Automatic upper airway segmentation in static and dynamic MRI via anatomy-guided convolutional neural networks

PURPOSE

Upper airway segmentation on MR images is a prerequisite step for quantitatively studying the anatomical structure and function of the upper airway and surrounding tissues. However, the complex variability of intensity and shape of anatomical structures and different modes of image acquisition commonly used in this application makes automatic upper airway segmentation challenging. In this paper, we develop and test a comprehensive deep learning-based segmentation system for use on MR images to address this problem.

MATERIALS AND METHODS

In our study, both static and dynamic MRI data sets are utilized, including 58 axial static 3D MRI studies, 22 mid-retropalatal dynamic 2D MRI studies, 21 mid-retroglossal dynamic 2D MRI studies, 36 mid-sagittal dynamic 2D MRI studies, and 23 isotropic dynamic 3D MRI studies, involving a total of 160 subjects and over 20 000 MRI slices. Samples of static and 2D dynamic MRI data sets were randomly divided into training, validation, and test sets by an approximate ratio of 5:2:3. Considering that the variability of annotation data among 3D dynamic MRIs was greater than for other MRI data sets, we increased the ratio of training data for these data to improve the robustness of the model. We designed a unified framework consisting of the following procedures. For static MRI, a generalized region-of-interest (GROI) strategy is applied to localize the partitions of nasal cavity and other portions of upper airway in axial data sets as two separate subobjects. Subsequently, the two subobjects are segmented by two separate 2D U-Nets. The two segmentation results are combined as the whole upper airway structure. The GROI strategy is also applied to other MRI modes. To minimize false-positive and false-negative rates in the segmentation results, we employed a novel loss function based explicitly on these rates to train the segmentation networks. An inter-reader study is conducted to test the performance of our system in comparison to human variability in ground truth (GT) segmentation of these challenging structures.

RESULTS

The proposed approach yielded mean Dice coefficients of 0.84±0.03, 0.89±0.13, 0.84±0.07, and 0.86±0.05 for static 3D MRI, mid-retropalatal/mid-retroglossal 2D dynamic MRI, mid-sagittal 2D dynamic MRI, and isotropic dynamic 3D MRI, respectively. The quantitative results show excellent agreement with manual delineation results. The inter-reader study results demonstrate that the segmentation performance of our approach is statistically indistinguishable from manual segmentations considering the inter-reader variability in GT.

CONCLUSIONS

The proposed method can be utilized for routine upper airway segmentation from static and dynamic MR images with high accuracy and efficiency. The proposed approach has the potential to be employed in other dynamic MRI-related applications, such as lung or heart segmentation.

Effect of Nasal Valve Shape on Downstream Volume, Airflow, and Pressure Drop: Importance of the Nasal Valve Revisited

OBJECTIVES

The relative importance of the nasal valve relative to the remainder of the nasal airway remains unknown. The goal of this article was to objectively measure the shape of the nasal inlet and its effect on downstream airflow and nasal cavity volume using a physical model and a physiologic flow model.

METHODS

A patient who had isolated nasal valve surgery and had pre- and postoperative computed tomography scans available for analysis was studied. Nasal inlet shape measurements, computational fluid dynamics, and nasal volume analysis were performed using the computed tomography data. In addition, a physical model was used to determine the effect of nasal obstruction on downstream soft tissue.

RESULTS

The postoperative shape of the nasal inlet was improved in terms of length and degree of tortuosity. Whereas the operated-on region at the nasal inlet showed an only 25% increase in cross-sectional area postoperatively, downstream nonoperated sites in the nasal cavity revealed increases in area ranging from 33% to 51%. Computational fluid dynamics analysis showed that airway resistance decreased by 42%, and pressure drop was reduced by 43%. Intraluminal mucosal expansion was found with nasal obstruction in the physical model.

CONCLUSION

By decreasing the degree of bending and length at the nasal valve, inspiratory downstream nonoperated sites of the nasal cavity showed improvement in volume and airflow, suggesting that the nasal valve could play an oversized role in modulating the aerodynamics of the airway. This was confirmed with the physical model of nasal obstruction on downstream mucosa.

Biomimetic design and fabrication of porous chitosan–gelatin liver scaffolds with hierarchical channel network

The presence of a hierarchical channel network in tissue engineering scaffold is essential to construct metabolically demanding liver tissue with thick and complex structures. In this research, chitosan–gelatin (C/G) scaffolds with fine three-dimensional channels were fabricated using indirect solid freeform fabrication and freeze-drying techniques. Fabrication processes were studied to create predesigned hierarchical channel network inside C/G scaffolds and achieve desired porous structure. Static in-vitro cell culture test showed that HepG2 cells attached on both micro-pores and micro-channels in C/G scaffolds successfully. HepG2 proliferated at much higher rates on C/G scaffolds with channel network, compared with those without channels. This approach demonstrated a promising way to engineer liver scaffolds with hierarchical channel network, and may lead to the development of thick and complex liver tissue equivalent in the future.

Image-based three-dimensional finite element modeling approach for upper airway mechanics

The goal of the project is to investigate new computational tools for understanding normal upper airway mechanics and the pathophysiology of obstructive sleep apnea (OSA), a disorder characterized by repetitive pharyngeal collapse and occlusion during sleep. To relate the airway mechanical parameters to anatomy, and explore the roles of primary muscle properties, airway shape and structure, a new computational system for modeling of rat upper airway passive mechanics based on magnetic resonance imaging (MRI) was developed. The model is capable of computing stress and strain distributions under arbitrary loading, and dynamic animation of three-dimensional airway motion.

Identifying the bad guy in a lineup using confidence judgments under deadline pressure

Eyewitness-identification tests often culminate in witnesses not picking the culprit or identifying innocent suspects. We tested a radical alternative to the traditional lineup procedure used in such tests. Rather than making a positive identification, witnesses made confidence judgments under a short deadline about whether each lineup member was the culprit. We compared this deadline procedure with the traditional sequential-lineup procedure in three experiments with retention intervals ranging from 5 min to 1 week. A classification algorithm that identified confidence criteria that optimally discriminated accurate from inaccurate decisions revealed that decision accuracy was 24% to 66% higher under the deadline procedure than under the traditional procedure. Confidence profiles across lineup stimuli were more informative than were identification decisions about the likelihood that an individual witness recognized the culprit or correctly recognized that the culprit was not present. Large differences between the maximum and the next-highest confidence value signaled very high accuracy. Future support for this procedure across varied conditions would highlight a viable alternative to the problematic lineup procedures that have traditionally been used by law enforcement.

Biocompatibility and biodegradation studies of PCL/β-TCP bone tissue scaffold fabricated by structural porogen method

Three-dimensional printer (3DP) (Z-Corp) is a solid freeform fabrication system capable of generating sub-millimeter physical features required for tissue engineering scaffolds. By using plaster composite materials, 3DP can fabricate a universal porogen which can be injected with a wide range of high melting temperature biomaterials. Here we report results toward the manufacture of either pure polycaprolactone (PCL) or homogeneous composites of 90/10 or 80/20 (w/w) PCL/beta-tricalcium phosphate (β-TCP) by injection molding into plaster composite porogens fabricated by 3DP. The resolution of printed plaster porogens and produced scaffolds was studied by scanning electron microscopy. Cytotoxicity test on scaffold extracts and biocompatibility test on the scaffolds as a matrix supporting murine osteoblast (7F2) and endothelial hybridoma (EAhy 926) cells growth for up to 4 days showed that the porogens removal process had only negligible effects on cell proliferation. The biodegradation tests of pure PCL and PCL/β-TCP composites were performed in DMEM with 10 % (v/v) FBS for up to 6 weeks. The PCL/β-TCP composites show faster degradation rate than that of pure PCL due to the addition of β-TCP, and the strength of 80/20 PCL/β-TCP composite is still suitable for human cancellous bone healing support after 6 weeks degradation. Combining precisely controlled porogen fabrication structure, good biocompatibility, and suitable mechanical properties after biodegradation, PCL/β-TCP scaffolds fabricated by 3DP porogen method provide essential capability for bone tissue engineering.

Identifying the Bad Guy in a Lineup Using Confidence Judgments Under Deadline Pressure

Eyewitness-identification tests often culminate in witnesses not picking the culprit or identifying innocent suspects. We tested a radical alternative to the traditional lineup procedure used in such tests. Rather than making a positive identification, witnesses made confidence judgments under a short deadline about whether each lineup member was the culprit. We compared this deadline procedure with the traditional sequential-lineup procedure in three experiments with retention intervals ranging from 5 min to 1 week. A classification algorithm that identified confidence criteria that optimally discriminated accurate from inaccurate decisions revealed that decision accuracy was 24% to 66% higher under the deadline procedure than under the traditional procedure. Confidence profiles across lineup stimuli were more informative than were identification decisions about the likelihood that an individual witness recognized the culprit or correctly recognized that the culprit was not present. Large differences between the maximum and the next-highest confidence value signaled very high accuracy. Future support for this procedure across varied conditions would highlight a viable alternative to the problematic lineup procedures that have traditionally been used by law enforcement.

Developmental aspects of the upper airway: report from an NHLBI Workshop, March 5-6, 2009

The upper airway serves three important functions: respiration, swallowing, and speech. During development it undergoes significant structural and functional changes that affect its size, shape, and mechanical properties. Abnormalities of the upper airway require prompt attention, because these often alter ventilatory patterns and gas exchange, particularly during sleep when upper airway motor tone and ventilatory drive are diminished. Recognizing the relationship of early life events to lung health and disease, the National Heart, Lung, and Blood Institute (NHLBI), with cofunding from the Office of Rare Diseases (ORD), convened a workshop of extramural experts, from many disciplines. The objective of the workshop was: (1) to review the state of science in pediatric upper airway disorders; (2) to make recommendations to the Institute to fill knowledge gaps; (3) to prioritize new research directions; and (4) to capitalize on scientific opportunities. This report provides recommendations that could facilitate translation of basic research findings into practice to better diagnose, treat, and prevent airway compromise in children.

Endothelial-dependent vasodilation is associated with increases in the phosphorylation of a small heat shock protein (HSP20)

PURPOSE

Increases in the phosphorylation of a small heat shock protein (HSP20) are associated with cyclic nucleotide-dependent vasorelaxation. The effect of pressure and flow on vessel diameter was studied. We hypothesized that physiologic conditions that induce vasorelaxation would lead to increases in HSP20 phosphorylation.

METHODS

Flow-dependent changes in vessel diameter, at different intraluminal pressures, were measured with a laser optical micrometer in intact bovine carotid arteries. Experiments were performed in the presence and absence of norepinephrine (10(-5) mol/L). Increases in the phosphorylation of HSP20 were determined with isoelectric focusing immunoblots.

RESULTS

The increase in vessel diameter was most significant at low intraluminal pressures (20 mm Hg), high flow rates (200 mL/min), and in the presence of the vasoconstrictor norepinephrine (10(-5) mol/L). The addition of methylene blue (a guanylate cyclase inhibitor) completely inhibited flow-induced vasodilation. Under conditions in which maximal flow induced vasodilation occurred, there were significant increases in the phosphorylation of HSP20.

CONCLUSION

Flow-dependent vasodilation in isolated perfused segments of bovine carotid arteries was maximal when the intraluminal pressures were low and when the vessels were precontracted with norepinephrine. Flow-dependent vasodilation was inhibited by methylene blue and was associated with increases in the phosphorylation of HSP20, suggesting that the vasodilation was mediated by endothelial production of nitric oxide.