SU‐F‐J‐207: Non‐Small Cell Lung Cancer Patient Survival Prediction with Quantitative Tumor Textures Analysis in Baseline CT

Abstract

Purpose: Chemo‐radiation therapy (CRT) is widely used in treating patients with locally advanced non‐small cell lung cancer (NSCLC). Determination of the likelihood of patient response to treatment and optimization of treatment regime is of clinical significance. Up to date, no imaging biomarker has reliably correlated to NSCLC patient survival rate. This pilot study is to extract CT texture information from tumor regions for patient survival prediction. Methods: Thirteen patients with stage II‐III NSCLC were treated using CRT with a median dose of 6210 cGy. Non‐contrast‐enhanced CT images were acquired for treatment planning and retrospectively collected for this study. Texture analysis was applied in segmented tumor regions using the Local Binary Pattern method (LBP). By comparing its HU with neighboring voxels, the LBPs of a voxel were measured in multiple scales with different group radiuses and numbers of neighbors. The LBP histograms formed a multi‐dimensional texture vector for each patient, which was then used to establish and test a Support Vector Machine (SVM) model to predict patients’ one year survival. The leave‐one‐out cross validation strategy was used recursively to enlarge the training set and derive a reliable predictor. The predictions were compared with the true clinical outcomes. Results: A 10‐dimensional LBP histogram was extracted from 3D segmented tumor region for each of the 13 patients. Using the SVM model with the leave‐one‐out strategy, only 1 out of 13 patients was misclassified. The experiments showed an accuracy of 93%, sensitivity of 100%, and specificity of 86%. Conclusion: Within the framework of a Support Vector Machine based model, the Local Binary Pattern method is able to extract a quantitative imaging biomarker in the prediction of NSCLC patient survival. More patients are to be included in the study.

Development of 365-day Life Expectancy Models: Application of Machine Learning Methods to a Prospective Study of Critically Ill Cirrhotic Patients (Preprint)

“…The main strength of this study is the use of models based on machine learning validated by leave-one-out cross-validation using prospective data, which demonstrated the clinical utility of the SVM/Ensemble/RANSAC/Random Forest methods in predicting survival [36][37][38][39]. We believe that our results may have positive clinical implications, because the information provided is objective and can be used in family meetings to help set expectations for reasonable clinical care.…”

Tracking knowledge evolution, hotspots and future directions of emerging technologies in cancers research: a bibliometrics review

“…Prediction of survival periods is generally difficult because of the lack of sufficient and comprehensive data. Nowadays, medical data and survey data after treatments can be acquired and combined with patients' electronic medical records 48, 49, 50. In the past, emerging information technologies have been applied in diagnosis and treatment, but application in the prognosis after treatments, such as the prediction of survival, is rare.…”