PURPOSE: Due to increased use of high-resolution computed tomography (HRCT), sub-solid pulmonary nodules of the adenocarcinoma spectrum with unknown clinical significance are increasingly identified. New strategies for non-invasive risk stratification to guide individualized treatment decisions regarding radiological observation versus surgical intervention, and the extent of the required resection are urgently needed. In these tumors, absent or minimal histopathologic tissue invasion has been associated with excellent patient outcomes.
METHODS: Histopathologic tissue invasion was quantified for 55 patients with surgically resected pulmonary nodules (≤ 3cm) of adenocarcinoma subtype spectrum (2005-2009) by at least two independent pulmonary pathologists. These were separated into “non-invasive” (≤ 5mm invasion) and “invasive” (> 5mm invasion) lesions. Every patient had a pre-operative HRCT with consecutive <1.5mm cuts. The computer-aided image analysis tool CALIPER was used to analyze 774 regions of interest (ROI) within 37 sub-solid nodules spanning the spectrum of histopathological invasiveness. ROI were stratified into natural clusters using affinity propagation, an unsupervised clustering technique. This yielded 9 unique exemplars (color coded radiological types) constituting the basic components of these subsolid lung nodules. Two groups of clusters were found to correlate with invasion (Violet-Indigo-Red-Orange) and lepidic growth (Blue-Green-Cyan). Individual voxels within the nodules of all 55 patients were mapped to the most relevant exemplar and this information was used to radiologically quantify invasion (V-I-R-O) in three-dimensional space. Sensitivity, specificity, positive predictive value and negative predictive value for the detection of tissue invasion by CALIPER were calculated.
RESULTS: The sensitivity of CALIPER was 88.2% [71.6-96.1%], the specificity was 100% [80.7-100%], the positive predictive value was 100% [95% CI 85.8-100%] and the negative predictive value was 84% [63.1-94.7%]. Four cases were identified as ”non-invasive” by CALIPER and as ”invasive” by pulmonary pathology consensus. None of these patients developed or died of recurrent disease.
CONCLUSIONS: CALIPER represents a promising tool for non-invasive risk stratification of pulmonary nodules of the adenocarcinoma spectrum, particularly sub-solid density nodules. Validation of this pilot data is ongoing.
CLINICAL IMPLICATIONS: CALIPER could be very valuable in the management of incidentally or screen-detected pulmonary nodules.
DISCLOSURE: The following authors have nothing to disclose: Fabien Maldonado, Sushravya Raghunath, Marie Christine Aubry, Brian Bartholmai, Jennifer Bohland, Ronald Karwoski, Srinivasan Rajagopalan, Richard Robb, Ping Yang, Eunhee Yi, Tobias Peikert
CALIPER computer aided image analysis tool