IMG-58. Radiomics can delineate lung cancer subtypes in metastatic lesions to the brain using magnetic resonance imaging
Year: 2025
Type: conference-abstract
Abstract: Lung cancer metastases to the brain represent nearly half of all metastatic lesions. Due to large differences in the clinical management, exact subtyping is required from surgically obtained samples either via intracranial biopsy or nodule/lesion sampling of the lung.. Near-instantaneous non-invasive subtyping of lung cancer lesions from routine diagnostic imaging would obviate the need for surgical sampling and greatly expedite treatment timelines. To accurately delineate CNS metastases from lung lesions as non-small cell lung cancer (NSCLC) lesions or small cell lung cancer (SCLC) using radiomic features derived from T1-weighted post-contrast imaging. A retrospective review was conducted from a large database of stereotactic radiosurgery-treated lesions at the University of Minnesota Medical Center (UMMC). T1-weighted post-contrast images from patients with histologically confirmed lung cancer with metastases to the brain were identified. Clinical demographics data for patients with non-small cell lung cancer (460 patients with 1896 metastatic lesions) and small cell lung cancer (78 patients with 648 lesions) were obtained. Radiomic features were extracted using the PyRadiomics library on semi-automatically obtained lesion segmentations. 199 radiomic features were extracted from each lesion. Using an XGBoost random classifier, lesions were split in a stratified manner 70:30 into training and test sets. Training data itself was split into 70:30 training-validation for hyperparameter tuning. A final F1 score of 0.61 (AUC-ROC score: 0.68) was obtained for the test set. Radiomic features have the ability to distinguish non-small cell lung cancer lesions from small cell lung cancer lesions. Histopathological differences among lung cancer subtypes may manifest as distinct radiographic phenotypes.
Source: Neuro-Oncology
Institution: University of Minnesota
Language: English
Subfield: Radiology, Nuclear Medicine and Imaging
Field: Medicine
Domain: Health Sciences
Open Access status: bronze
OpenAlex ID: W4416140474
Cites:
Cited by:
Related to: