Manual
User Manual:
Open the PDF directly: View PDF 
.
Page Count: 5
Manual ESD toolbox
Simon Geirnaert, Tom Francart, Alexander Bertrand
June 17, 2019
CONTENTS 2 GETTING STARTED
Contents
1 Important information 2
2 Getting started 2
2.1 Installation.................................. 2
2.2 Contents.m.................................. 3
2.3 Quickstartguide............................... 3
2.3.1 Computation of the ESD performance metric . . . . . . . . . . . . 3
2.3.2 Designing an optimal Markov chain model for a neuro-steered hear-
ingprosthesis............................. 4
3 Function documentation 4
The ESD toolbox provides an implementation of the expected switching time (ESD) metric
to evaluate the performance of neural decoders for auditory attention detection in the
context of neuro-steered hearing prostheses as published in [1, 3]. The ESD metric is
an interpretable, single-number metric that combines both accuracy and decision time. It
allows easy comparison between neural decoders based on a relevant criterion, independent
of the evaluated window lengths.
1 Important information
The toolbox can be found online at Github. By downloading and installing this software
and the associated files on your computing system, you agree to use the software under
the terms and conditions as specified in the License agreement (which can be found in
the directory or online). By distributing the EST code freely online, we hope that it
can be used by other researchers as a new standard performance metric in the auditory
attention detection community, possibly with extensions to other brain-computer interface
applications. If you wish to use this toolbox, please cite [1, 3, 2] in any related or resulting
publications.
Note that this code (as it is a first version) is still prone to errors. If you find any bugs,
please report them to simon.geirnaert@esat.kuleuven.be. Furthermore, we welcome any
suggestion or improvement for the toolbox.
2 Getting started
2.1 Installation
Download the ESD toolbox, unzip it to any directory and add the directory to the MATLAB
path. This can be done by browsing to the chosen directory and running
a ddp at h ( pwd ) ; % Add the c u r r e n t d i r e c t o r y to the MATLAB path
s a v e p a t h ; % Save the path
2
2 GETTING STARTED 2.2 Contents.m
2.2 Contents.m
To show a complete listing of all the functions in the ESD toolbox, once it is installed to
the directory esd−toolbox , run:
doc esd−toolbox
The command displays the file Contents.m , which shows grouped functions together with
a concise description:
% ESD TOOLBOX
% V e r s i o n 1 . 0 , 06−03−2019
%
% MAIN FUNCTION
% computeESD −Compute th e e x p ec te d s w i t c h d u r a t i o n
based
% on ( tau , p )−points .
%
% OPTIMIZATION MARKOV CHAIN MODEL
% optimizeMarkovChain −Compute th e o p t i m a l AAD Markov c h a i n .
%
% BASIC FEATURES MARKOV CHAIN MODEL
% ssDistribution −Compute th e s teady −state distribution of
% t he AAD Markov c h a i n .
% stateToAmpl −Convert a s t a t e i n d e x t o a r e l a t i v e
% a m p l i f i c a t i o n l e v e l .
% l b C f d I n t −Compute t h e l o w e r bound o f t h e
% P0−c o n f i d e n c e i n t e r v a l o f t he AAD Markov
% c h a i n .
% e v S t a t e −Compute t h e e x p e c t e d v a l u e o f t h e AAD
% Markov c h a i n .
%
% FUNDAMENTAL METRICS
% meanHittingTime −Compute t h e mean h i t t i n g time from s t a t e i
% to k .
% emtt −Compute t h e e x p e c t e d Markov t r a n s i t t im e t o s t a t e
% k .
%
% CONSTRUCTION PERFORMANCE CURVE
% interpolatePerfCurve −I n t e r p o l a t e the per f o r m a n c e c u r v e t hrough
% e v a l u a t e d p er fo rm an ce p o i n t s .
2.3 Quick start guide
2.3.1 Computation of the ESD performance metric
The ESD is computed in four steps:
1. Construction of the p(τ)-performance curve by interpolating through the evaluated
(on real EEG and audio data) (τi, pi)-points (decision time,accuracy).
2. Optimization of the Markov chain in the number of states Nfor each sampled τon
the p(τ)-performance curve.
3
3 FUNCTION DOCUMENTATION
3. Computation of the expected Markov transit time T(p(τ), τ, ˆ
Nτ)per sampled τand
corresponding optimal number of states ˆ
Nτ.
4. The ESD is equal to the minimal expected Markov transit time over all evaluated
expected Markov transit times: ESD =min
τT(p(τ), τ, ˆ
Nτ).
These steps are implemented in the main-function computeESD.m . Given the evaluated
(τi, pi)-performance points (tau,p) , the ESD can be computed with:
esd = computeESD ( tau , p ) ;
The default hyperparameter values are P0= 0.8(confidence level), c= 0.65 (lower bound
confidence interval), Nmin = 5 (minimal number of states) and K= 1000 (number of
samples evaluated on the performance curve). These hyperparameters can be adapted via
extra arguments in the ESD-function.
2.3.2 Designing an optimal Markov chain model for a neuro-steered hearing
prosthesis
In Section II.D of [3], a methodology is proposed to design an optimal Markov chain model
for an adaptive gain control system in a neuro-steered hearing prosthesis. For a fixed
accuracy pand hyperparameters P0,cand Nmin, the optimal number of states can be
found with:
Nopt = optimizeMarkovChain( p , Nmin , P0 , c ) ;
The optimal model for a certain neural decoder (represented by evaluated (τi, pi)-points)
can be identified by extra outputs of the ESD-function:
[ esd , Nopt , tauOpt , pOpt ] = computeESD ( tau , p , ’ Nmin ’ , Nmin , ’ P0 ’ , P0 , ’ c ’ , c ) ;
3 Function documentation
The MATLAB files in the ESD toolbox contain a documentation consisting of:
•a short description
•a more extensive description with in- and output arguments
•a detailed description of the input arguments
•contact information.
In addition, a short demo can be found in esdDemo.m to show an example of how to
compute the ESD metric with the toolbox.
[more to be completed]
4
REFERENCES REFERENCES
References
[1] Simon Geirnaert, Tom Francart, and Alexander Bertrand. A New Metric to Evaluate
Auditory Attention Detection Performance Based on a Markov Chain. Internal Report,
March 2019.
[2] Simon Geirnaert, Tom Francart, and Alexander Bertrand. EST toolbox, March 2019.
Available online.
[3] Simon Geirnaert, Tom Francart, and Alexander Bertrand. Expected Switching Time: a
Interpretable Performance Metric to Evaluate Neural Decoders for Auditory Attention
Detection. Internal Report, March 2019.
5
