ITHscore: comprehensive quantification of intra-tumor heterogeneity in NSCLC by multi-scale radiomic features

Objectives To quantify intra-tumor heterogeneity (ITH) in non-small cell lung cancer (NSCLC) from computed tomography (CT) images. Methods We developed a quantitative ITH measurement—ITHscore—by integrating local radiomic features and global pixel distribution patterns. The associations of ITHscore with tumor phenotypes, genotypes, and patient’s prognosis were examined on six patient cohorts ( n = 1399) to validate its effectiveness in characterizing ITH. Results For stage I NSCLC, ITHscore was consistent with tumor progression from stage IA1 to IA3 ( p < 0.001) and captured key pathological change in terms of malignancy ( p < 0.001). ITHscore distinguished the presence of lymphovascular invasion ( p = 0.003) and pleural invasion ( p = 0.001) in tumors. ITHscore also separated patient groups with different overall survival ( p = 0.004) and disease-free survival conditions ( p = 0.005). Radiogenomic analysis showed that the level of ITHscore in stage I and stage II NSCLC is correlated with heterogeneity-related pathways. In addition, ITHscore was proved to be a stable measurement and can be applied to ITH quantification in head-and-neck cancer (HNC). Conclusions ITH in NSCLC can be quantified from CT images by ITHscore, which is an indicator for tumor phenotypes and patient’s prognosis. Key Points • ITHscore provides a radiomic quantification of intra-tumor heterogeneity in NSCLC. • ITHscore is an indicator for tumor phenotypes and patient’s prognosis. • ITHscore has the potential to be generalized to other cancer types such as HNC.