Premium
WE‐D‐M100J‐04: Prediction of Lung Radiation‐Induced Pneumonitis Using the Support Vector Machine Algorithm
Author(s) -
Chen S,
Zhou S,
Zhang J,
Marks L,
Das S
Publication year - 2007
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.2761565
Subject(s) - support vector machine , radiation therapy , pneumonitis , lung cancer , computer science , dose volume histogram , lung volumes , nuclear medicine , lung , artificial intelligence , medicine , pattern recognition (psychology) , mathematics , machine learning , radiation treatment planning , radiology , pathology
Purpose: To build and test a Support Vector Machine (SVM) model to predict the occurrence of lung radiation‐induced Grade 2+ pneumonitis. SVM is a sophisticated statistical technique that is capable of using complex hypersurfaces to separate the cases with and without pneumonitis. Method and Materials: Two SVM models were built using data from 235 patients with lung cancer treated using radiotherapy (34 diagnosed with pneumonitis). One model (SVM all ) selected input features from all dose‐volume and non‐dose factors. For comparison, the other model (SVM dose ) selected input features only from lung dose‐volume factors. The models were built with in‐house developed software that employed a unique strategy to sequentially add/remove/substitute features. The SVM models were tested using ten‐fold cross‐validation, wherein 1/10 th of the data were tested, in turn, using the model built with the remaining 9/10 th of the data. Results: The input features selected to build SVM all were the lung generalized equivalent uniform dose (EUD) with exponents a=1.2, 1.3, 1.4, chemotherapy prior to radiotherapy (yes/no), tumor location (central/peripheral), gender, and histology (adenocarcinoma/other; small cell/other). The input features for SVM dose were EUD a = 1.1, 1.3, 1.4, lung volume receiving > 48 Gy (V48), and V50. Both models selected EUD a ≈ 1 (EUD a=1 is the mean lung dose, which frequently appears as a strong predictor of radiation pneumonitis in literature). The area under the cross‐validated SVM all Receiver Operating Characteristics curve was 0.76 (sensitivity/specificity = 74%/75%), compared to the corresponding SVM dose area of 0.71 (sensitivity/specificity = 68%/68%). SVM all was statistically superior (p=0.01), indicating that non‐dose features significantly contribute to separating patients with and without pneumonitis. Conclusions: The SVM model constructed from dose and non‐dose input factors is a valuable prospective tool for predicting the occurrence of radiation‐induced lung pneumonitis.