Reproducibility of a combined artificial intelligence and optimal-surface graph-cut method to automate bronchial parameter extraction

OBJECTIVES

Computed tomography (CT)-based bronchial parameters correlate with disease status. Segmentation and measurement of the bronchial lumen and walls usually require significant manpower. We evaluate the reproducibility of a deep learning and optimal-surface graph-cut method to automatically segment the airway lumen and wall, and calculate bronchial parameters.

METHODS

A deep-learning airway segmentation model was newly trained on 24 Imaging in Lifelines (ImaLife) low-dose chest CT scans. This model was combined with an optimal-surface graph-cut for airway wall segmentation. These tools were used to calculate bronchial parameters in CT scans of 188 ImaLife participants with two scans an average of 3 months apart. Bronchial parameters were compared for reproducibility assessment, assuming no change between scans.

RESULTS

Of 376 CT scans, 374 (99%) were successfully measured. Segmented airway trees contained a mean of 10 generations and 250 branches. The coefficient of determination (R2) for the luminal area (LA) ranged from 0.93 at the trachea to 0.68 at the 6th generation, decreasing to 0.51 at the 8th generation. Corresponding values for Wall Area Percentage (WAP) were 0.86, 0.67, and 0.42, respectively. Bland-Altman analysis of LA and WAP per generation demonstrated mean differences close to 0; limits of agreement (LoA) were narrow for WAP and Pi10 (± 3.7% of mean) and wider for LA (± 16.4-22.8% for 2-6th generations). From the 7th generation onwards, there was a sharp decrease in reproducibility and a widening LoA.

CONCLUSION

The outlined approach for automatic bronchial parameter measurement on low-dose chest CT scans is a reliable way to assess the airway tree down to the 6th generation.

STATEMENT ON CLINICAL RELEVANCE

This reliable and fully automatic pipeline for bronchial parameter measurement on low-dose CT scans has potential applications in screening for early disease and clinical tasks such as virtual bronchoscopy or surgical planning, while also enabling the exploration of bronchial parameters in large datasets.

KEY POINTS

• Deep learning combined with optimal-surface graph-cut provides accurate airway lumen and wall segmentations on low-dose CT scans. • Analysis of repeat scans showed that the automated tools had moderate-to-good reproducibility of bronchial measurements down to the 6th generation airway. • Automated measurement of bronchial parameters enables the assessment of large datasets with less man-hours.

Bronchial wall parameters on CT in healthy never-smoking, smoking, COPD, and asthma populations: a systematic review and meta-analysis

Objective

Research on computed tomography (CT) bronchial parameter measurements shows that there are conflicting results on the values for bronchial parameters in the never-smoking, smoking, asthma, and chronic obstructive pulmonary disease (COPD) populations. This review assesses the current CT methods for obtaining bronchial wall parameters and their comparison between populations.

Methods

A systematic review of MEDLINE and Embase was conducted following PRISMA guidelines (last search date 25th October 2021). Methodology data was collected and summarised. Values of percentage wall area (WA%), wall thickness (WT), summary airway measure (Pi10), and luminal area (Ai) were pooled and compared between populations.

Results

A total of 169 articles were included for methodologic review; 66 of these were included for meta-analysis. Most measurements were obtained from multiplanar reconstructions of segmented airways (93 of 169 articles), using various tools and algorithms; third generation airways in the upper and lower lobes were most frequently studied. COPD (12,746) and smoking (15,092) populations were largest across studies and mostly consisted of men (median 64.4%, IQR 61.5 – 66.1%). There were significant differences between populations; the largest WA% was found in COPD (mean SD 62.93 ± 7.41%, n = 6,045), and the asthma population had the largest Pi10 (4.03 ± 0.27 mm, n = 442). Ai normalised to body surface area (Ai/BSA) (12.46 ± 4 mm², n = 134) was largest in the never-smoking population.

Conclusions

Studies on CT-derived bronchial parameter measurements are heterogenous in methodology and population, resulting in challenges to compare outcomes between studies. Significant differences between populations exist for several parameters, most notably in the wall area percentage; however, there is a large overlap in their ranges.

Key Points

• Diverse methodology in measuring airways contributes to overlap in ranges of bronchial parameters among the never-smoking, smoking, COPD, and asthma populations.

• The combined number of never-smoking participants in studies is low, limiting insight into this population and the impact of participant characteristics on bronchial parameters.

• Wall area percent of the right upper lobe apical segment is the most studied (87 articles) and differentiates all except smoking vs asthma populations.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00330-022-08600-1.

Creating a training set for artificial intelligence from initial segmentations of airways

Airways segmentation is important for research about pulmonary disease but require a large amount of time by trained specialists. We used an openly available software to improve airways segmentations obtained from an artificial intelligence (AI) tool and retrained the tool to get a better performance. Fifteen initial airway segmentations from low-dose chest computed tomography scans were obtained with a 3D-Unet AI tool previously trained on Danish Lung Cancer Screening Trial and Erasmus-MC Sophia datasets. Segmentations were manually corrected in 3D Slicer. The corrected airway segmentations were used to retrain the 3D-Unet. Airway measurements were automatically obtained and included count, airway length and luminal diameter per generation from the segmentations. Correcting segmentations required 2–4 h per scan. Manually corrected segmentations had more branches (p < 0.001), longer airways (p < 0.001) and smaller luminal diameters (p = 0.004) than initial segmentations. Segmentations from retrained 3D-Unets trended towards more branches and longer airways compared to the initial segmentations. The largest changes were seen in airways from 6th generation onwards. Manual correction results in significantly improved segmentations and is potentially a useful and time-efficient method to improve the AI tool performance on a specific hospital or research dataset.

Anisotropic Shape-Memory Alginate Scaffolds Functionalized with Either Type I or Type II Collagen for Cartilage Tissue Engineering

Regenerating articular cartilage and fibrocartilaginous tissue such as the meniscus is still a challenge in orthopedic medicine. While a range of different scaffolds have been developed for joint repair, none have facilitated the development of a tissue that mimics the complexity of soft tissues such as articular cartilage. Furthermore, many of these scaffolds are not designed to function in mechanically challenging joint environments. The overall goal of this study was to develop a porous, biomimetic, shape-memory alginate scaffold for directing cartilage regeneration. To this end, a scaffold was designed with architectural cues to guide cellular and neo-tissue alignment, which was additionally functionalized with a range of extracellular matrix cues to direct stem cell differentiation toward the chondrogenic lineage. Shape-memory properties were introduced by covalent cross-linking alginate using carbodiimide chemistry, while the architecture of the scaffold was modified using a directional freezing technique. Introducing such an aligned pore structure was found to improve the mechanical properties of the scaffold, and promoted higher levels of sulfated glycosaminoglycans (sGAG) and collagen deposition compared to an isotropic (nonaligned) pore geometry when seeded with adult human stem cells. Functionalization with collagen improved stem cell recruitment into the scaffold and facilitated more homogenous cartilage tissue deposition throughout the construct. Incorporating type II collagen into the scaffolds led to greater cell proliferation, higher sGAG and collagen accumulation, and the development of a stiffer tissue compared to scaffolds functionalized with type I collagen. The results of this study demonstrate how both scaffold architecture and composition can be tailored in a shape-memory alginate scaffold to direct stem cell differentiation and support the development of complex cartilaginous tissues.