public ARArtifactDetector(ARModel model)
{
if (model == null)
{
throw new ArgumentException();
}
this.model = model;
errorDist = new OnlineVariance();
confusionDist = new OnlineVariance();
}
protected override bool Process(object item)
{
EEGEvent evt = (EEGEvent)item;
var n = evt.data.Length;
double[] buffer = new double[evt.data.Length];
for (int i = 0; i < n; i++)
{
buffer[i] = signalFilters[i].Filter(evt.data[i]);
}
// TODO test artifact detection
if (isLearning)
{
// add samples to window and test if window is sufficiently large
bool stillLearning = false;
for (int i = 0; i < n; i++)
{
artifactLearningSamples[i].Enqueue(evt.data[i]);
stillLearning = artifactLearningSamples[i].Count < artifactLearningSize;
}
// sufficient samples to fit AR model
if (!stillLearning)
{
for (int i = 0; i < n; i++)
{
double[] x = artifactLearningSamples[i].ToArray();
var p = StatsUtils.EstimateAROrder(x, 50);
double mean = StatsUtils.SampleMean(x);
double[] arParams = StatsUtils.FitAR(p, x);
Logger.Log("Estimated AR params: p={0}, c={1}, phi={2}", p, mean, string.Join(", ", arParams));
arPredictors[i] = new ARModel(mean, arParams);
lastPredictions[i] = arPredictors[i].Predict(evt.data[i]);
}
}
isLearning = stillLearning;
}
else
{
bool isArtifact = false;
for (int i = 0; i < n; i++)
{
var x = evt.data[i];
var error = x - lastPredictions[i];
var errorDist = arError[i];
errorDist.Update(error);
lastPredictions[i] = arPredictors[i].Predict(x);
if (errorDist.isValid)
{
var errorS = Math.Sqrt(errorDist.var);
var errorMaxCI95 = errorDist.mean + 2 * errorS;
var errorMinCI95 = errorDist.mean - 2 * errorS;
//Logger.Log("Error={0}, 95% CI = [{1}, {2}]", error, errorMinCI95, errorMaxCI95);
isArtifact = error > errorMaxCI95 || error < errorMinCI95;
}
}
if (isArtifact)
{
Logger.Log("Detected large amplitude artifact via AR model");
return(true);
}
}
Add(new EEGEvent(evt.timestamp, evt.type, buffer, evt.extra));
return(true);
}