An integrated baseline CT model for predicting growth heterogeneity in solid lung cancer with transcriptomic annotation
Related Experiment Videos
Jiaqi Chen1, Linlin Qi1, Jianing Liu2
1Department of Diagnostic Radiology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China.
Background:
Solid nodular lung cancer exhibits significant heterogeneity in growth behavior, and the speed of growth is significantly correlated with prognosis. This study aimed to develop and validate an integrated baseline model for predicting growth heterogeneity in solid lung cancer and to provide exploratory transcriptomic annotation.
Methods:
This study included 585 retrospectively collected surgically resected solid lung cancer nodules from March 2011 to December 2025, randomly divided into training and internal test cohorts at a 7:3 ratio. Forty-two nodules prospectively enrolled from January 2026 to June 2026 were used for prospective time validation. Fast and slow growth were defined by first-to-last volume doubling time (VDT), with a threshold of 400 days. Clinical, computer tomography (CT) morphological, intratumoral and peritumoral radiomics features were extracted from baseline CT. Least absolute shrinkage and selection operator (LASSO) logistic regression was used for feature selection, and three score systems and seven predictive models were constructed. Whole transcriptome sequencing was performed in a quality-controlled subset of 54 lung adenocarcinoma nodules.
Results:
The VDT of all enrolled nodules was 410.0 [interquartile range (IQR), 206.0-828.0] days, with a median follow-up interval of 173.0 (IQR, 80.0-481.0) days. The training, testing, and prospective time validation cohorts consisted of 409 (rapid vs. slow: 199 vs. 210), 176 (rapid vs. slow: 87 vs. 89), and 42 nodules (rapid vs. slow: 21 vs. 21), respectively. The Clinical-Intra-Peri model achieved the highest area under the curve (AUC) in the training and internal test cohorts, with AUCs of 0.754 and 0.770, respectively. Bootstrap internal validation yielded an optimism-corrected AUC of 0.750. In the prospective temporal validation cohort, the AUC was 0.825. Transcriptomic annotation showed that MYC targets V2 and unfolded protein response pathways, with leading-edge genes H2AX and MCM4, suggesting potential links between model-derived risk and growth-related transcriptomic activity.
Conclusions:
The baseline Clinical-Intra-Peri CT model showed potential for predicting growth heterogeneity in solid lung cancer. Exploratory transcriptomic annotation suggested that model-defined risk was associated with growth-related transcriptional programs.