Use of Transfer Learning for the Automated Segmentation and Detection of Swallows via Digital Cervical Auscultation in Children.
Stephen So, Timothy Tadj, Belinda Schwerin and 2 others
PMID 40459758WHAT IT FOUND
A new machine learning tool automatically found 81% of swallows in children's cervical auscultation recordings.
This removes the need for time-consuming manual audio sorting, but the tool detects swallows only, not aspiration itself.
Key findings
01The model achieved 91% total accuracy, correctly identifying 81% of true swallows (sensitivity) and 94% of non-swallows.
02The system was trained on VFSS data and tested on mealtime observations, showing it can handle real-world clinical recording conditions.
03This is the first study to use transfer learning to segment swallowing sounds in audio recordings.
STILL TO COME
How it was doneWhat they foundWhat it means for SLPs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
The model detects swallows, not aspiration. It cannot tell if a child is aspirating. Training and testing data came from different groups (VFSS vs. mealtime), which is realistic but means the model has not been validated on the same type of data it was trained on. The study used a small number of participants (35 children). The model had false positives and false negatives, and no post-processing was used to clean the output.
Declared interests
The University of Queensland funded the work. No other conflicts of interest were declared.
The easy way to misread this
Do not interpret this as a validated tool for detecting aspiration. The model only segments swallow sounds from audio; it does not classify whether those swallows are safe or if aspiration is occurring.
Summarised by AI from the full paper, without a clinician reviewing it. Check it against the source before it changes what you do. Read it on PubMed →