Room: Exhibit Hall | Forum 8
Purpose: We tested the hypothesis: radiomic features could predict the cancer prognosis for lung cancer patients with stereotactic body radiotherapy (SBRT) using slow CT images.
Methods: Sixty-eight lung cancer patients treated with SBRT were analyzed. Patients received a dose of 6.25-12 GyÃ—4-8 fractions. Pre-treatment slow CT images (4 s/slice) obtained by CT scanner (GE) were used for radiomics analysis. A total of 1610 radiomic features describing morphology (e.g., volume, compactness), first-order histogram-based features (e.g., skewness, kurtosis), second-order texture-based features (e.g., Gray Level Co-occurrence Matrix textures, Run-length metrices) were extracted for each patient. To clarify the prognostic power of radiomics, the relationship between radiomic features and cancer specific survival was evaluated by Kaplan-Meier method.
Results: For 23 radiomic features based on stability and variance, Kaplan-Meier curves were significantly different (p<0.05) between groups with high and low radiomic feature values (morphology: n=2, first-order histogram-based features: n=1, second-order texture-based features: n=20). For example, Cluster Prominence and Gray level nonuniformity was significantly prognostic for overall survival (Cluster Prominence: p<0.01, Gray level nonuniformity: p<0.046).
Conclusion: Our result showed that a significant difference occurred in the survival curves of 23 radiomic features using slow CT-based radiomics. This study showed that radiomics using slow CT images has the potential for useful tool to predict the prognosis in SBRT lung cancer patients, even though slow CT had larger motion blur caused by respiratory motion.
IM/TH- Image Analysis (Single modality or Multi-modality): Imaging biomarkers and radiomics