Status of CT textural/shape features as prognostic indicators for NSCLC

Intense research is being conducted throughout the world to identify biomarkers seen on computed tomography (CT) imaging that can help predict prognoses in patients with non-small- cell lung cancer (NSCLC). Such biomarkers could help improve and personalize treatment plans for patients.

Many researchers believe that textural features are potential imaging biomarkers, based on extensive research suggesting that they are an index of the degree of tumor heterogeneity. Much emphasis has been placed on the assessment of tumor heterogeneity through shape and textural features, because heterogeneity is believed to be associated with worse response to therapy and poor prognosis. However, studies conducted during the past 10 years have reported highly diverse and conflicting findings.

Researchers at the University of Perugia in Italy sought to determine if statistical criteria and protocols used were responsible for conflicting discrepancies and such diverse findings from analyses of shape and textural features of NSCLC seen on CT. Their concerns had been supported by Anastasia Chalkidou, PhD, and colleagues in the Division of Imaging Sciences and Bioengineering of King’s College London. After conducting an analysis of 15 published studies, they reported in PLoS One that they found insufficient evidence to support a relationship between PET or CT texture features and patient outcome.1 The King’s College London researchers said they were unable to find any two studies that identified the same texture feature and/or cut-off value as of prognostic significance. They also stated that “the most alarming finding was that in some cases the same texture feature was linked to both positive and negative patient outcomes in different studies.”


A team of radiologists, engineers, and biomedical specialists at the University of Perugia evaluated 30 three-dimensional shape and textural CT-derived features as potential biomarkers predictive of overall survival of 203 NSCLC patients. They employed two types of statistical analysis that were used in related studies with four different statistical protocols. Statistically significant features associated with prognosis differed dramatically based on the protocols used, ranging from 2 to 17. The two features identified as significant by each type of analysis were volume and gray-level run length matrix (GRLM) run length non-uniformity, both of which correlated negatively with overall survival.

Patient data was obtained from a subset of the publicly available NSCLC Radiomics collection at the Cancer Imaging Archive, a large archive of anonymized medical images hosted by the University of Arkansas for Medical Science in Little Rock. The authors used nine shape and 21 textural features. They used Cox proportional hazards univariate regression analysis and Kaplan-Meier survival analysis to assess the predictive power of the image features.They then analyzed the results of the two statistical models with four evaluation protocols differing in strictness.

The least strict model identified 17 statistically significant features, whereas the most identified two. Lead author Francesco Bianconi, PhD, of the Department of Engineering reported that there was an increased relative risk of negative outcome of about 48% per 1 cm3 increment of lesion volume . The increase in the relative risk was approximately 6% per unit with respect to GLRLM run length uniformity, and because of this, the authors recommend further studies to validate the use of this parameter in clinical study.

“We found insufficient evidence to claim a relationship between heterogeneity and overall survival in this study,” the authors wrote in the April issue of Anticancer Research. “Classic textural features that are indicative of heterogeneity failed to reach statistical significance in our calculations. Likewise, the average tissue density, which was considered significant in previous studies, gave divergent results in this work.”

The authors said that the results of their study seemed to confirm on a quantitative basis the risk of type-I error inflation and of increased false-discovery rates in studies with radiomic features stated in the PLoS One article.


  1. Chalkidou A, O'Doherty MJ, Marsden PK. False discovery rates in PET and CT studies with texture features: A systematic review. PLoS One. 2015. 10;5:e0124165.
  2. Bianconi F, Fravolini ML, Bello-Cerezo R, et al. Evaluation of shape and textural features from CT as prognostic biomarkers in non-small-cell lung cancer. Anticancer Res. 2018 38;4:2155-2160.
© Anderson Publishing, Ltd. 2024 All rights reserved. Reproduction in whole or part without express written permission Is strictly prohibited.