Nevin Manimala Statistics

Standard deviation of CT radiomic features among malignancies in each individual: prognostic ability in lung cancer patients

J Cancer Res Clin Oncol. 2023 Mar 8. doi: 10.1007/s00432-023-04649-7. Online ahead of print.


PURPOSE: It was reported that individual heterogeneity among malignancies (IHAM) might correlate well to the prognosis of lung cancer; however, seldom radiomic study is on this field. Standard deviation (SD) in statistics could scale average amount of variability of a variable; therefore, we used SD of CT feature (FeatureSD) among primary tumor and malignant lymph nodes (LNs) in an individual to represent IHAM, and its prognostic ability was explored.

METHODS: The enrolled patients who had accepted PET/CT scans were selected from our previous study (, NCT03648151). The patients had primary tumor and at least one LN, and standardized uptake value of LN higher than 2.0 and 2.5 were enrolled as the cohort 1 (n = 94) and 2 (n = 88), respectively. FeatureSD from the combined or thin-section CT were calculated among primary tumor and malignant LNs in each patient, and were separately selected by the survival XGBoost method. Finally, their prognostic ability was compared to the significant patient characteristics identified by the Cox regression.

RESULTS: In the univariate and multi-variate Cox analysis, surgery, target therapy, and TNM stage were significantly against OS in the both cohorts. In the survival XGBoost analysis of the thin-section CT dataset, none FeatureSD could be repeatably ranked on the top list of the both cohorts. For the combined CT dataset, only one FeatureSD ranked in the top three of both cohorts, but the three significant factors in the Cox regression were not even on the list. Both in the cohort 1 and 2, C-index of the model composed of the three factors could be improved by integrating the continuous FeatureSD; furthermore, that of each factor was obviously lower than FeatureSD.

CONCLUSION: Standard deviation of CT features among malignant foci within an individual was a powerful prognostic factor in vivo for lung cancer patients.

PMID:36884114 | DOI:10.1007/s00432-023-04649-7

By Nevin Manimala

Portfolio Website for Nevin Manimala