Grading Chondroid Tumors through Magnetic Resonance ImagingMRI Radiomics: Enchondroma, Low-grade Chondrosarcoma, and High-grade Chondrosarcoma
Seminars in Musculoskeletal Radiology · 01.06.2024 · 28 (S 01 S 01) · S1-S24
DOI: 10.1055/s-0044-1787520 ↗Kurz gefasst
Fachartikel in der Zeitschrift Seminars in Musculoskeletal Radiology, erschienen am 01.06.2024. Verfasst von H. Park.
Abstract (Originalsprache)
Purpose or Learning Objective: To develop athe multiclass radiomics model to differentiate the spectrum of enchondroma, low-grade chondrosarcoma (CSA), and intermediate-- to- high- grade CSA. Methods or Background: A total of 119 patients with clinically or pathologically confirmed enchondroma or well-differentiated CSAchondrosarcoma who had performed contrast- enhanced magnetic resonance imaging MRI between 2009 and –2019 were enrolled retrospectively enrolled. Open- source software ITK-SNAP v. (version 3.9.0, (http://www.itksnap.org/) was used for lesion segmentation. Delineation was performed by a musculoskeletal radiologist. Image preprocessing and feature extraction were performed using PyRadiomics v.(Ver 3.0). We used Least Absolute Shrinkage and Selection Operator (LASSO) and random forest (RF) for feature selection and development of the classification model. We applied two2 different schemes for feature selection by LASSO. First, features were selected once from the entire set by LASSO, and the selected features were used consistently used to train and validate each model. Second, features were repeatedly selected by LASSO for each training set using the among five5-fold cross--validation process. Additionally, to overcome the imbalance of samples in the study, the Synthetic Minority Oversampling Technique (SMOTE) method was applied. We developed four4 different models using different combinations of the aforementioned algorithms just described. For the input of features, We used various input features from T1-weightedWI, T2-weightedWI, contrast-enhanced CE-T1-weighted imagesT1WI, and a combination of all sequences. We used averaged accuracy and area under the curve (AUC) with the from five5-fold cross- validation process for each model. Agreements between pathologic results and machine learning were analyzed using the weighted κkappa statistics. Results or Findings: The overall accuracy wasis >more than 0.8, regardless of among the different models., LASSO plus+ RF showed the highest accuracy and AUC for three-3 class classification and the binary classification. A classification model with LASSO plus+ SMOTE plus+ RF, T2-weighted image–WI derived features showed the highest accuracy, and the combination of all sequences showed the highest AUC in our test set. In the second try 2, we used the LASSO feature selection at every time in the training set and validated it in the test set. The combined Lasso, SMOTE, and RF model using T2-weightedWI–- derived features showed the highest accuracy and combination of all features and showed the highest AUC. Conclusion: By integrating radiomic features and machine learning methods, we constructed an accurate machine learning classifier to categorize for classification of cartilaginous bone tumors. Our model showed good diagnostic performance. Publication History Article published online: 22 May 2024 © 2024. Thieme. All rights reserved. Thieme Medical Publishers, Inc. 333 Seventh Avenue, 18th Floor, New York, NY 10001, USA
Zitieren
APA
Park, H. (2024). Grading Chondroid Tumors through Magnetic Resonance ImagingMRI Radiomics: Enchondroma, Low-grade Chondrosarcoma, and High-grade Chondrosarcoma. Seminars in Musculoskeletal Radiology, 28(S 01 S 01), S1-S24. https://doi.org/10.1055/s-0044-1787520
Vancouver
Park H. Grading Chondroid Tumors through Magnetic Resonance ImagingMRI Radiomics: Enchondroma, Low-grade Chondrosarcoma, and High-grade Chondrosarcoma. Seminars in Musculoskeletal Radiology. 2024;28(S 01 S 01):S1-S24. doi:10.1055/s-0044-1787520
BibTeX
@article{Park2024,
author = {Park, H.},
title = {Grading Chondroid Tumors through Magnetic Resonance ImagingMRI Radiomics: Enchondroma, Low-grade Chondrosarcoma, and High-grade Chondrosarcoma},
journal = {Seminars in Musculoskeletal Radiology},
year = {2024},
volume = {28},
number = {S 01 S 01},
pages = {S1--S24},
doi = {10.1055/s-0044-1787520}
}
EUVIMED