Related content: Contrast agents, Pulmonary, Oncology

Yoon SH, Park CM, Park SJ, Yoon JH, Hahn S, Goo JM. Tumor Heterogeneity in Lung Cancer: Assessment with Dynamic Contrast-enhanced MR Imaging. Radiology. 2016;280(3):940-8.

Predicting patient survival by texture analysis

Entropy values derived from dynamic contrast-enhanced MRI might serve as a reliable parameter to describe intratumoral heterogeneity of lung cancers. The result is linked to patient survival.

Prediction of survival for patients with lung cancer beyond TNM classification (size of original tumor, lymph nodes involved, distant metastasis) is possible by characterizing intratumoral heterogeneity. All patients had undergone a DCE-MRI with gadopentetate dimeglumine (Magnevist®) at a 1.5T device for treatment decision. Soon Ho Yoon and colleagues, Seoul National University College of Medicine, Seoul, Korea, evaluated the ability of several parameters derived from DCE-MRI for assessment of texture analysis and prediction of survival.


Images of 38 patients with pathologically proven lung cancer were included in the retrospective study. All patients had undergone a DCE-MRI with Magnevist® at a 1.5T device for treatment decision. Sequences were performed every 30 seconds up to 300 seconds, revealing eleven image sets, plus one sequence at 480 seconds after contrast injection.

Texture analysis

Lesion segmentation became possible by using specialized software. One experienced radiologist and one experienced MR technician manually encircled the cancer independently of each other and analyzed mean signal intensity, histogram and texture in the region of interest (ROI) using the following parameters:

  • Standard deviation (of the pixel distribution histogram)
  • Skewness (asymmetry of the histogram)
  • Kurtosis (flatness of the histogram)
  • Entropy (irregularity of gray-level distribution)
  • Homogeneity (of the image)

High standard deviation, negative or positive skewness, high kurtosis, high entropy and lower uniformity are thought to represent increased heterogeneity in the ROI. Yoon et al. performed ROC curve analysis to find the optimal parameters and the optimal time point for describing texture.

Prediction of survival

Yoon et al. included the following parameters into their two-year progression-free survival analysis:

  • Age
  • Sex
  • Smoking history
  • Overall cancer stage
  • Treatment purpose (palliative or curative)
  • Maximum standardized uptake value measured at PET/CT
  • Eastern Cooperative Oncology Group performance status score

Yoon et al. performed univariate and multivariate COX regression analysis for imaging parameters and survival rates to find the most suitable parameter for prediction of survival.


Texture analysis: Significant correlation was observed in entropy with changes in histogram and texture parameters. Entropy fitted best with changes in histogram and texture parameters over time.

Prediction of survival: With univariate COX analysis, standard deviation at 150 seconds (p=0.006) and entropy at 120 seconds (p=0.001) or 180 seconds (p<0.001) showed to be good predictors for progression-free survival.
With multivariate COX analysis, entropy at 120 seconds or 180 seconds turned out to be the best predictor for two-year survival (p=0.010 and p=0.015, respectively). Interobserver variability was least in entropy measurements.


Yoon et al. identified entropy measurement 120 seconds after contrast injection as best parameter for prediction of survival of patients with lung cancer. Entropy showed the best correlation changes in mean signal intensity, texture parameters and histogram over time after contrast injection.

However, Yoon et al. admitted that two-year overall survival is not a surrogate endpoint for five-year overall survival. Their study also lacked a reference standard, e.g., histopathologic correlation analysis.