Digital Health Innovation and Informatics

SS 16 - Digital Health Innovation & Informatics 1

121 - Prediction of Pseudoprogression Versus Progression Using Machine Learning Algorithm in Glioblastoma

Monday, October 22
5:05 PM - 5:15 PM
Location: Room 007 A/B

Prediction of Pseudoprogression Versus Progression Using Machine Learning Algorithm in Glioblastoma
B. S. Jang1, S. H. Jeon2, I. H. Kim2, and I. A. Kim3; 1Seoul National University Hospital, Seoul, Korea, Republic of (South), 2Department of Radiation Oncology, Seoul National University Hospital, Seoul, Korea, Republic of (South), 3Department of Radiation Oncology, Seoul National University Bundang Hospital, Seongnam, Gyeonggi, Korea, Republic of (South)

Purpose/Objective(s): We aimed to investigate the feasibility of machine learning (ML) algorithm to distinguish pseudoprogression (PsPD) from progression (PD) in follow-up glioblastoma (GBM) patients who completed the standard concurrent chemoradiation therapy after gross total resection (GTR).

Materials/Methods: From April 2010 to April 2017, we collected the clinical and imaging data from primary GBM patients who showed suspicious contrast-enhanced lesion in the follow-up brain magnetic resonance (MR) imaging after the standard therapy. We developed a convolutional neural network combined with a long short-term memory ML structure. The ML model was trained to classify PsPD and PD with nine axial gadolinium-enhanced T1-weighted MR images representing suspicious lesions and clinical factors. To evaluate the performance of the model, the area under the receiver operating characteristic curve (AUC), precision, recall, and F1-score were estimated.

Results: Data from two independent institutions were obtained for training (N=59) and testing (N=19), respectively. ML model was trained in the training set, resulting AUC of 0.90. The trained ML model showed the AUC of 0.76 in the testing set. By using the optimal threshold, our model achieved the sensitivity of 100% and the specificity of 73% for detecting PsPD. The overall precision and recall rate were 0.89 and 0.84, respectively, generating F1-score of 0.84.

Conclusion: We developed an ML algorithm that can predict PsPD versus PD with the selected nine axial MR images and clinical factors in follow-up GBM patients. The trained ML model showed an acceptable performance in the independent dataset, which could serve as a decision-making tool for early salvage treatment.
Summary of Model Evaluation
Precision rate Recall Rate F1-score
Pseudoprogression 0.73 1.00 0.84
True Progression 1.00 0.73 0.84
Average 0.89 0.84 0.84

Author Disclosure: B. Jang: None. I. Kim: None.

Send Email for Bum Sup Jang


Assets

121 - Prediction of Pseudoprogression Versus Progression Using Machine Learning Algorithm in Glioblastoma



Attendees who have favorited this

Please enter your access key

The asset you are trying to access is locked. Please enter your access key to unlock.

Send Email for Prediction of Pseudoprogression Versus Progression Using Machine Learning Algorithm in Glioblastoma