Sleep Apnea Detection

Model Overview

The following table provides the latest pre-trained models for sleep apnea. Below we also provide additional details including training configuration, accuracy metrics, and hardware performance results for the models.

NAME	LOCATION	# CLASSES	MODEL	PARAMS	FLOPS	ACCURACY	F1
SA-2-TCN-SM	Wrist	2	TCN	5K	20M/hr	91.0%	91.0%

Model Details

SA-2-TCN-SM

The SA-2-TCN-SM model is a 2-stage sleep apnea detection model that uses a Temporal convolutional network (TCN). The model is trained on PPG sensor data collected from the wrist and is able to locate apnea/hypopnea events.

Location: Wrist
Classes: Normal, Apnea
Frame Size: 10 minutes
Datasets: MESA
Features: FS-W-P-5

Model Performance

SA-2-TCN-SM

The following plots show the model's performance in detecting apnea/hypopnea events. The first plot shows the confusion matrix for apnea detection.

The following plot shows the model's ability to detect AHI (Apnea-Hypopnea Index) compared to the ground truth AHI values. The x-axis represents the true AHI values, while the y-axis represents the predicted AHI values. The plot shows a strong correlation between the true and predicted AHI values.

The following table provides the corresponding confusion matrix for AHI.

EVB Performance

The following table provides the latest performance results when running on Apollo4 Plus EVB. These results are obtained using neuralSPOTs Autodeploy tool. From neuralSPOT repo, the following command can be used to capture EVB results via Autodeploy:

python -m ns_autodeploy \
--tflite-filename model.tflite \
--model-name model \
--cpu-mode 192 \
--arena-size-scratch-buffer-padding 0 \
--max-arena-size 60 \

Name	Params	FLOPS	Metric	Time	Arena	Energy
SA-2-TCN-SM	10K	44M/hr	91.0% F1	1.8s/hr	61KB	17.0mJ/hr

Downloads

SA-2-TCN-SM

Asset	Description
configuration.json	Configuration file
model.keras	Keras Model file
model.tflite	TFLite Model file
metrics.json	Metrics file