Pre-Implantation Peri-Valve Jacobian Texture Radiomics As a Biomarker of Endobronchial Valve Treatment Success
Abstract
Purpose
Endobronchial Valves (EBV) are one of the few treatment options for patients with moderate to severe emphysema. Eligibility is typically assessed from CT image data analysis including Emphysema Score (Relative Area, RA950) and Fissure Integrity Score (FIS). However not all EBV candidates are responders. The purpose of this work was to develop an approach to quantify CT texture of biomechanical lung expansion and contraction during lung inspiration and expiration using Jacobian-derived texture-based radiomics features, and to investigate the association of localized peri-valve biomechanical lung texture in pre-implantation scans in identifying candidates who will benefit from EBV treatment.
Methods
We performed a retrospective study using data from 20 patients with moderate to severe emphysema who underwent EBV treatment. Each patient underwent baseline (pre-treatment) CT scanning at both Total Lung Capacity (TLC) and Residual Volume (RV) breath-holds. Lung lobe segmentations were performed on CT image data to automatically identify lobar regions. RA910 and RA950 were extracted for each lobe. Using the deformation vector field from deformable registered in RV-TLC images, Jacobian determinant was calculated to quantify lung expansion and contraction and discretized for second-order texture-based radiomics. 72 features from 3 texture-radiomic feature sets were extracted. Peri‑valve regions of interest (ROIs) were generated using a novel valve detection method on follow-up CT and rigid mapping to baseline RV geometry. Univariate discrimination analyses were performed to assess associations between peri-valve radiomic features and EBV responder status.
Results
Univariate ROC analysis demonstrated modest discrimination for selected peri-valve Jacobian radiomic features, including GLRLM Short Run Low Gray Level Emphasis and Run Percentage (AUC = 0.63), suggesting localized biomechanical differences between responders and non-responders.
Conclusion
A fully automated pipeline for peri‑valve biomechanical texture analysis is feasible and shows proof‑of‑concept signal for EBV response stratification. Future work will improve valve localization specificity.