Table 2 Summary of base blink features, background and reference features

From: Spontaneous eye blink-based machine learning for tracking clinical fluctuations in Parkinson’s disease

Category

Feature Name

Equation

Description

Base Blink Features #1

RATE

\(n\)

The total number of blinks observed within the time window.

ENERGY

\(E=\mathop{\sum }\limits_{i=0}^{n}\frac{1}{{D}_{i}^{2}}\)

The sum of the reciprocal of the squared blink duration for each blink within the time window.

Base Blink Features #2

INTERVAL

\(I={t}_{{onset\; i}}-{t}_{{onset\; i}+1}\)

The time elapsed between the onset of one blink and the onset of the subsequent blink.

DURATION

\(D={t}_{{offset}}-{t}_{{onset}}\)

The time elapsed between the onset and offset of the blink.

CONFIDENCE

\(C=1-\frac{1}{D}{\int}_{{t}_{onset}}^{{t}_{offset}}{C}_{{pupil}}\)

Blink confidence reflects the quality of individual blinks. It is defined as one minus the mean of pupil confidences between the onset and offset of each blink. See Fig. 4(B) for graphical explanation.

DEPTH

\(C/D\)

Blink depth is defined as blink confidence divided by blink duration.

Background Features

L-DOPA_TIME

\({T}_{{noisy}}=T-{T}_{L-{dopa}}+\epsilon\)

Elapsed time after L-dopa administration (min) + Gaussian white noise \(\epsilon \sim {\mathscr{N}}\left(0,{\sigma }^{2}\right),\sigma =30\).

AGE

\({{Age}}_{{noisy}}={Age}+\epsilon\)

Patient age + Gaussian white noise \(\epsilon \sim {\mathscr{N}}\left(0,{\sigma }^{2}\right),\sigma =3\).

Reference Feature

L-DOPA_CONC

\({C}_{L-{dopa}}\)

Linearly interpolated plasma L-dopa concentration (ng/mL) sampled every 15−30 min.