An increased interest in longitudinal neurodevelopment during the first few years after birth has emerged in recent years. Noninvasive magnetic resonance imaging (MRI) can provide crucial information about the development of brain structures in the early months of life. Despite the success of MRI collections and analysis for adults, it remains a challenge for researchers to collect high-quality multimodal MRIs from developing infant brains because of their irregular sleep pattern, limited attention, inability to follow instructions to stay still during scanning. In addition, there are limited analytic approaches available. These challenges often lead to a significant reduction of usable MRI scans and pose a problem for modeling neurodevelopmental trajectories. Researchers have explored solving this problem by synthesizing realistic MRIs to replace corrupted ones. Among synthesis methods, the convolutional neural network-based (CNN-based) generative adversarial networks (GANs) have demonstrated promising performance. In this study, we introduced a novel 3D MRI synthesis framework– pyramid transformer network (PTNet3D)– which relies on attention mechanisms through transformer and performer layers. We conducted extensive experiments on high-resolution Developing Human Connectome Project (dHCP) and longitudinal Baby Connectome Project (BCP) datasets. Compared with CNN-based GANs, PTNet3D consistently shows superior synthesis accuracy and superior generalization on two independent, large-scale infant brain MRI datasets. Notably, we demonstrate that PTNet3D synthesized more realistic scans than CNN-based models when the input is from multi-age subjects. Potential applications of PTNet3D include synthesizing corrupted or missing images. By replacing corrupted scans with synthesized ones, we observed significant improvement in infant whole brain segmentation.
1 February 2025 in Radiology: Cardiothoracic Imaging
Metronome-paced tachypnea during real-time cine pulmonary MRI is a newly proposed technique to measure dynamic hyperinflation in chronic obstructive pulmonary disease, which yielded reproducible results.
Chronic obstructive pulmonary disease (COPD) and emphysema are characterized by functional and structural damage which increases the spaces for gaseous diffusion and impairs oxygen exchange. Here we explore the potential for hyperpolarized (HP) 3He MRI to characterize lung structure and function in a large-scale population-based study. Participants (n = 54) from the Multi-Ethnic Study of Atherosclerosis (MESA) COPD Study, a nested case-control study of COPD among participants with 10+ packyears underwent HP 3He MRI measuring pAO2, apparent diffusion coefficient (ADC), and ventilation. HP MRI measures were compared to full-lung CT and pulmonary function testing. High ADC values (>0.4 cm2/s) correlated with emphysema and heterogeneity in pAO2 measurements. Strong correlations were found between the heterogeneity of global pAO2 as summarized by its standard deviation (SD) (p < 0.0002) and non-physiologic pAO2 values (p < 0.0001) with percent emphysema on CT. A regional study revealed a strong association between pAO2 SD and visual emphysema severity (p < 0.003) and an association with the paraseptal emphysema subtype (p < 0.04) after adjustment for demographics and smoking status. HP noble gas pAO2 heterogeneity and the fraction of non-physiological pAO2 results increase in mild to moderate COPD. Measurements of pAO2 are sensitive to regional emphysematous damage detected by CT and may be used to probe pulmonary emphysema subtypes. HP noble gas lung MRI provides non-invasive information about COPD severity and lung function without ionizing radiation.
Purpose: Hyperpolarized helium (3He) MRI is used to assess ventilation defects in COPD and other pulmonary diseases. Existing segmentation methods are very operator-dependent and effort consuming. We aimed to develop a deep-learning based method to accelerate segmentation, eliminate operator dependence and improve reliability. Methods and Materials: The MESA COPD Study is a nested case-control study of COPD (post-bronchodilator FEV1/FVC ratio<0.70) among 10+ pack-year smokers. 56 participants (74±8 yrs) underwent same breath-hold 1H and 3He MRI. Using semi-automatic region growing techniques, labels of total lung were created on 543 1H slices and labels of ventilation defects were created on 543 3He slices. 42 randomly selected subjects were used for training, with the remaining 14 subjects used for validation. Two separate convolutional neural networks were trained to segment 1H MRI and 3He MRI with conventional data augmentation. Total lung masks via 1H MRI were imported into a second neural network to define lung boundaries on 3He MRI. Results: The percentage of non-, low-, and normal- ventilated regions were 12±9%, 25±14%, and 62±20%, respectively. At patient level, the dice coefficient scores were not significantly different between participants with and without COPD (p=0.29-0.95, Table). Conclusion: The proposed deep learning method yields accurate, automated segmentation of ventilation defects among older smokers, independent of COPD status. Funding NIH/NHLBI R01-HL093081, R01-HL077612, R01-HL121270