TF_peak_selection

TF_peak_selection(candidate_signals, candidate_times, varargin)

** DETERMINING TF PEAK TIMES FROM CANDIDATE PEAKS**

🔶

function used to calculate which prominence curve peaks are TF peaks. This function uses kmeans to create two separate clusters and then extracts TF peak times from one of the clusters that is labelled to be TF peaks based on higher peak prominence (in time) values. An optional rectification is provided by running a second round of kmeans that tries to significantly reduce false positive rates, with the cost of sacrificing some “true” TF peaks.

Usage: [ spindle_times ] = TF_peak_selection(candidate_signals, candidate_times, ‘<flag#1>’,<arg#1>…’<flag#n>’,<arg#n>); - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - ##### Declared Inputs: [[[NEEDS UPDATES!!!]]]

The following variables can be acquired from the function find_prominence_peaks.m:

• can_signals: This is the input signal(s) of candidate

  spindle peaks from which spindles are detected. It can either be a vector of peak prominence values with the same length as number of candidate spindles, or a matrix with dimensions [#row = number of candidate spindles] and [#column = different features about each candidate spindle].

• can_times: A matrix with two columns, specifying the

  start and end times of candidate spindle times. Output detected spindle times are subsets from this input can_times, so it should have the same length as the number of candidate spindles.

##### Optional Inputs:

• ‘detection_method’: Method used to detect sleep spindles,

  can be either ‘kmeans’ or ‘threshold’. default: ‘kmeans’

• ‘prominence_column’: A column number specifying which column

  of can_signals is prominence value. This is used to disambiguate which group clustered by kmeans is spindles. This column should normally be in logarithmic scale already. If not, log it before constructing can_signals. default: 1

• ‘bandwidth_column’: A column number for the bandwidth. Will

  perform an arbitrary cutoff below half of specified spectral resolution to be non-spindles. default: []

• ‘spectral_resol’: A value specifying spectral resolution,

  will be used to mark non-spindles if bandwidth_column is non-empty. Everything with bandwidth below half of this value will be non-spindles. default: 4

• ‘kmeans_column’: A parameter specifying the columns of

  can_signals to use in the kmeans algorithm. If this value is Inf and method is ‘kmeans’, then all columns are used. default: Inf

• ‘kmeans_class’: A parameter specifying the number of

  classes to detect in the kmeans algorithm. default: 2

• ‘threshold_percentile’: A parameter specifying percentile

  threshold used for cut-off detection of spindles from candidate spindles (range 0-1). default: 0.75

• ‘verbose’: Whether to print some verbose messages.

  default: true

##### Outputs:

• spindle_times: A two column vector containing

  [spindle_start, spindle_end] times in seconds.

• non_spindle_times: A two column vector containing

  [nonspindle_start, nonspindle_end] times in seconds.

• spindle_index: A logical index that is the same length

  as the number of candidate spindles (i.e., length of can_signals and can_times). 1 for detected spindles, 0 for detected non-spindles.

• group_idx: A double array specifying the group

  assignment of all candidate spindles by kmeans, will be redundant with spindle_index if ‘threshold’ is the method.

• spindle_group: An integer specifying which integer in

  the group_idx array is spindle. Will be 1 if ‘threshold’ is the method.

• group_order: An array encoding the group order from

  higher peak prominence values to lower.