Automated identification of bolus types in modified barium swallow studies using deep learning: a preliminary study

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Swallowing Disorders

Background:

• Modified Barium Swallow (MBS) studies use videofluoroscopy to assess swallowing.
• Accurate classification of bolus consistency is crucial for interpreting swallowing dynamics and aspiration risks.
• Current manual classification is time-consuming and prone to variability.

Purpose of the Study:

• To develop and evaluate a deep learning-based approach for automated bolus type classification in MBS videos.
• To provide a standardized and efficient framework for automated swallowing assessment.
• To reduce reliance on manual labeling in MBS analysis.

Main Methods:

• A dataset of 2,752 bolus-level video segments from 277 MBS studies was used.
• Videos were temporally segmented into fixed-length frame sequences.
• An Inflated 3D convolutional neural network was employed for deep learning model development.

Main Results:

• The deep learning model achieved 96.13% overall accuracy in classifying bolus types.
• The macro F1-score for bolus type classification was 95.05%.
• The system demonstrated effective automated classification of liquid, pudding, and solid boluses.

Conclusions:

• The AI-based system effectively automates food bolus type classification in MBS videos.
• This represents a significant advancement toward fully automated MBS analysis for swallowing efficiency.
• The AI model promises to streamline clinical and research workflows by reducing manual labeling.