Back

Automated Tracheal Morphometrics Using Deep Learning: Toward Custom Tracheostomy Tubes

Ton, B.; Farooq, S.; Veenstra, J.

2025-11-30 radiology and imaging
10.1101/2025.11.27.25341143 medRxiv
Show abstract

PurposeTracheostomy is a critical, life-saving procedure, but complications often arise due to suboptimal tracheostomy tube fit. Customised tubes could mitigate these risks, yet their development hinges on precise tracheal morphometric data. This study aims to establish an automated workflow for extracting high-resolution tracheal measurements from CT scans to support the design of demographic-specific tracheostomy tubes. MethodsThis retrospective study utilised 100 anonymised CT scans, with 90 included after exclusions. An automated workflow was developed using 3D Slicer and a deep learning segmentation model (VISTA-3D) to extract tracheal measurements. Key parameters analysed included tracheal length, anteroposterior and transverse dimensions, and maximum inscribed circle diameter. ResultsAutomated measurements revealed the following average tracheal dimensions: O_LILength: 66.1 {+/-} 12.2 mm C_LIO_LIAnteroposterior dimension: 16.2 {+/-} 2.5 mm C_LIO_LITransverse dimension: 17.9 {+/-} 2.4 mm C_LIO_LIMaximum inscribed circle diameter: 14.1 {+/-} 1.9 mm C_LI Sex-specific differences were observed, with males exhibiting larger tracheal dimensions across all metrics. ConclusionThe results align with existing literature, validating the automated approach for efficient, objective, and high-resolution morphometric analysis. This method supports the development of customised tracheostomy tubes tailored to demographic-specific needs, ultimately improving patient wellbeing.

Published in BMJ Innovations · not in our set (fewer than 10 published preprints to learn from) · training set

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.