ESTRO 2020 Abstract Book

S825 ESTRO 2020

PO-1527 When small is too small? Training Deep Learning models in limited datasets. G. Valdes 1 , M. Romero 2 , Y. Interian 2 , T. Solberg 1 1 University of California UCSF, Radiation Oncology, San Francisco CA, USA ; 2 University of San Francisco, Data Science, San Francisco, USA Purpose or Objective To perform an in depth evaluation of current state of the art techniques in training neural networks to identify appropriate approaches in small datasets. Material and Methods 112,120 frontal-view x-ray images from the NIH ChestXray14 dataset were used in our analysis . Two tasks were studied: unbalanced multi-label classification of 14 diseases, and binary classification of pneumonia vs non- pneumonia . All datasets were randomly split into training, validation and testing (70%, 10%, 20%). A popular convolution neural network (CNN), DensNet121, was trained using PyTorch. We performed several experiments to test: 1) whether transfer learning using pre-trained networks on ImageNet are of value to medical imaging / physics tasks (e.g., predicting toxicity from radiographic images after training on images from the internet), 2) whether using pre-trained networks trained on problems that are similar to the target task helps transfer learning (e.g., using x-ray pre-trained networks for x-ray target tasks), 3) whether freezing deep layers or changing all weights provides an optimal transfer learning strategy, 4) the best strategy for the learning rate policy and, 5) what quantity of data is needed in order to appropriately deploy these various strategies (N=50 to N=77,880). Finally, linear models using radiomics features (79 in total) were trained as baseline and confident intervals were calculated using bootstrap. Results In the multi-label problem, DensNet121 needed at least 1600 patients to be comparable to, and 10,000 to outperform, radiomics-based logistic regression. In classifying pneumonia vs. non-pneumonia, both CNN and radiomics-based methods performed poorly when N < 2000. For small datasets (< 2000), however, a significant boost in performance (> 15% increase on AUC) comes from a good selection of the transfer learning dataset, dropout, cycling learning rate and freezing and unfreezing of deep layers as training progresses. In contrast, if sufficient data is available (>35000), little or no tweaking is needed to obtain impressive performance. While transfer learning using x-ray images from other anatomical sites improves performance, we also observed a similar boost by using pre-trained networks from ImageNet, Figure 1. Having source images from the same anatomical site, however, outperforms every other methodology, by up to 15%., Figure 1 In this case, DL models can be trained with as little as N=50, Figure 1.

(N>35 000). Using transfer learning with images from the same anatomical site can yield remarkable performance in new tasks with as few as N=50. Surprisingly, we did not find any advantage for using images from other anatomical sites over networks that have been trained using ImageNet. This indicates that features learned may not be as general as currently believed, and performance decays rapidly even by just changing the anatomical site of the images. PO-1528 Predicting response to neoadjuvant chemoradiation in esophageal cancer using CT radiomic features I. Mallick 1 , V.H. Lal 1 , M. Arunsingh 1 , T. Bhattacharyya 1 , S. Chatterjee 1 , S. Chakraborty 1 , R. Achari 1 1 Tata Medical Center, Department of radiation Oncology, Kolkata, India Purpose or Objective To determine if CT based radiomic feature extraction can identify complete pathological responders from neoadjuvant chemoradiation (NACRT) in operable esophageal cancer. Material and Methods From a dataset of 41 patients with squamous cell carcinoma (SCC) of the esophagus who underwent NACRT with 41.4 Gy of radiation and concurrent weekly paclitaxel and carboplatin followed by esophagectomy we identified 20 patients with a complete pathological response (pT0) and 21 patients with a poor response (pT3). The tumor was delineated on the planning CT scan. DICOM files were read and radiomic features extracted using open source libraries (pydicom and PyRadiomics). Univariate feature selection was performed and several classifiers (logistic regression, k-nearest neighbors, random forest and support vector machines) were trained. The model was tested on a test data sub-set. Results Of the 107 radiomic parameters tested, using univariate feature extraction we identified eight radiomic parameters that were different in the two groups. These included first order features (kurtosis), gray level co- occurrence matrix features (inverse difference, inverse difference normalized, inverse difference moment normalized), gray level run length matrix features (runlength non-uniformity normalized, run percentage), neighbouring gray tone difference matrix features (contrast) and gray level difference matrix features (dependence non-uniformity normalized). The random forest classifier performed the best and had a precision of 0.67 for predicting pCR, an F1 score of 0.67 and area under curve of 0.78. The accuracy score was modest at 0.56. Conclusion In this exploratory dataset, CT based radiomic features had modest success at classifying complete pathological response. Larger datasets and multimodality imaging need to be used to improve the accuracy of prediction. PO-1529 Increased cerebral radiosensitivity of the periventricular region in proton therapy of gliomas A. Lühr 1 , F. Raschke 2 , C. Karpowitz 3 , F. Permatasari 1 , B. Lutz 4 , W. Enghardt 1 , M. Krause 1 , E. Troost 1 , J. Eulitz 1 1 OncoRay - National Center for Radiation Research in Oncology, Institute of Radiooncology – OncoRay, Dresden, Germany ; 2 Helmholtz-Zentrum Dresden - Rossendorf, Institute of Radiooncology – OncoRay, Dresden, Germany ; 3 Faculty of Medicine and University Hospital Carl Gustav Carus, Department of Radiotherapy and Radiation Oncology, Dresden, Germany ; 4 Helmholtz-Zentrum Dresden - Rossendorf, Institute for Radiation Physics, Dresden, Germany Purpose or Objective In neuro-radiooncology, a constant radiosensitivity is assumed within the brain. However, recent findings

Conclusion While training DL models in small datasets (N<2000) is challenging, no tweaking is necessary for bigger datasets