Correlative Hierarchical Clustering-based Low-Rank dimensionality reduction of radiomics-driven phenotype in Non-Small Cell Lung Cancer

摘要

Background: Lung cancer is one of the most common cancers in the United States and the most fatal, with 142,670 deathsin 2019. Accurately determining tumor response is critical to clinical treatment decisions, ultimately impacting patientsurvival. To better differentiate between non-small cell lung cancer (NSCLC) responders and non-responders to therapy,radiomic analysis is emerging as a promising approach to identify associated imaging features undetectable by the humaneye. However, the plethora of variables extracted from an image may actually undermine the performance of computer-aidedprognostic assessment, known as the curse of dimensionality. In the present study, we show that correlative-drivenhierarchical clustering improves high-dimensional radiomics-based feature selection and dimensionality reduction, ultimatelypredicting overall survival in NSCLC patients.Methods: To select features for high-dimensional radiomics data, a correlation-incorporated hierarchical clustering algorithmautomatically categorizes features into several groups. The truncation distance in the resulting dendrogram graph is used tocontrol the categorization of the features, initiating low-rank dimensionality reduction in each cluster, and providingdescriptive features for Cox proportional hazards (CPH)-based survival analysis. Using a publicly available non- NSCLCradiogenomic dataset of 204 patients’ CT images, 429 established radiomics features were extracted. Low-rankdimensionality reduction via principal component analysis (PCA) was employed ( = , > ) to find the representativecomponents of each cluster of features and calculate cluster robustness using the relative weighted consistency metric.Results: Hierarchical clustering categorized radiomic features into several groups without primary initialization of clusternumbers using the correlation distance metric (as a function) to truncate the resulting dendrogram into different distances.The dimensionality was reduced from 429 to 67 features (for truncation distance of 0.1). The robustness within the featuresin clusters was varied from -1.12 to -30.02 for truncation distances of 0.1 to 1.8, respectively, which indicated that therobustness decreases with increasing truncation distance when smaller number of feature classes (i.e., clusters) are selected.The best multivariate CPH survival model had a C-statistic of 0.71 for truncation distance of 0.1, outperforming conventionalPCA approaches by 0.04, even when the same number of principal components was considered for feature dimensionality.Conclusions: Correlative hierarchical clustering algorithm truncation distance is directly associated with robustness of theclusters of features selected and can effectively reduce feature dimensionality while improving outcome prediction.