1
|
Schmidt AM, Desai AD, Watkins LE, Crowder HA, Black MS, Mazzoli V, Rubin EB, Lu Q, MacKay JW, Boutin RD, Kogan F, Gold GE, Hargreaves BA, Chaudhari AS. Generalizability of Deep Learning Segmentation Algorithms for Automated Assessment of Cartilage Morphology and MRI Relaxometry. J Magn Reson Imaging 2023; 57:1029-1039. [PMID: 35852498 PMCID: PMC9849481 DOI: 10.1002/jmri.28365] [Citation(s) in RCA: 4] [Impact Index Per Article: 4.0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Grants] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 04/04/2022] [Revised: 07/06/2022] [Accepted: 07/07/2022] [Indexed: 01/21/2023] Open
Abstract
BACKGROUND Deep learning (DL)-based automatic segmentation models can expedite manual segmentation yet require resource-intensive fine-tuning before deployment on new datasets. The generalizability of DL methods to new datasets without fine-tuning is not well characterized. PURPOSE Evaluate the generalizability of DL-based models by deploying pretrained models on independent datasets varying by MR scanner, acquisition parameters, and subject population. STUDY TYPE Retrospective based on prospectively acquired data. POPULATION Overall test dataset: 59 subjects (26 females); Study 1: 5 healthy subjects (zero females), Study 2: 8 healthy subjects (eight females), Study 3: 10 subjects with osteoarthritis (eight females), Study 4: 36 subjects with various knee pathology (10 females). FIELD STRENGTH/SEQUENCE A 3-T, quantitative double-echo steady state (qDESS). ASSESSMENT Four annotators manually segmented knee cartilage. Each reader segmented one of four qDESS datasets in the test dataset. Two DL models, one trained on qDESS data and another on Osteoarthritis Initiative (OAI)-DESS data, were assessed. Manual and automatic segmentations were compared by quantifying variations in segmentation accuracy, volume, and T2 relaxation times for superficial and deep cartilage. STATISTICAL TESTS Dice similarity coefficient (DSC) for segmentation accuracy. Lin's concordance correlation coefficient (CCC), Wilcoxon rank-sum tests, root-mean-squared error-coefficient-of-variation to quantify manual vs. automatic T2 and volume variations. Bland-Altman plots for manual vs. automatic T2 agreement. A P value < 0.05 was considered statistically significant. RESULTS DSCs for the qDESS-trained model, 0.79-0.93, were higher than those for the OAI-DESS-trained model, 0.59-0.79. T2 and volume CCCs for the qDESS-trained model, 0.75-0.98 and 0.47-0.95, were higher than respective CCCs for the OAI-DESS-trained model, 0.35-0.90 and 0.13-0.84. Bland-Altman 95% limits of agreement for superficial and deep cartilage T2 were lower for the qDESS-trained model, ±2.4 msec and ±4.0 msec, than the OAI-DESS-trained model, ±4.4 msec and ±5.2 msec. DATA CONCLUSION The qDESS-trained model may generalize well to independent qDESS datasets regardless of MR scanner, acquisition parameters, and subject population. EVIDENCE LEVEL 1 TECHNICAL EFFICACY: Stage 1.
Collapse
Affiliation(s)
- Andrew M Schmidt
- Department of Radiology, Stanford University, Palo Alto, California, USA
| | - Arjun D Desai
- Department of Radiology, Stanford University, Palo Alto, California, USA
- Electrical Engineering, Stanford University, Palo Alto, California, USA
| | - Lauren E Watkins
- Department of Radiology, Stanford University, Palo Alto, California, USA
- Bioengineering, Stanford University, Palo Alto, California, USA
| | - Hollis A Crowder
- Mechanical Engineering, Stanford University, Palo Alto, California, USA
| | - Marianne S Black
- Department of Radiology, Stanford University, Palo Alto, California, USA
- Mechanical Engineering, Stanford University, Palo Alto, California, USA
| | - Valentina Mazzoli
- Department of Radiology, Stanford University, Palo Alto, California, USA
| | - Elka B Rubin
- Department of Radiology, Stanford University, Palo Alto, California, USA
| | - Quin Lu
- Philips Healthcare North America, Gainesville, Florida, USA
| | - James W MacKay
- Department of Radiology, University of Cambridge, Cambridge, UK
- Norwich Medical School, University of East Anglia, Norwich, UK
| | - Robert D Boutin
- Department of Radiology, Stanford University, Palo Alto, California, USA
| | - Feliks Kogan
- Department of Radiology, Stanford University, Palo Alto, California, USA
| | - Garry E Gold
- Department of Radiology, Stanford University, Palo Alto, California, USA
- Bioengineering, Stanford University, Palo Alto, California, USA
| | - Brian A Hargreaves
- Department of Radiology, Stanford University, Palo Alto, California, USA
- Electrical Engineering, Stanford University, Palo Alto, California, USA
- Bioengineering, Stanford University, Palo Alto, California, USA
| | - Akshay S Chaudhari
- Department of Radiology, Stanford University, Palo Alto, California, USA
- Biomedical Data Science, Stanford University, Palo Alto, California, USA
| |
Collapse
|
2
|
Hayashi D, Roemer FW, Link T, Li X, Kogan F, Segal NA, Omoumi P, Guermazi A. Latest advancements in imaging techniques in OA. Ther Adv Musculoskelet Dis 2022; 14:1759720X221146621. [PMID: 36601087 PMCID: PMC9806406 DOI: 10.1177/1759720x221146621] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 08/27/2022] [Accepted: 12/05/2022] [Indexed: 12/28/2022] Open
Abstract
The osteoarthritis (OA) research community has been advocating a shift from radiography-based screening criteria and outcome measures in OA clinical trials to a magnetic resonance imaging (MRI)-based definition of eligibility and endpoint. For conventional morphological MRI, various semiquantitative evaluation tools are available. We have lately witnessed a remarkable technological advance in MRI techniques, including compositional/physiologic imaging and automated quantitative analyses of articular and periarticular structures. More recently, additional technologies were introduced, including positron emission tomography (PET)-MRI, weight-bearing computed tomography (CT), photon-counting spectral CT, shear wave elastography, contrast-enhanced ultrasound, multiscale X-ray phase contrast imaging, and spectroscopic photoacoustic imaging of cartilage. On top of these, we now live in an era in which artificial intelligence is increasingly utilized in medicine. Osteoarthritis imaging is no exception. Successful implementation of artificial intelligence (AI) will hopefully improve the workflow of radiologists, as well as the level of precision and reproducibility in the interpretation of images.
Collapse
Affiliation(s)
- Daichi Hayashi
- Department of Radiology, Renaissance School of Medicine at Stony Brook University, Stony Brook, NY, USA,Department of Radiology, Chobanian & Avedisian School of Medicine, Boston University, Boston, MA, USA
| | - Frank W. Roemer
- Department of Radiology, Chobanian & Avedisian School of Medicine, Boston University, Boston, MA, USA,Department of Radiology, Universitätsklinikum Erlangen, Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), Erlangen, Germany
| | - Thomas Link
- Department of Radiology, University of California San Francisco, San Franciso, CA, USA
| | - Xiaojuan Li
- Department of Radiology, Cleveland Clinic Lerner College of Medicine, Cleveland, OH, USA
| | - Feliks Kogan
- Department of Radiology, Stanford University, Stanford, CA, USA
| | - Neil A. Segal
- Department of Rehabilitation Medicine, The University of Kansas, Kansas City, KS, USA
| | - Patrick Omoumi
- Department of Radiology, Lausanne University Hospital, University of Lausanne, Lausanne, Switzerland
| | | |
Collapse
|