CT-Based Radiomic Biomarker Features Predict Prognosis in Patients with Limited Stage Small Cell Lung Cancer

Abstract

Limited stage small cell lung cancer (LS-SCLC) has a poor prognosis (5-year survival <20%). Despite being initially sensitive to chemoradiation (CRT), most patients develop recurrences with limited salvage options. Early stratification of risk of local-regional recurrence (LRR)/poor survival may provide valuable information for selective intensification of treatment. Radiomics, large-scale quantitative imaging features, may be used as a potential biomarker to identify patients at higher risk of recurrence. We investigate the ability of radiomic features extracted from CT images to predict overall survival (OS), LRR, and distant metastases (DM). Between 2000 and 2013, 105 patients with biopsy-proven LS-SCLC treated with CRT were analyzed retrospectively. Planning CT scans were reviewed, and the gross tumor volume (GTV) was manually contoured. Based on unsupervised selection, we evaluated 18 stable and non-redundant radiomic features, and compared these with conventional features (volume and maximum diameter). Cox regression hazard ratio (HR) and concordance index (CI) were computed for clinical and imaging features, respectively. P-values were computed by comparing significance from random prediction. Clinical and treatment data are described in the table. With a median follow-up of 21.3 months (range, 0.6-113.4), 2-year local control was 62% (n = 38 failures; median time to failure of 8.6 months), 2-year metastasis-free survival was 48% (n = 55 events, median time to metastasis of 20.6 months), and 2-year OS was 47% (median survival of 22.3 months). One conventional imaging feature (maximum diameter) was found to be significantly associated with LRR (0.61 CI, P = 0.008). GTV volume did not show any association with clinical outcomes (CI range 0.5-0.58, P-values > 0.09). One radiomic feature capturing tumor elongation (shape) was associated with OS, LRR, and DM (CI 0.58, 0.63, 0.61, respectively, P-values< 0.05). Two textural radiomics features (patterns) were associated with LRR and OS with one sensitive to homogeneity (informal measure of correlation, CI = 0.58, P-value = 0.05) and one sensitive to tumor heterogeneity (entropy, CI = 0.43, P-value = 0.02), respectively. For LS-SCLC, we found several radiomic features associated with treatment outcomes after CRT. In particular, tumor elongation at time of treatment planning was established as a potential biomarker for LRR, DM, and OS. These features need further validation but demonstrate promise in the era of precision medicine.Tabled 1Abstract 27; Table 1 Clinical and treatment dataCharacteristicN (%)GenderFemale42 (40%)Male63 (60%)AgeMedian (range)63.8 (44.4-87.5)Performance score032 (30%)≥172 (70%)Smoking HistoryCurrent56 (53%)Former46 (44%)Never3 (3%)Pack-yearsMedian (range)45 (1.3-127.5)Tumor size (cm)Median (range)4.5 (0.7-11.5)Concurrent CRT92 (88%)Median RT dose (range)45 (20-66.6) Open table in a new tab