Valentina Pedoia1
1UC San Francisco, San Francisco, CA, United States
Synopsis
In an effort to develop quantitative biomarkers for degenerative joint disease and fill the void that exists for diagnosing, monitoring and assessing the extent of whole joint degeneration, the past decade has been marked by a greatly increased role of noninvasive imaging. This coupled with recent advances in image processing and deep learning opens new possibilities for promising quantitative techniques including ability in automatically segment multiple musculoskeletal tissues and detect and stage the severity of morphological and biochemical abnormalities. In this lecture, we aim to summarize recent advances in quantitative imaging, image processing and deep learning techniques to study OA.
Introduction
OA is a multifactorial disease that causes joint degeneration, affects 27 million U.S. adults, with symptoms such as stiffness, limited joint function and pain which lead to severe disability and impact the overall quality of life.While MRI can exploit the complexity of the joint with capacity in imaging soft tissues from morphological and biochemical point of views, substantial challenges in image analysis and quantitative image biomarker extraction still hamper clinical translation of promising quantitative techniques widely used in research setting. However, with novel computerized image post processing and more recently machine/deep learning techniques the translation of quantitative MRI is now a tangible goal.In this educational talk we will review recent advances in quantitative imaging and application of image processing and deep learning techniques to study knee and hip osteoarthritis (OA). We aim to give a prospective on how these techniques can help clinical translation of morphological and functional musculoskeletal imaging. Detection of Structural Changes
OA assessment and diagnosis are most commonly done with radiographs (x-rays) using the Kellgren Lawrence (KL) grading system. KL grading is widely used for clinical assessment of OA, usually on a high volume of radiographs, however it is still subject to inter- and intra-user variability, making its automation highly relevant. In the last few years several groups proposed to use deep learning models to automatize OA diagnosis and severity staging from plain radiographs1-2. While x-ray remain the clinical standard, Magnetic Resonance Imaging (MRI), with its’ ability to provide a rich array of structural and functional features of musculoskeletal tissues, has shed light on disease etiology, potential treatment pathways and predictors of long-range outcomes in OA. The processing pipeline of extracting quantitative morphological measurements from MRI often includes a step of image segmentation. An efficient and repeatable fully automatic algorithm for bone and cartilage segmentation could establish a standardized practice for identifying OA imaging biomarkers, as well as the ability to better analyze large datasets in a reasonable amount of time. Musculoskeletal tissue automatic segmentation was an unmet challenge for the last several years. However, recent advances in deep learning have brought new end-to-end models that have showed performances never achieved before 3-5. Additionally, automation of morphological grading of the tissues in the joint would be a significant breakthrough in both OA research and for translation to clinical practice. We will review recent advancement in automatic image grading as well 6-8. Detection of Biochemical Changes
While the analysis of MRI derived morphological OA grades coupled with novel image processing and deep learning techniques have great potential, it is well known that morphological changes are preceded by changes in the cartilage extra-cellular matrix, which are not captured by the metrics above. Quantitative MR, including T1ρ and T2 mapping has been extensively used to probe biochemical changes in the articular cartilage in the early stage. The task of determining quantitative MRI degenerative changes is usually accomplished through region of interest (ROI)-based approaches. In this class of techniques, compartments of the cartilage are segmented and each ROI is described by average encompassed relaxation time values. Sensitivity of relaxation time evaluation trough the extraction of simple averages of cartilage global compartments was widely discussed and previous studies reported that spatially assessing MR images of the knee cartilage relaxation times using laminar and sub-compartmental analyses could lead to better and probably earlier identification of cartilage matrix abnormalities. While the value of quantitative MRI techniques is extensively proven by the body of recent literature, the clinical translation of these advanced imagining techniques is still hampered by the tedious and non-scalable post processing or image segmentation required and the too simplistic feature extraction techniques used for the quantification. Accordingly, the last few years of quantitative MRI research were characterized by a growing interest in exploring fully automatic techniques to assess spatial distribution and local patterns in relaxation time maps. Extraction of second order statistical information or texture analysis has been widely used to overcome the limitation of the average ROI-based approaches with promising results 9-10. However, texture analysis does not address the problem of regional or compartmental differences between the two groups, does not enable the extraction of salient relaxometry patterns and still require segmentation. A fully automatic, local and unbiased algorithm for studying knee T1ρ and T2 relaxation times by creating an atlas and using voxel-based relaxometry (VBR) was proposed 11. This technique allows for the investigation of local cartilage compositional differences between two cohorts, or between different time points in the same cohort using statistical parametric mapping. the usage of VBR coupled with modern deep learning feature extraction techniques will be presented 12.Acknowledgements
No acknowledgement found.References
1. Tiulpin A, Thevenot J, Rahtu E, Lehenkari P, Saarakkala S. Automatic Knee Osteoarthritis Diagnosis from Plain Radiographs: A Deep Learning-Based Approach. Sci Rep. 2018 Jan 29;8(1):1727. doi: 10.1038/s41598-018-20132-7
2. Norman B, Peoia V, Noworolski, A, Link T, Majumdar S, Applying Densely Connected Convolutional Neural Networks for Staging Osteoarthritis Severityfrom Plain Radiographs. J Digit Imaging. 2019 Jun;32(3):471-477. doi: 10.1007/s10278-018-0098-3
3. Liu F, Zhou Z, Jang H, Samsonov A, Zhao G, Kijowski R. Deep convolutional neural network and 3D deformable approach for tissue segmentation in musculoskeletal magnetic resonance imaging. Magnetic resonance in medicine. 2018;79(4):2379-91.
4. Norman B, Pedoia V, Majumdar S. Use of 2D U-Net Convolutional Neural Networks for Automated Cartilage and Meniscus Segmentation of Knee MR Imaging Data to Determine Relaxometry and Morphometry. Radiology. 2018;288(1):177-85
5. Gaj S, Yang M, Nakamura K, Li X. Automated cartilage and meniscus segmentation of knee MRI with conditional generative adversarial networks. Magnetic resonance in medicine. 2019. Epub 2019/12/04. doi: 10.1002/mrm.28111. PubMed PMID: 31793071.
6 Liu FZ, Z.; Samsonov, A.; Blankenbaker, D.; Larison, W.; Kamarek, A.; Lian, K.; Kambhampati, S.; Kijwiski, R. Deep Learning Approach for Evaluating Knee MR Images: Achieving High Diagnostic Performance for Cartilage Lesion Detection. Radiology. 2018 Oct;289(1):160-169
7. Pedoia V, Norman B, Mehany N, Bucknor MD, Link TM, Majumdar S, 3D Convolutional Neural Networks for Detection and Severity Staging of Meniscus and PFJ Cartilage Morphological Degenerative Changes in Osteoarthritis and ACL Subjects. J Magn Reson Imaging. 2019 Feb;49(2):400-410.
8. Bien N, Rajpurkar P, Ball RL, Irvin J, Park A, Jones E, Bereket M, Patel BN, Yeom KW, Shpanskaya K, Halabi S, Zucker E, Fanton G, Amanatullah DF, Beaulieu CF, Riley GM, Stewart RJ, Blankenberg FG, Larson DB, Jones RH, Langlotz CP, Ng AY, Lungren MP, Deep-learning-assisted diagnosis for knee magnetic resonance imaging: Development and retrospective validation of MRNet PLoS Med. 2018 Nov 27;15(11):e1002699.
9 Carballido-Gamio J, Stahl R, Blumenkrantz G, et al. 2009. Spatial analysis of magnetic resonance T1rho and T2 relaxation times improves classification between subjects with and without osteoarthritis. Medical physics 36:4059-4067.
10 Urish K.L., Keffalas M.Durkin.G. J.R,.Miller‖ D.J,.Chu C.R,T.J.Mosher T2 texture index of cartilage can predict early symptomatic OA progression: data from the osteoarthritis initiative Osteoarthritis Cartilage. 2013 Oct;21(10):1550-7
11 Pedoia V, Li X, Su F, Calixto N, Majumdar S.Fully automatic analysis of the knee articular cartilage T1ρ relaxation time using voxel-basedrelaxometry. Magn Reson Imaging. 2016 Apr;43(4):970-80
12 Pedoia V, Lee J, Norman B, Link TM, Majumdar S.Diagnosing osteoarthritis from T2 maps using deep learning: an analysis of the entire Osteoarthritis Initiative baseline cohort..Osteoarthritis Cartilage. 2019 Jul;27(7):1002-1010