Identification of non-small-cell lung cancer subtypes by unsupervised clustering of CT image features with distinct prognoses and gene pathway activities

Abstract

ObjectiveIntra-tumour heterogeneity exists in non-small-cell lung cancer (NSCLC). Medical imaging can provide a comprehensive and detailed description of the tumour. Moreover, quantitative features can reflect the potentially valuable information of tumours that is neglected by visual diagnosis. To this end, we identified imaging subtypes of NSCLC by computed tomography (CT) images. MethodsWe analysed CT image data from a single-centre discovery cohort (n = 239), an internal validation cohort (n = 80) and an independent multi-centre external validation cohort (n = 63). Quantitative image features were extracted to identify intra-tumour heterogeneity. An unsupervised consensus clustering algorithm was used to identify and validate robust imaging subtypes. Then, we assessed the clinical relevance and prognostic value of these subtypes. Furthermore, differential gene expression profiles with copy number variation (CNV) data were used to identify the underlying biological patterns associated with subtypes and to map molecular activation patterns in dysregulated pathways. ResultsThree distinct imaging subtypes, spherical tumour clusters, irregular large tumour clusters and hollow tumour clusters, were identified and validated in the discovery and validation cohorts. Imaging subtypes showed significant survival differences in all cohorts. Moreover, each imaging subtype had a specific activated pattern in dysregulated pathways that can be targeted. ConclusionThe imaging subtypes provide complimentary prognostic value to establish non-invasive histopathological subtypes with molecular signatures and may help stratify NSCLC patients for targeted diagnosis and treatment. SignificanceThis study identified three imaging NSCLC subtypes that implied different prognosis and dysregulated pathways. These results may contribute to personalized diagnosis and treatment for NSCLC patients.