University of Twente Student Theses


Objective assessment for tracheomalacia severity in neonates born with an esophageal atresia

Muijen, Nienke (2024) Objective assessment for tracheomalacia severity in neonates born with an esophageal atresia.

[img] PDF
Abstract:Goal: The goal of this thesis was to explore how the tracheomalacia severity in neonates born with an esophageal atresia could objectively be assessed, using bronchoscopy videos. Method: Three subgoals were established to achieve the goal. The first subgoal aimed to determine the appropriate region of interest and image representation, considering the ’red, green, blue’ (RGB), ’hue, saturation, value’ (HSV), and gray color-spaces. The second subgoal aimed to select the most suitable segmentation techniques, considering Otsu-thresholding, Multi-Otsu thresholding, K-means clustering, and Canny edge detection. The final subgoal aimed to develop an objective scoring system, using the obtained segmentations. Therefore, the area and anteroposterior : transverse (APT) diameter ratio was calculated for the end-inspiratory and -expiratory phase at the malacia and non-malacia part of the trachea. Five formulas were proposed to relate the APT-ratio outcomes from these frames to the clinically estimated tracheal lumen collapse (TLC). Linear regression was used to assess whether the calculated outcomes could predict the percentage of TLC. Results: The pixel values corresponding to the tracheal lumen exhibit a different range than the cartilage ring and posterior wall in the red-, value- and gray channel. Following dice similarity coefficient (DSC) values were obtained for tracheal lumen segmentation in these channels; Otsu-thresholding: DSCred=0.9567, DSCvalue=0.9571 & DSCgray=0.9584; Multi-Otsu thresholding: DSCred=0.9674, DSCvalue=0.9681 & DSCgray=0.9615; K-means clustering: DSCred=0.9650, DSCvalue=0.9638 & DSCgray=0.9667. Linear regression resulted in the following outcomes; Formula 1: R2=0.046, RMSE=22.12 & p=0.365; Formula 2: R2=0.064, RMSE=21.91 & p=0.283; Formula 3: R2=0.003, RMSE=22.61 & p=0.819; Formula 4: R2=0.007, RMSE=22.56 & p=0.718; Formula 5: R2=0.002, RMSE=22.62 & p=0.840. Conclusion: The tracheal lumen is the ROI that can be segmented from a bronchoscopy video. The highest DSC value (DSC = 0.9681) was obtained for three-threshold Multi-Otsu thresholding in the value channel. However, the examined techniques had all quite similar DSC values. It remains uncertain whether this particular combination is really the best, given the constraints of the limited data. The APT-ratio seems not to be a useful measurement because none of the proposed formula’s had a statistically significant linear regression analysis. Further research should focus on how the area could be used as measurement.
Item Type:Essay (Master)
Faculty:TNW: Science and Technology
Subject:44 medicine
Programme:Technical Medicine MSc (60033)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page