Task

To analyze and extract useful insights from a large clinical trial dataset of bowel sound recordings, and then to create an unsupervised machine learning model for the automatic clustering of gut sound snippets to facilitate the diagnosis of irritable bowel syndrome.

Challanges

• Complex and noisy audio data, with subject-to-subject variability.
• Bowel acoustics is a new field with little prior knowledge and limited academic research available.

How we helped

• Created and implemented a de-noising algorithm specific to microphone-rubbing noise.
• Developed a Python-based framework for the unsupervised clustering (data mining) of bowel sounds and for understanding hierarchical relationships within the data.

Results

Our contributions helped our client, NoisyGuts, de-noise and improve their understanding of clinical trial data.