Deformable Image Registration of Dark-Field Chest Radiographs for Local Lung Signal Change Assessment

TL;DR

This paper addresses how to quantitatively compare lung signals in dark-field chest radiographs across inspiratory and expiratory states. It introduces an iterative, deformable registration framework tailored for dark-field images using a multiresolution SVFFD model with LNCC similarity and bending-energy regularization, and evaluates it against a baseline attenuation-registration approach on a COPD-focused dataset. The results show strong registration performance and reveal that registered dark-field signal changes correlate more strongly with breathing capacity and emphysema severity than attenuation-based metrics, suggesting potential for low-dose dynamic lung-function assessment using dark-field radiography. Overall, the work provides a practical method and evidence that dark-field imaging can add functional insight beyond conventional radiography, with implications for rapid, low-dose ventilation analysis.

Abstract

Dark-field radiography of the human chest has been demonstrated to have promising potential for the analysis of the lung microstructure and the diagnosis of respiratory diseases. However, previous studies of dark-field chest radiographs evaluated the lung signal only in the inspiratory breathing state. Our work aims to add a new perspective to these previous assessments by locally comparing dark-field lung information between different respiratory states. To this end, we discuss suitable image registration methods for dark-field chest radiographs to enable consistent spatial alignment of the lung in distinct breathing states. Utilizing full inspiration and expiration scans from a clinical chronic obstructive pulmonary disease study, we assess the performance of the proposed registration framework and outline applicable evaluation approaches. Our regional characterization of lung dark-field signal changes between the breathing states provides a proof-of-principle that dynamic radiography-based lung function assessment approaches may benefit from considering registered dark-field images in addition to standard plain chest radiographs.

Figures (6)

• Figure 1: Dark-field-compatible deformable image registration methods enable matching chest radiographs in full expiration and full inspiration. The ratio of the transformed expiration and inspiration dark-field images is used to analyze local lung signal changes.
• Figure 2: a) Schematic view of the dark-field chest X-ray prototype system with the main components. The dark-field lung image formation in the expiratory and inspiratory breath-hold states is indicated in green and blue. b) Examples of the corresponding attenuation and dark-field images with an overlay of the manually drawn full (solid line) and partial (dotted line) lung masks and the annotated landmarks (blue and green dots).
• Figure 3: Overview of the dark-field chest radiograph registration framework (top) and the accuracy evaluation approaches (bottom). The expiration dark-field image is registered to the inspiration dark-field radiograph utilizing a multiresolution strategy and an SVFFD deformation model. The LNCC image similarity between the inspiration and transformed expiration image as well as the bending energy regularization guide the transformation optimization. Manually annotated full and partial lung masks as well as landmarks are utilized for the registration accuracy evaluation. The registration transformation is applied to the respective masks and landmarks. The Dice score, the mean surface distance, the Hausdorff distance of the masks, and the target registration error of the landmarks are calculated to assess the registration accuracy.
• Figure 4: Registration evaluation metrics considering full lung mask metrics and landmark distances (TRE) for the dark-field image registration framework within the dataset. The top row shows the scores before, and the bottom row the values after the registration procedure. Red lines indicate the median, black boxes the interquartile range (25th to 75th percentile), and blue dots the individual study participant scores.
• Figure 5: Attenuation and dark-field image registration results for two study participants with signal ratio analysis. a) Fleischner scale score 0, $VLC_{\mathrm{rel}}=0.28$. b) Fleischner scale score 2, $VLC_{\mathrm{rel}}=0.14$. The linear fit slopes to the CC signal ratio projection differ substantially for the two study participant cases in the dark-field domain (0.029 right and left for case a), 0.008 right and 0.009 left for case b)) in contrast to the attenuation domain (0.001 right and left for case a), 0.003 right and 0.004 left for case b)). Black regions within the ratio images indicate pixels where the transformed exp. or insp. signal was below $10^{-15}$ and thus set to NaN to avoid division by zero errors.