomarkers in OA
2011, Osteoarthritis and Cartilage
To summarize literature on the concurrent and predictive validity of MRI-based measures of osteoarthritis (OA) structural change.
An online literature search was conducted of the OVID, EMBASE, CINAHL, PsychInfo and Cochrane databases of articles published up to the time of the search, April 2009. 1338 abstracts obtained with this search were preliminarily screened for relevance by two reviewers. Of these, 243 were selected for data extraction for this analysis on validity as well as separate reviews on discriminate validity and diagnostic performance. Of these 142 manuscripts included data pertinent to concurrent validity and 61 manuscripts for the predictive validity review. For this analysis we extracted data on criterion (concurrent and predictive) validity from both longitudinal and cross-sectional studies for all synovial joint tissues as it relates to MRI measurement in OA.
Concurrent validity of MRI in OA has been examined compared to symptoms, radiography, histology/pathology, arthroscopy, CT, and alignment. The relation of bone marrow lesions, synovitis and effusion to pain was moderate to strong. There was a weak or no relation of cartilage morphology or meniscal tears to pain. The relation of cartilage morphology to radiographic OA and radiographic joint space was inconsistent. There was a higher frequency of meniscal tears, synovitis and other features in persons with radiographic OA. The relation of cartilage to other constructs including histology and arthroscopy was stronger. Predictive validity of MRI in OA has been examined for ability to predict total knee replacement (TKR), change in symptoms, radiographic progression as well as MRI progression. Quantitative cartilage volume change and presence of cartilage defects or bone marrow lesions are potential predictors of TKR.
MRI has inherent strengths and unique advantages in its ability to visualize multiple individual tissue pathologies relating to pain and also predict clinical outcome. The complex disease of OA which involves an array of tissue abnormalities is best imaged using this imaging tool.
Magnetic Resonance Imaging-Based Semiquantitative and Quantitative Assessment in Osteoarthritis
2009, Rheumatic Disease Clinics of North America
Citation Excerpt :
Some reports found that the sensitivity to change or ThCtAB or VCtAB was higher than for VC,100,101 whereas others found comparable standardized response means (SRMs) for VC, VCtAB, or ThCtAB.102,103 Quantitative measures of surface curvature and joint incongruity104–106 have been explored and reported to be associated with cartilage loss (at 0.2 T).107 Cartilage homogeneity was also reported to be significantly different between subjects without and with radiographic OA at 0.2 T,84,108 although other validated MRI techniques of composition cartilage imaging have often been less successful in discriminating healthy volunteers and subjects with OA.70
Quantitative MR Imaging of Cartilage and Trabecular Bone in Osteoarthritis
2009, Radiologic Clinics of North America
Citation Excerpt :
This finding needs to be confirmed, however, in other cohorts and pathophysiologic conditions. Quantitative measures of surface curvature and joint incongruity have been determined36 and were reported to discriminate among subjects with various radiographic OA grades cross sectionally at 0.2 T.37,38 Curvature estimates at different scales (at 0.2 T) were also reported to predict cartilage loss longitudinally.39 Also, at 0.2 T, cartilage homogeneity (quantified by measuring entropy from the distribution of signal intensities in tibial cartilage) was reported to discriminate between subjects without and with early radiographic OA19 and was proposed to be particularly sensitive in peripheral regions, wherein the cartilage is covered by the meniscus.40

View all citing articles on Scopus

View full text

MICCAI Joint Disease WorkshopAccuracy Evaluation of Automatic Quantification of the Articular Cartilage Surface Curvature from MRI

Rationale and Objectives

Materials and Methods

Results

Conclusion

Section snippets

Population and Image Acquisition

Evaluation on Phantoms

Discussion

Osteoarthr Cartilage

Osteoarthr Cartilage

Osteoarthr Cartilage

Arthroscopy

Symptomatic articular cartilage degeneration: the impact in the new millennium

Clin Orthop Related Res

Radiological assessment of osteo-arthrosis

Ann Rheum Dis

Quantitative assessment of cartilage status in osteoarthritis by quantitative magnetic resonance imaging

Arthr Rheum

Determination of 3D cartilage thickness data from MR imaging: computational method and reproducibility in the living

Magnet Res Med

Automatic cartilage thickness quantification using a statistical shape model

MICCAI Joint Disease Workshop, Copenhagen, Denmark

The use of active shape models for making thickness measurements of articular cartilage from MR images

Magn Res Med

Cartilage segmentation of 3d MRI scans of the osteoarthritic knee combining user knowledge and active contours

SPIE

MICCAI Joint Disease Workshop
Accuracy Evaluation of Automatic Quantification of the Articular Cartilage Surface Curvature from MRI