Yupeng Zhang1, Baorui Zhang1, Fei Liang1, Shikai Liang2, Yuxiang Zhang1, Peng Yan1, Chao Ma1, Aihua Liu1, Feng Guo3, Chuhan Jiang4. 1. Department of Interventional Neuroradiology, Beijing Neurosurgical Institute and Beijing Tiantan Hospital, Capital Medical University, Room 603, No. 6 Tiantan Xili, Dongcheng District, Beijing, China. 2. Department of Neurosurgery, Beijing Tsinghua Changgung Hospital, School of Clinical Medicine, Tsinghua University, Beijing, China. 3. Department of Neurosurgery, Linyi People's Hospital, No. 49 Yizhou Road, Linyi City, Shandong Province, China. drguofeng@163.com. 4. Department of Interventional Neuroradiology, Beijing Neurosurgical Institute and Beijing Tiantan Hospital, Capital Medical University, Room 603, No. 6 Tiantan Xili, Dongcheng District, Beijing, China. jiangchuhan126@126.com.
Abstract
OBJECTIVE: To investigate the classification ability of quantitative radiomics features extracted on non-contrast-enhanced CT (NECT) image for discrimination of AVM-related hematomas from those caused by other etiologies. METHODS: Two hundred sixty-one cases with intraparenchymal hematomas underwent baseline CT scan between 2012 and 2017 in our center. Cases were split into a training dataset (n = 180) and a test dataset (n = 81). Hematoma types were dichotomized into two classes, namely, AVM-related hematomas (AVM-H) and hematomas caused by other etiologies. A total of 576 radiomics features of 6 feature groups were extracted from NECT. We applied 11 feature selection methods to select informative features from each feature group. Selected radiomics features and the clinical feature age were then used to fit machine learning classifiers. In combination of the 11 feature selection methods and 8 classifiers, we constructed 88 predictive models. Predictive models were evaluated and the optimal one was selected and evaluated. RESULTS: The selected radiomics model was RELF_Ada, which was trained with Adaboost classifier and features selected by Relief method. Cross-validated area under the curve (AUC) on training dataset was 0.988 and the relative standard deviation (RSD%) was 0.062. AUC on the test dataset was 0.957. Accuracy (ACC), sensitivity, specificity, positive prediction value (PPV), and negative predictive value (NPV) were 0.926, 0.889, 0.937, 0.800, and 0.967, respectively. CONCLUSIONS: Machine learning models with radiomics features extracted from NECT scan accurately discriminated AVM-related intraparenchymal hematomas from those caused by other etiologies. This technique provided a fast, non-invasive approach without use of contrast to diagnose this disease. KEY POINTS: • Radiomics features from non-contrast-enhanced CT accurately discriminated AVM-related hematomas from those caused by other etiologies. • AVM-related hematomas tended to be larger in diameter, coarser in texture, and more heterogeneous in composition. • Adaboost classifier is an efficient approach for analyzing radiomics features.
OBJECTIVE: To investigate the classification ability of quantitative radiomics features extracted on non-contrast-enhanced CT (NECT) image for discrimination of AVM-related hematomas from those caused by other etiologies. METHODS: Two hundred sixty-one cases with intraparenchymal hematomas underwent baseline CT scan between 2012 and 2017 in our center. Cases were split into a training dataset (n = 180) and a test dataset (n = 81). Hematoma types were dichotomized into two classes, namely, AVM-related hematomas (AVM-H) and hematomas caused by other etiologies. A total of 576 radiomics features of 6 feature groups were extracted from NECT. We applied 11 feature selection methods to select informative features from each feature group. Selected radiomics features and the clinical feature age were then used to fit machine learning classifiers. In combination of the 11 feature selection methods and 8 classifiers, we constructed 88 predictive models. Predictive models were evaluated and the optimal one was selected and evaluated. RESULTS: The selected radiomics model was RELF_Ada, which was trained with Adaboost classifier and features selected by Relief method. Cross-validated area under the curve (AUC) on training dataset was 0.988 and the relative standard deviation (RSD%) was 0.062. AUC on the test dataset was 0.957. Accuracy (ACC), sensitivity, specificity, positive prediction value (PPV), and negative predictive value (NPV) were 0.926, 0.889, 0.937, 0.800, and 0.967, respectively. CONCLUSIONS: Machine learning models with radiomics features extracted from NECT scan accurately discriminated AVM-related intraparenchymal hematomas from those caused by other etiologies. This technique provided a fast, non-invasive approach without use of contrast to diagnose this disease. KEY POINTS: • Radiomics features from non-contrast-enhanced CT accurately discriminated AVM-related hematomas from those caused by other etiologies. • AVM-related hematomas tended to be larger in diameter, coarser in texture, and more heterogeneous in composition. • Adaboost classifier is an efficient approach for analyzing radiomics features.
Authors: Damiano G Barone; Hani J Marcus; Mathew R Guilfoyle; J Nicholas P Higgins; Nagui Antoun; Thomas Santarius; Rikin A Trivedi; Ramez W Kirollos Journal: Neurosurgery Date: 2017-07-01 Impact factor: 4.654
Authors: David Roh; Chung-Huan Sun; J Michael Schmidt; Edip Gurol; Santosh Murthy; Soojin Park; Sachin Agarwal; E Sander Connolly; Jan Claassen Journal: Neurocrit Care Date: 2018-08 Impact factor: 3.210
Authors: Joost J M van Griethuysen; Andriy Fedorov; Chintan Parmar; Ahmed Hosny; Nicole Aucoin; Vivek Narayan; Regina G H Beets-Tan; Jean-Christophe Fillion-Robin; Steve Pieper; Hugo J W L Aerts Journal: Cancer Res Date: 2017-11-01 Impact factor: 12.701
Authors: Hugo J W L Aerts; Emmanuel Rios Velazquez; Ralph T H Leijenaar; Chintan Parmar; Patrick Grossmann; Sara Carvalho; Sara Cavalho; Johan Bussink; René Monshouwer; Benjamin Haibe-Kains; Derek Rietveld; Frank Hoebers; Michelle M Rietbergen; C René Leemans; Andre Dekker; John Quackenbush; Robert J Gillies; Philippe Lambin Journal: Nat Commun Date: 2014-06-03 Impact factor: 14.919
Authors: Ji Eun Park; Donghyun Kim; Ho Sung Kim; Seo Young Park; Jung Youn Kim; Se Jin Cho; Jae Ho Shin; Jeong Hoon Kim Journal: Eur Radiol Date: 2019-07-26 Impact factor: 5.315