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.

Uncovering associations between data-driven learned qMRI biomarkers and chronic pain

“…The bone segmentation was also previously extensively evaluated, with a stratified analysis showing no significant differences in segmentation performances at different KL gradings. Additionally, high performance in detecting small, relevant osteophytes was previously shown 24 .…”

“…For full details of this automatic cartilage thickness method, we refer to a previous study 23 . The bone shape was intrinsically described by the distance from the bone surface of each bone mask to its volumetric centroid 24 .…”

Emergence of Deep Learning in Knee Osteoarthritis Diagnosis

“…Osteoarthritis (OA) is one of the most prevalent degenerative musculoskeletal diseases. This disease is affecting almost 5% of the global population [ 1 ]. The knee is the most common joint affected by OA, which is characterized by irreversible degeneration of the articular cartilage at the ends of the bones such as femoral, tibial, and patella cartilages [ 2 ].…”

“…However, this approach causes delay in OA diagnosis because the bony changes only appear in advanced conditions. Besides X-ray, other imaging modalities such as magnetic resonance imaging can utilize several OA soft tissue biomarkers such as cartilage and meniscus degeneration and also deformation of the subchondral and trabecular bone to determine the onset of knee OA [ 1 ]. There exist different types of OA-related segmentation or classification models for assessing the knee which are generally classified into classical methods and deep learning (DL) methods [ 7 , 8 ].…”

Characterization of anisotropic T2W signals from human knee femoral cartilage: The magic angle effect on a spherical surface

“…Two schematic sagittal imaging slices cutting through one femoral condyle's center (red) and off-center (green) in two-dimensional (coronal view) (B) and three-dimensional (unit sphere) diagrams (C) assuming a spherical surface of cartilage. LFC, lateral femoral condyle; MFC, medial femoral condyle As illustrated in Figure 1A, human knee femoral cartilage has a curved surface because of two extruding condyles, that is, lateral (LFC) and medial (MFC) femoral condyles, 21,22 which could be approximately described by a spherical surface. Even although a circle could be defined for the angularly segmented femoral cartilage in a sagittal imaging plane, 20,21,23 the normal vector (m ! )…”