Radiomic feature stability across 4D respiratory phases and its impact on lung tumor prognosis prediction

• DOI: 10.1371/journal.pone.0216480
• Publication Date: 2019-05-07T18:56:13Z
• Authors: Qian Du; Michael Baine; Kyle Bavitz; Josiah McAllister; Xiaoying Liang; Hongfeng Yu; Jeffrey Ryckman; Lina Yu; Hengle Jiang; Sumin Zhou; Chi Zhang; Dandan Zheng
• Keywords: Feature (linguistics)

Radiomic analysis has recently demonstrated versatile uses in improving diagnostic and prognostic prediction accuracy for lung cancer. However, since tumors are subject to substantial motion due respiration, the stability of radiomic features over respiratory cycle patient needs be investigated better evaluate robustness inter-patient feature variability clinical applications, its impact such applications assessed. A full panel 841 features, including tumor intensity, shape, texture, wavelet were extracted from individual phases a four-dimensional (4D) computed tomography on 20 early-stage non-small-cell cancer (NSCLC) patients. The each was assessed across different phase images same using coefficient variation (COV). relationship between COVs magnitude inspected. Population COVs, mean all patients, used categorize into 4 groups. two extremes, Very Small group (COV≤5%) Large (COV>20%), accounted about quarter features. Shape most stable, with COV≤10% study subsequently conducted 140 NSCLC employed predict overall survival 500-round bootstrapping. Identical multiple regression model development process applied, performance compared models without pre-selection step based 4D COV pre-exclude unstable Among systematically tested cutoff values, COV≤5% achieved optimal performance. resulting 3-feature significantly outperformed counterpart no pre-selection, P = 2.16x10-27 one-tailed t-test comparing performances models.