Automated delineation of non-small cell lung cancer

Abstract

PurposeThe tendencies of non-small cell lung cancers (NSCLC) to be large-sized, irregularly shaped, and to grow against the surrounding structures can cause even expert clinicians to experience difficulty with accurate segmentation.\n\nMethodsAn automated delineation tool based on spatial analysis was developed and studied on 25 sets of CT scans of primary NSCLCs with diverse radiological characteristics (sizes, shapes, contouring, localization, and microenvironment). Manual and automated gross delineations of the gross tumor were compared using a specific metric built based on spatially overlapping pixels.\n\nResultsThe proposed algorithm exhibited robustness in terms of the tumor size (5.32-18.24 mm), shape (spherical or non-spherical), contouring (lobulated, spiculated, or cavitated), localization (solitary, pleural, mediastinal, endobronchial, or tagging), and microenvironment (left or right lobe), with sensitivity, specificity, and accuracy rates of 80-98%, 85-99%, and 84-99%, respectively.\n\nConclusionsSmall discrepancies were observed between the manual and automated delineations. These might have risen from the variability in the practitioners definitions of ROIs or from imaging artifacts that reduced the tissue resolution.