Clinical and Radiomics Features for Prognostication in Juvenile Myoclonic Epilepsy Patients
Abstract number :
2.182
Submission category :
5. Neuro Imaging / 5A. Structural Imaging
Year :
2022
Submission ID :
2204146
Source :
www.aesnet.org
Presentation date :
12/4/2022 12:00:00 PM
Published date :
Nov 22, 2022, 05:23 AM
Authors :
Kyung Min Kim, MD – Yonsei University College of Medicine; Hye Jeong Lee, MD – Clinical Fellow, Neurology, Yonsei University College of Medicine; Won-Joo Kim, MD, PhD – Professor, Neurology, Yonsei University College of Medicine; Kyoung Heo, MD, PhD – Professor, Neurology, Yonsei University College of Medicine
Rationale: This study aimed to build and validate clinical and radiomics prediction models that could predict the prognosis of patients with juvenile myoclonic epilepsy (JME).
Methods: A total of 92 subjects were assigned to a training (n=64) or a test set (n=28) group. Clinical characteristics of JME patients including gender, onset age, epilepsy duration, family history, febrile seizure history, presence of absence seizure, and 2-year seizure freedom were collected. Radiomics features were extracted from 20 regions of interest from T1-weighted magnetic resonance images (MRI). Several machine learning models were trained with the patients’ clinical and radiomics features during training sets, and they were validated with test sets.
Results: The six tested clinical and radiomics models – light gradient boosting machine, extreme gradient boosting, random forest, decision tree, support vector classifier, and logistic regression – showed an area under the curve (AUC) of 0.72, 0.80, 0.58, 0.46, 0.50, 0.51, respectively. The best-performing model, the extreme gradient boosting, demonstrated an accuracy, precision, recall, and F1 score of 0.79, 0.78, 0.95, and 0.86, respectively. The model combined with clinical and radiomics features (0.80) showed better AUC results than that with only clinical characteristics (0.63).
Conclusions: Machine learning models combined with clinical characteristics and radiomics features could help predict the prognosis of JME patients.
Funding: None
Neuro Imaging