Using Deep Learning to Improve Diagnosis of Sleep Apnea with Electromyography
Author Information
Author(s): Mandeville Ross, Sedghamiz Hooman, Mansfield Perry, Sheean Geoffrey, Studer Chris, Cordice Derrick, Ghanbari Ghodsieh, Malhotra Atul, Nemati Shamim, Koola Jejo
Primary Institution: Powell Mansfield, Inc.
Hypothesis
Transformers can effectively model transmembranous electromyography (tmEMG) data to distinguish between control, neurogenic, and sleep apnea patients.
Conclusion
The study suggests that deep learning can enhance the diagnostic ability of tmEMG for identifying obstructive sleep apnea and its endotypes.
Supporting Evidence
- The model achieved 98% sensitivity and 73% specificity for classifying neurogenic cases from controls.
- The model achieved 88% sensitivity and 64% specificity for classifying OSA from controls.
- By averaging predicted probabilities, the model correctly classified up to 82% of control and OSA patients.
Takeaway
Researchers used a special computer model to help doctors better understand signals from muscles in the throat, which can help find out if someone has sleep apnea.
Methodology
The study used a deep learning model with transformers to analyze 177 EMG recordings from patients and controls, employing techniques like transfer learning and data augmentation.
Potential Biases
Potential bias due to the limited diversity in the patient sample.
Limitations
The small sample size limits the generalizability of the findings.
Participant Demographics
The study included six healthy controls, five moderate to severe OSA patients, and five ALS patients.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website