The potential of radiomics with PET/CT: study of correlations with the metabolic profile and its discriminative power

Abstract

Treatment of lung cancer is still a challenge, partly due to late-stage diagnosis of patients. Focusing on non-small cell lung cancer (NSCLC), metabolic biomarkers from blood plasma (metabolomics) and features from medical images (radiomics) are examined. This study combines metabolomics and radiomics datasets from NSCLC patients, unravels the underlying correlations and generate discriminative models based on radiomics features. Two patient cohorts of (i) 39 patients and (ii) 85 patients were used. All patients were diagnosed with early-stage, locally advanced NSCLC and underwent a chirurgical resection of the lung tumor. PET/CT images of all patients were collected and segmented. From each volume of interest, 483 parameters are extracted. From 39 patients, 238 metabolic parameters representing 62 plasma metabolites are determined using proton nuclear magnetic resonance (1H-NMR) spectroscopy. A correlation test is used on the total omics-dataset of 39 patients. Logistic regression is used to generate the discriminative models based on radiomics parameters from 85 patients. The correlation matrices show that glucose and glycerol are strongly correlated with specific radiomics features. These results suggest new insights in PET/CT interpretation and that more plasma metabolites might be correlated with features out of PET/CT images. The discriminative models are built using two radiomics features and can distinguish between malignant/non-malignant PET-positive lung nodules, and between adenocarcinoma/ squamous cell carcinoma.