Accuracy and longitudinal reproducibility of quantitative femorotibial cartilage measures derived from automated U-Net-based segmentation of two different MRI contrasts: data from the osteoarthritis initiative healthy reference cohort

Abstract

Objective To evaluate the agreement, accuracy, and longitudinal reproducibility of quantitative cartilage morphometry from 2D U-Net-based automated segmentations for 3T coronal fast low angle shot (corFLASH) and sagittal double echo at steady-state (sagDESS) MRI. Methods 2D U-Nets were trained using manual, quality-controlled femorotibial cartilage segmentations available for 92 Osteoarthritis Initiative healthy reference cohort participants from both corFLASH and sagDESS (n = 50/21/21 training/validation/test-set). Cartilage morphometry was computed from automated and manual segmentations for knees from the test-set. Agreement and accuracy were evaluated from baseline visits (dice similarity coefficient: DSC, correlation analysis, systematic offset). The longitudinal reproducibility was assessed from year-1 and -2 follow-up visits (root-mean-squared coefficient of variation, RMSCV%). Results Automated segmentations showed high agreement (DSC 0.89–0.92) and high correlations (r ≥ 0.92) with manual ground truth for both corFLASH and sagDESS and only small systematic offsets (≤ 10.1%). The automated measurements showed a similar test–retest reproducibility over 1 year (RMSCV% 1.0–4.5%) as manual measurements (RMSCV% 0.5–2.5%). Discussion The 2D U-Net-based automated segmentation method yielded high agreement compared with manual segmentation and also demonstrated high accuracy and longitudinal test–retest reproducibility for morphometric analysis of articular cartilage derived from it, using both corFLASH and sagDESS.

Imaging in Osteoarthritis

“…Many studies have described DL methods for fully-automated segmentation of articular cartilage using a wide variety of approaches including statistical shape modelling 151 , 3D active appearance models 152 , 2D and 3D CNNs 153,154 , and sophisticated combinations or variants of generative adversarial networks (GANs) 155 . Although progress in OA image segmentation has led to a point where algorithm errors lie within the intra-reader variability range, deployment in clinical practice still requires some form of supervision, which may include visual inspection of segmentation outputs as well as error maps when a ground truth is available 146 .…”

Osteoarthritis year in review 2021: imaging

“…AI approaches were applied most frequently in MRI and second most commonly in radiography. The applications of AI included for example tissue segmentation [49][50][51][52][53][54][55][56] , lesion detection 57,58 , and diagnosis [59][60][61][62][63][64][65][66][67][68] and prediction of OA [69][70][71][72] . Many of the AI approaches were based on deep learning which is a subfield of AI.…”

Emergence of Deep Learning in Knee Osteoarthritis Diagnosis

“…One of the main advantages of CNN is that they are easier to train and have fewer parameters compared to other architectures [ 30 ]. CNN, basically the U-Net architecture, is popularly used in knee OA for automated segmentation of the cartilage, menisci, bone, or total knee joint anatomy [ 31 , 32 ]. Segmentation of the anatomical structures is important in the clinical practice to evaluate the disease progression and morphological changes where the recent breakthrough of this field is segmenting the cartilage from magnetic resonance (MR) images [ 28 , 33 ].…”

“…The DSCs of cartilage compartments obtained from this study are in the range of 0.76–0.87. Wirth et al [ 31 ] also used 2D U-Net for segmenting femorotibial cartilages to test the cartilage morphometry longitudinal test-retest reproducibility and had demonstrated high DSC for both coronal FLASH and sagittal DESS images. For both studies by Si et al [ 47 ] and Wirth et al [ 31 ], only subjects without OA were included.…”