A Robust Method for Estimating Regional Pulmonary Parameters in Presence of Noise

Abstract

The reliability of parameters estimation in imaging-based regional functional measurements can be improved in the presence of noise by utilizing principal component analysis-based clustering without sacrificing spatial resolution compared to Cartesian binning. Results suggest that this approach has a great potential for robust grouping of pixels in hyperpolarized (3)He MRI maps of lung oxygen tension.

Chronic Obstructive Pulmonary Disease: Quantification of Bronchodilator Effects by Using Hyperpolarized He MR Imaging

“…(33)(34)(35), to classify the 3 He MR imaging voxel intensity values into clusters ranging from 1 to 5, representing gradations of signal intensity from no signal (cluster 1) and hypointense signal (cluster 2) to hyperintense signal (cluster 5) and generating a 3 He voxel cluster map ( Fig 1b ). To obtain the external contour of the thoracic cavity to differentiate ventilation defects (cluster 1) from the edge of the lung, 1 H MR images were segmented by using a seeded region-growing algorithm ( 36 ) and were registered to the 3 He MR ventilation images as previously described ( 37 ).…”

Comparison of single breath hyperpolarized 129Xe MRI with dynamic 19F MRI in cystic fibrosis lung disease

“…The algorithmic VDP segmentations commonly used in 129 Xe studies, such as k‐means clustering, are SNR dependent, and therefore, it is difficult to use these segmentations to make comparisons across modalities with different SNRs 23 . In this study, the SNR was calculated as the ratio of the 90th percentile of the lung interior distribution to the noise signal SD (taken from an ROI outside the thorax).…”

“…Xe studies, such as k-means clustering, are SNR dependent, and therefore, it is difficult to use these segmentations to make comparisons across modalities with different SNRs. 23 In this study, the SNR was calculated as the ratio of the 90th percentile of the lung interior distribution to the noise signal SD (taken from an ROI outside the thorax). As expected, 19 F images had markedly lower SNR than 129 Xe images, and this difference between SNR values drove the decision to use a threshold-based VDP analysis for all image types.…”