Suter, Yannick Raphael; Jungo, Alain; Rebsamen, Michael; Knecht, Urspeter; Herrmann, Evelyn; Wiest, Roland; Reyes, Mauricio (26 January 2019). Deep Learning Versus Classical Regression for Brain Tumor Patient Survival Prediction. Lecture notes in computer science, 11384, pp. 429-440. Cham: Springer 10.1007/978-3-030-11726-9_38
![[img]](https://boris.unibe.ch/style/images/fileicons/text.png) |
Text
Suter2019_Chapter_DeepLearningVersusClassicalReg.pdf - Published Version Restricted to registered users only Available under License Publisher holds Copyright. Download (2MB) |
Deep learning for regression tasks on medical imaging datahas shown promising results. However, compared to other approaches,their power is strongly linked to the dataset size. In this study, we eval-uate 3D-convolutional neural networks (CNNs) and classical regressionmethods with hand-crafted features for survival time regression of pa-tients with high-grade brain tumors. The tested CNNs for regressionshowed promising but unstable results. The best performing deep learn-ing approach reached an accuracy of 51.5% on held-out samples of thetraining set. All tested deep learning experiments were outperformed bya Support Vector Classifier (SVC) using 30 radiomic features. The inves-tigated features included intensity, shape, location and deep features.The submitted method to the BraTS 2018 survival prediction challenge isan ensemble of SVCs, which reached a cross-validated accuracy of 72.2%on the BraTS 2018 training set, 57.1% on the validation set, and 42.9%on the testing set.The results suggest that more training data is necessary for a stable per-formance of a CNN model for direct regression from magnetic resonanceimages, and that non-imaging clinical patient information is crucial alongwith imaging information.
Actions (login required)
 |
Edit item |