SU-F-R-27: Use Local Shape Descriptor Based On Geodesic Distance to Predict Survival in Non-Small Cell Lung Cancer After Radiotherapy

Abstract

Purpose: The shape of the Positron Emission Tomography (PET) image represents the heterogeneity of tumor growth in various directions, and thus could be associated with tumor malignancy. We have proposed a median geodesic distance (MGD) to represent the local complexity of the shape and use a normalized MGD (NMGD) to quantify the shape, and found a potential correlation of NMGD to survival in a 20-patient pilot study. This study was to verify the finding in a larger patient cohort. Methods: Geodesic distance of two vertices on a surface is defined as the shortest path on the surface connecting the two vertices. The MGD was calculated for each vertex on the surface to display the local complexity of the shape. The NMGD was determined as: NMGD = 100*standard deviation(MGDs)/mean(MGDs). We applied the NMGD to 40 NSCLC patients who were enrolled in prospective PET image protocols and received radiotherapy. Each patient had a pre-treatment PET scan with the resolution of 4mm*4mm*5mm. Tumors were contoured by a professional radiation oncologist and triangulation meshes were built up based on the contours. Results: The mean and standard deviation of NMGD was 6.4±3.0. The OS was 33.1±16.9 months for low NMGD group, and 15.4±15.6 months for the high NMGD group. The low NMGD group had significant better OS than the high NMGD group (p=0.0013). Conclusion: NMGD could be used as a shape biomarker to predict survival and the MGD could be combined with image texture in future to increase prediction accuracy. This study was supported by Award Number 1R01CA166948 from the NIH and National Cancer Institute.