Search

IntroductionSegmentation of lung structures in medical imaging is crucial for the application of automated post-processing steps on lung diseases like cystic fibrosis (CF). Recently, machine learning methods, particularly neural networks, have demonstrated remarkable improvements, often outperforming conventional segmentation methods. Nonetheless, challenges still remain when attempting to segment various imaging modalities and diseases, especially when the visual characteristics of pathologic findings significantly deviate from healthy tissue.MethodsOur study focuses on imaging of pediatric CF patients [mean age, standard deviation (7.50 ± 4.6)], utilizing deep learning-based methods for automated lung segmentation from chest magnetic resonance imaging (MRI). A total of 165 standardized annual surveillance MRI scans from 84 patients with CF were segmented using the nnU-Net framework. Patient cases represented a range of disease severities and ages. The nnU-Net was trained and evaluated on three MRI sequences (BLADE, VIBE, and HASTE), which are highly relevant for the evaluation of CF induced lung changes. We utilized 40 cases for training per sequence, and tested with 15 cases per sequence, using the Sørensen-Dice-Score, Pearson’s correlation coefficient (r), a segmentation questionnaire, and slice-based analysis.ResultsThe results demonstrated a high level of segmentation performance across all sequences, with only minor differences observed in the mean Dice coefficient: BLADE (0.96 ± 0.05), VIBE (0.96 ± 0.04), and HASTE (0.95 ± 0.05). Additionally, the segmentation quality was consistent across different disease severities, patient ages, and sizes. Manual evaluation identified specific challenges, such as incomplete segmentations near the diaphragm and dorsal regions. Validation on a separate, external dataset of nine toddlers (2–24 months) demonstrated generalizability of the trained model achieving a Dice coefficient of 0.85 ± 0.03.Discussion and conclusionOverall, our study demonstrates the feasibility and effectiveness of using nnU-Net for automated segmentation of lung halves in pediatric CF patients, showing promising directions for advanced image analysis techniques to assist in clinical decision-making and monitoring of CF lung disease progression. Despite these achievements, further improvements are needed to address specific segmentation challenges and enhance generalizability.

IntroductionDue to hypoxic vasoconstriction, perfusion is interesting in the lungs. Magnetic Resonance Imaging (MRI) perfusion imaging based on Dynamic Contrast Enhancement (DCE) has been demonstrated in patients with Chronic Obstructive Pulmonary Diseases (COPD) using visual scores, and quantification methods were recently developed further. Inter-patient correlations of echo time-dependent observed T1 [T1(TE)] have been shown with perfusion scores, pulmonary function testing, and quantitative computed tomography. Here, we examined T1(TE) quantification and quantitative perfusion MRI together and investigated both inter-patient and local correlations between T1(TE) and quantitative perfusion.Methods22 patients (age 68.0 ± 6.2) with COPD were examined using morphological MRI, inversion recovery multi-echo 2D ultra-short TE (UTE) in 1–2 slices for T1(TE) mapping, and 4D Time-resolved angiography With Stochastic Trajectories (TWIST) for DCE. T1(TE) maps were calculated from 2D UTE at five TEs from 70 to 2,300 μs. Pulmonary Blood Flow (PBF) and perfusion defect (QDP) maps were produced from DCE measurements. Lungs were automatically segmented on UTE images and morphological MRI and these segmentations registered to DCE images. DCE images were separately registered to UTE in corresponding slices and divided into corresponding subdivisions. Spearman’s correlation coefficients were calculated for inter-patient correlations using the entire segmented slices and for local correlations separately using registered images and subdivisions for each TE. Median T1(TE) in normal and defect areas according to QDP maps were compared.ResultsInter-patient correlations were strongest on average at TE2 = 500 μs, reaching up to |ρ| = 0.64 for T1 with PBF and |ρ| = 0.76 with QDP. Generally, local correlations of T1 with PBF were weaker at TE2 than at TE1 or TE3 and with maximum values of |ρ| = 0.66 (from registration) and |ρ| = 0.69 (from subdivision). In 18 patients, T1 was shorter in defect areas than in normal areas, with the relative difference smallest at TE2.DiscussionThe inter-patient correlations of T1 with PBF and QDP found show similar strength and TE-dependence as those previously reported for visual perfusion scores and quantitative computed tomography. The local correlations and median T1 suggest that not only base T1 but also the TE-dependence of observed T1 in normal areas is closer to that found previously in healthy volunteers than in defect areas.

BackgroundDynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) allows the assessment of pulmonary perfusion, which may play a key role in the development of muco-obstructive lung disease. One problem with quantifying pulmonary perfusion is the high variability of metrics. Quantifying the extent of abnormalities using unsupervised clustering algorithms in residue function maps leads to intrinsic normalization and could reduce variability.PurposeWe investigated the reproducibility of perfusion defects in percent (QDP) in clinically stable patients with cystic fibrosis (CF) and chronic obstructive pulmonary disease (COPD).Methods15 CF (29.3 ± 9.3y, FEV1%predicted = 66.6 ± 15.8%) and 20 COPD (66.5 ± 8.9y, FEV1%predicted = 42.0 ± 13.3%) patients underwent DCE-MRI twice 1 month apart. QDP, pulmonary blood flow (PBF), and pulmonary blood volume (PBV) were computed from residue function maps using an in-house quantification pipeline. A previously validated MRI perfusion score was visually assessed by an expert reader.ResultsOverall, mean QDP, PBF, and PBV did not change within 1 month, except for QDP in COPD (p < 0.05). We observed smaller limits of agreement (± 1.96 SD) related to the median for QDP (CF: ± 38%, COPD: ± 37%) compared to PBF (CF: ± 89%, COPD: ± 55%) and PBV (CF: ± 55%, COPD: ± 51%). QDP correlated moderately with the MRI perfusion score in CF (r = 0.46, p < 0.05) and COPD (r = 0.66, p < 0.001). PBF and PBV correlated poorly with the MRI perfusion score in CF (r =−0.29, p = 0.132 and r =−0.35, p = 0.067, respectively) and moderately in COPD (r =−0.57 and r =−0.57, p < 0.001, respectively).ConclusionIn patients with muco-obstructive lung diseases, QDP was more robust and showed a higher correlation with the MRI perfusion score compared to the traditionally used perfusion metrics PBF and PBV.