Download This Paper
Preprints with The Lancet is a collaboration between The Lancet Group of journals and SSRN to facilitate the open sharing of preprints for early engagement, community comment, and collaboration. Preprints available here are not Lancet publications or necessarily under review with a Lancet journal. These preprints are early-stage research papers that have not been peer-reviewed. The usual SSRN checks and a Lancet-specific check for appropriateness and transparency have been applied. The findings should not be used for clinical or public health decision-making or presented without highlighting these facts. For more information, please see the FAQs.
Machine-Learning-Assisted Radiomic Biomarkers for Early Detection of Radiation-Induced Brain Injury in Nasopharyngeal Carcinoma
30 Pages Posted: 30 Jan 2019
More...Abstract
Background: Radiation-induced temporal lobe injury (RTLI) is a progressive neurological complication in nasopharyngeal carcinoma (NPC) after radiotherapy, and is associated with severe intelligence, memory, and speech impairments. Early RTLI diagnosis is very challenging, and predictive biomarkers of RTLI are lacking. Thus, we aimed to develop noninvasive radiomic biomarkers for early detection of RTLI.
Methods: Between January 2006 and August 2016, we retrospectively evaluated 242 NPC patients who had undergone baseline and regular follow-up magnetic resonance imaging (MRI) scans. The clinical endpoint was RTLI, which was diagnosed using MRI. The radiomic pipeline involved MRI data pre-processing, medial temporal lobe segmentation, feature extraction, feature selection, and predictive modeling. Total 24 non-texture and 61,920 texture features were extracted from the medial temporal lobe. A 0.632 + bootstrap and the area under the receiver operating characteristic curve (AUC) were applied for feature selection. The data were randomly divided into training set (n = 63.2%) and validation set (n = 36.8%). Random forest was used to construct the prediction model. Three models, 1, 2 and 3, were developed for prediction 1, 2, and 3 years before RTLI onset, respectively. Different numbers of selected features were tested for optimization. AUC was used to assess the predictive performance.
Findings: Of the 242 included patients, 171 (70.7%) were men, and the mean (standard deviation) age was 48.5 (10.4) years. The median follow-up time and latency from radiotherapy until the first RTLI were 46 and 41 months, respectively. In the validation set, models 1, 2, and 3, with 20 radiomic features derived from the whole medial temporal lobe, yielded maximum AUCs of 0.830 (95% CI: 0.823-0.837), 0.773 (0.763-0.782), and 0.716 (0.699-0.733), respectively.
Interpretation: The three machine-learning-assisted radiomic models can dynamically predict RTLI 1-3 years in advance, enabling early detection and preventive neurological intervention.
Funding Statement: This research was supported by a grant of the National Scientific Foundation of China (81571664, 81801665), the Science and Technology Planning Project of Guangdong Province (2014A020212244, 2016A020216020), the Scientific Research General Project of Guangzhou Science Technology and Innovation Commission (201605110912158), the China Postdoctoral Science Foundation (2016M600145), and the Guangdong Grand Science and Technology Special Project (2015B010106008).
Declaration of Interests: The authors have declared that no competing interest exists.
Ethics Approval Statement: This retrospective, longitudinal study was approved by the Ethics Committee of our institution, which waived the requirement for informed patient consent because the data for all subjects were anonymized.
Keywords: Radiation-induced temporal lobe injury; Nasopharyngeal carcinoma; Radiomics; Machine learning
Suggested Citation: Suggested Citation