Ventilation Segmentation Accuracy Enhancement using Kirsch Operator in Pulmonary MRI

Hyperpolarized (HP) noble gas magnetic resonance imaging (MRI) and inert fluorinated gas MRI are the main pulmonary functional imaging modalities capable of regional ventilation assessment. To quantify pulmonary ventilation, K-means-based segmentation of ventilation images is usually performed. K-means clustering, however, performs poorly on images with relatively low SNR and cannot effectively distinguish background noise from low-ventilated areas. Here, we demonstrate a novel ventilation segmentation approach based on the subsequent implementation of lung edge detection using the Kirsch Compass operator followed by K-means segmentation. Our segmentation algorithm resulted in significantly higher accuracy of ventilation defect volume (VDV) assessment compared to a conventional ventilation segmentation algorithm. There were no significant changes in ventilation volume (VV) between our segmentation approach and conventional K-means segmentation. These indicate that implementation of digital lung edge detection prior to K-means segmentation substantially decreases segmentation sensitivity to background noise and allows a much more accurate assessment of low-ventilated regions of the lungs.