Automated Detection and Characterization of Small Cell Lung Cancer Liver Metastases on CT

Sophia Ty, Fahmida Haque, Parth Desai, Nobuyuki Takahashi, Usamah Chaudhary, Benjamin Simon, Peter L. Choyke, Anish Thomas, Baris Turkbey, Stephanie A. Harmon

Abstract: Small cell lung cancer patients with metastatic disease often present with liver lesions that can be detected in computed tomography (CT). These lesions have highly diverse phenotypes resulting from heterogenous expression of various neuroendocrine (NE) genes. In this work, we present a two-step machine learning framework to automatically detect liver lesions in CT scans using a 3D segmentation model followed by radiomics-based analysis to stratify the patients based on their NE tumor profile. Our liver lesion detection model achieved a lesion sensitivity greater than 66% and up to 83% for two datasets. The radiomics-based neuroendocrine phenotype classifier can distinguish patients based on their NE expression profile (NE vs. non-NE) and achieved an F1 score of 85%.