Comparing Mean-anchored Generalized Linear Binning with Established Methods to Quantify Xenon-129 Ventilation Defect...

Hyperpolarized xenon-129 (¹²⁹Xe) MRI is an established, sensitive method to assess ventilation but lacks a standardized ventilation defect quantification. We introduce mean-anchored generalized linear binning (Mean_GLB, GLB) to calculate ventilation defect percentage (VDP), benchmarking its performance against reader-segmented VDP (Reader) and five widely-used automated VDP calculation methods.

¹²⁹Xe images were acquired from 38 people with cystic fibrosis (CF) and 25 age-matched healthy controls (HCs) using 2D-spiral sequence. Flexible chest coil-induced inhomogeneities were mitigated using both N4- or a flip-angle (FA)-based bias correction. VDP was quantified using Reader, Hierarchical and Adaptive K-means clustering, mean-normalized linear binning, 99th Percentile-normalized GLB, binary Thresholding, and Mean_GLB; and compared for all techniques. Additionally, VDPs obtained via automated methods were evaluated against Reader. Voxel-wise agreement with Reader-based defect maps was assessed using sensitivity, specificity and Youden's index.

Using N4-corrected images, all methods-except Hierarchical-showed significant difference in VDP between participants with CF and HCs (P<0.05). For FA-corrected images, similar significance was observed for all methods (P<0.05), except Hierarchical and Adaptive. While Mean_GLB overestimated defects in participants with CF, for both N4- and FA-corrected images (Bland-Altman biases: -3.1% and -1.9%, respectively), it strongly correlated with Reader (r=0.97, P<0.001 both). Mean_GLB achieved the highest Youden's index (∼0.75) across N4- and FA-corrected images, indicating balance of sensitivity and specificity.

Mean_GLB enabled robust quantification of VDP in ¹²⁹Xe ventilation MRI, independent of bias-correction algorithm. Its ability to capture subtle impairments indicates its potential for applications in pulmonary imaging.