Dataset Information

Learning osteoarthritis imaging biomarkers from bone surface spherical encoding.

ABSTRACT: PURPOSE:To learn bone shape features from spherical bone map of knee MRI images using established convolutional neural networks (CNN) and use these features to diagnose and predict osteoarthritis (OA). METHODS:A bone segmentation model was trained on 25 manually annotated 3D MRI volumes to segment the femur, tibia, and patella from 47 078 3D MRI volumes. Each bone segmentation was converted to a 3D point cloud and transformed into spherical coordinates. Different fusion strategies were performed to merge spherical maps obtained by each bone. A total of 41 822 merged spherical maps with corresponding Kellgren-Lawrence grades for radiographic OA were used to train a CNN classifier model to diagnose OA using bone shape learned features. Several OA Diagnosis models were tested and the weights for each trained model were transferred to the OA Incidence models. The OA incidence task consisted of predicting OA from a healthy scan within a range of eight time points, from 1 y to 8 y. The validation performance was compared and the test set performance was reported. RESULTS:The OA Diagnosis model had an area-under-the-curve (AUC) of 0.905 on the test set with a sensitivity and specificity of 0.815 and 0.839. The OA Incidence models had an AUC ranging from 0.841 to 0.646 on the test set for the range from 1 y to 8 y. CONCLUSION:Bone shape was successfully used as a predictive imaging biomarker for OA. This approach is novel in the field of deep learning applications for musculoskeletal imaging and can be expanded to other OA biomarkers.

SUBMITTER: Morales Martinez A

PROVIDER: S-EPMC7329596 | biostudies-literature | 2020 Oct

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Learning osteoarthritis imaging biomarkers from bone surface spherical encoding.

Morales Martinez Alejandro A Caliva Francesco F Flament Io I Liu Felix F Lee Jinhee J Cao Peng P Shah Rutwik R Majumdar Sharmila S Pedoia Valentina V

Magnetic resonance in medicine 20200403 4

<h4>Purpose</h4>To learn bone shape features from spherical bone map of knee MRI images using established convolutional neural networks (CNN) and use these features to diagnose and predict osteoarthritis (OA).<h4>Methods</h4>A bone segmentation model was trained on 25 manually annotated 3D MRI volumes to segment the femur, tibia, and patella from 47 078 3D MRI volumes. Each bone segmentation was converted to a 3D point cloud and transformed into spherical coordinates. Different fusion strategies ...[more]

PMID: 32243657

Similar Datasets

Project description:ObjectivesBone material properties are a major determinant of bone health in older age, both in terms of fracture risk and implant fixation, in orthopaedics and dentistry. Bone is an anisotropic and hierarchical material so its measured material properties depend upon the scale of metric used. The scale used should reflect the clinical problem, whether it is fracture risk, a whole bone problem, or implant stability, at the millimetre-scale. Indentation, an engineering technique involving pressing a hard-tipped material into another material with a known force, may be able to assess bone stiffness at the millimetre-scale (the apparent elastic modulus). We aimed to investigate whether spherical-tip indentation could reliably measure the apparent elastic modulus of human cortical bone.Materials and methodsCortical bone samples were retrieved from the femoral necks of nineteen patients undergoing total hip replacement surgery (10 females, 9 males, mean age: 69 years). The samples underwent indentation using a 1.5 mm diameter, ruby, spherical indenter tip, with sixty indentations per patient sample, across six locations on the bone surfaces, with ten repeated indentations at each of the six locations. The samples then underwent mechanical compression testing. The repeatability of indentation measurements of elastic modulus was assessed using the co-efficient of repeatability and the correlation between the bone elastic modulus measured by indentation and compression testing was analysed by least-squares regression.ResultsIn total, 1140 indentations in total were performed. Indentation was found to be repeatable for indentations performed at the same locations on the bone samples with a mean co-efficient of repeatability of 0.4 GigaPascals (GPa), confidence interval (C.I): 0.33-0.42 GPa. There was variation in the indentation modulus results between different locations on the bone samples (mean co-efficient of repeatability: 3.1 GPa, C.I: 2.2-3.90 GPa). No clear correlation was observed between indentation and compression values of bone elastic modulus (r = 0.33, p = 0.17). The mean apparent elastic modulus obtained by spherical indentation was 9.9 GPa, the standard deviation for each indent cycle was 0.11 GPa, and the standard deviation between locations on the same sample was 1.01 GPa. The mean compression apparent elastic modulus was 4.42 GPa, standard deviation 1.02 GPa.DiscussionSpherical-tip indentation was found to be a repeatable test for measuring the elastic modulus of human cortical bone, demonstrated by a low co-efficient of repeatability in this study. It could not, however, reliably predict cortical bone elastic modulus determined by platens compression testing in this study. This may be due to indentation only probing mechanical properties at the micro-scale while platens compression testing assesses millimetre length-scale properties. Improvements to the testing technique, including the use of a larger diameter spherical indenter tip, may improve the measurement of bone stiffness at the millimetre scale and should be investigated further.

Project description:OBJECTIVES:To assess whether initial or 12-18-month change in magnetic resonance imaging (MRI) subchondral bone texture is predictive of radiographic knee osteoarthritis (OA) progression over 36 months. METHODS:This was a nested case-control study including 122 knees/122 participants in the Osteoarthritis Initiative (OAI) Bone Ancillary Study, who underwent MRI optimised for subchondral bone assessment at either the 30- or 36-month and 48-month OAI visits. Case knees (n = 61) had radiographic OA progression between the 36- and 72-month OAI visits, defined as ? 0.7 mm minimum medial tibiofemoral radiographic joint space (minJSW) loss. Control knees (n = 61) without radiographic OA progression were matched (1:1) to cases for age, sex, body mass index and initial medial minJSW. Texture analysis was performed on the medial femoral and tibial subchondral bone. We assessed the association of texture features with radiographic progression by creating a composite texture score using penalised logistic regression and calculating odds ratios. We evaluated the predictive performance of texture features for predicting radiographic progression using c-statistics. RESULTS:Initial (odds ratio [95% confidence interval] = 2.13 [1.41-3.40]) and 12- 18-month change (3.76 [2.04-7.82]) texture scores were significantly associated with radiographic OA progression. Combinations of texture features were significant predictors of radiographic progression using initial (c-statistic [95% confidence interval] = 0.65 [0.64-0.65], p = 0.003) and 12-18-month change (0.68 [0.68-0.68], p < 0.001) data. CONCLUSIONS:Initial and 12-18-month changes in MRI subchondral bone texture score were significantly associated with radiographic progression at 36 months, with better predictive performance for 12-18-month change in texture. These results suggest that texture analysis may be a useful biomarker of subchondral bone in OA. KEY POINTS:• Subchondral bone MRI texture analysis is a promising knee osteoarthritis imaging biomarker. • In this study, subchondral bone texture was associated with knee osteoarthritis progression. • This demonstrates predictive and concurrent validity of MRI subchondral bone texture analysis. • This method may be useful in clinical trials with interventions targeting bone.

Dataset Information

Learning osteoarthritis imaging biomarkers from bone surface spherical encoding.

Publications

Learning osteoarthritis imaging biomarkers from bone surface spherical encoding.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets