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

Wolfgang Wirth,Ender Konukoglu,Akshay Sanjay Chaudhari,Christian Frederik Baumgartner,Jana Kemnitz,David Fuerst,Felix Eckstein

doi:10.1007/s10334-020-00889-7

Abstract

ObjectiveTo 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.Methods2D 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%).ResultsAutomated 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%).DiscussionThe 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.

Highlights

Osteoarthritis (OA) is a highly prevalent, chronic disease that affects more than 300 million people world-wide [1, 2]
Convolutional neural networks (CNNs), primarily based on the U-Net architecture [13], have been employed for automated cartilage segmentations and have demonstrated a good segmentation agreement between automated vs. ground-truth approaches [14,15,16,17,18,19,20,21,22,23]. Only few of these convolutional neural networks (CNNs)-based studies examined the accuracy of quantitative cartilage measures derived from CNN-based segmentations [14, 16, 23]. None of these reported the longitudinal stability or test–retest precision of quantitative cartilage measures derived from CNN-based cartilage segmentation, which is an important prerequisite before a segmentation methodology can be applied to data from a clinical trial, or compared the segmentation and analysis performance between different magnetic resonance images (MRI) sequences typically used in osteoarthritis studies [24]
During the training of the networks, the best segmentation agreement with data from the validation set was achieved for corFLASH/sagittal double echo at steady-state (sagDESS) lateral femorotibial compartment (LFTC)/sagDESS medial femorotibial compartment (MFTC) after 14/33/34 epochs (99/167/159 min of training), and these U-Net weights were subsequently chosen for the automated segmentations on the hold-out test set

Summary

Introduction

Osteoarthritis (OA) is a highly prevalent, chronic disease that affects more than 300 million people world-wide [1, 2]. While radiography was previously used to assess the structural progression of OA, quantitative measurement of articular cartilage based on serial magnetic resonance images (MRI) is the method of choice and provides the high test–retest precision and sensitivity to longitudinal change required for. Only few of these CNN-based studies examined the accuracy of quantitative cartilage measures (e.g. thickness, volume, and surface area) derived from CNN-based segmentations [14, 16, 23] None of these reported the longitudinal stability or test–retest precision of quantitative cartilage measures derived from CNN-based cartilage segmentation, which is an important prerequisite before a segmentation methodology can be applied to data from a clinical trial, or compared the segmentation and analysis performance between different MRI sequences typically used in osteoarthritis studies [24]

Methods

Results

Conclusion