Skin layer classification by feedforward neural network in bioelectrical impedance spectroscopy

Kiagus Aufa Ibrahim; Marlin Ramadhan Baidillah; Ridwan Wicaksono; Masahiro Takei

doi:10.2478/joeb-2023-0004

.blurhash-client-img { display: none !important; }

Skin layer classification by feedforward neural network in bioelectrical impedance spectroscopy

Journal of Electrical Bioimpedance

Volume 14 (2023): Issue 1 (January 2023)

By: Kiagus Aufa Ibrahim, Marlin Ramadhan Baidillah, Ridwan Wicaksono and Masahiro Takei

Open Access

|Aug 2023

Figures & Tables

Schematic flow of skin layer classification of conductivity change

Table 1:

Parameters for training data set

Configuration parameter

Measurement parameter

Injection pattern φ_s

Electrode-length-to-skin-thickness ratio δ_b

Conductivity-changed layer ψ_p

Conductivity change Δσ_q [%]

Frequency pair

f_{r}^{l h}

[kHz]

I. Bipolar

II. Tetrapolar

D/H < 1, D/H = 1, D/H > 1

Fixed electrode gap:

d_g = 1 [mm]

Variable electrode diameter:

d_e = {1, 2, 3}[mm]

Electrode length:

D = d_g + d_e

Fixed each layer thickness:

h^S = 50 [μm], h^E = 0.45 [mm],

h^D = 2.5 [mm], h^F = 5 [mm]

Fixed skin thickness:

H = h^S + h^E + h^D + h^F

ψ₁ = Stratum Corneum only (S)

ψ₂ = Epidermis (E) only

ψ₃ = Dermis (D) only

ψ₄ = Fat (F) only

ψ₅ = S + E

ψ₆ = E + D

ψ₇ = D + F

ψ₈ = S + E + D

ψ₉ = E + D + F

ψ₁₀ = S + E + D + F

Δσ₁ = −20

Δσ₂ = −17.5

Δσ₃ = −15

Δσ₄ = −12.5

Δσ₅ = −10

Δσ₆ = −7.5

Δσ₇ = −5

Δσ₈ = −2.5

Δσ₉ = 0

Δσ₁₀ = 2.5

Δσ₁₁ = 5

Δσ₁₂ = 7.5

Δσ₁₃ = 10

Δσ₁₄ = 12.5

Δσ₁₅ = 15

Δσ₁₆ = 17.5

Δσ₁₇ = 20

\begin{array}{l} f_{1}^{l h} = 2 & 10 \\ f_{2}^{l h} = 2 & 35 \\ f_{3}^{l h} = 2 & 100 \\ f_{4}^{l h} = 2 & 225 \\ f_{5}^{l h} = 10 & 35 \\ f_{6}^{l h} = 10 & 100 \\ f_{7}^{l h} = 10 & 225 \\ f_{8}^{l h} = 35 & 100 \\ f_{9}^{l h} = 35 & 225 \\ f_{10}^{l h} = 100 & 225 \end{array}

Δσ₉ = 0 [%] is considered as standard.
For each conductivity-changed layer, the other layer(s) are fixed with standard conductivity change value.
Case number follow this rule: $φ_{s_} δ_{b_} ψ_{p_} Δ σ_{q_} f_{r}^{l h}$
(for example: I_D/H < 1_S_-20_2&10 which refers to injection pattern: φ_I = Bipolar_δ₁ = D/H < 1, ψ₁ = S, Δσ₁ = −20, and $f_{3}^{l h} = 2 & 100$ )
Total case: $2 (φ_{s}) \times 3 (δ_{b}) \times 10 (ψ_{p}) \times 17 (Δ σ_{q}) \times 10 (f_{r}^{l h}) = 10200$

Electrical properties of skin, fat, and muscle.

Nyquist plot based on numerical simulation of *φ_s* in the Δσ and *D/H* < 1 in all frequency range.

DRT results based on simulation of *φ_s* in the case of *D/H* < 1 in all frequency range.

Table 2:

Matrix profile of α_ξ

Matrix Profile	Number of input features α_ξ	Details
One matrix	4	$\overset{´}{Z}, \overset{´}{θ}, \overset{´}{R}, \overset{´}{X}$
Two matrices	6	$\overset{´}{Z} + \overset{´}{θ}, \overset{´}{Z} + \overset{´}{R}, \overset{´}{Z} + \overset{´}{X}, \overset{´}{θ} + \overset{´}{R}, \overset{´}{θ} + \overset{´}{X}, \overset{´}{R} + \overset{´}{X}$
Three matrices	3	$\overset{´}{Z} + \overset{´}{θ} + \overset{´}{R}, \overset{´}{Z} + \overset{´}{θ} + \overset{´}{X}, \overset{´}{θ} + \overset{´}{R} + \overset{´}{X}$
Four matrices	1	$\overset{´}{Z} + \overset{´}{θ} + \overset{´}{R} + \overset{´}{X}$

The comparison of FNN results related to feature extraction of *α_ξ* effect on bipolar and tetrapolar.

The comparison of FNN results related to feature extraction of *α_ξ* effect on bipolar and tetrapolar using four matrices profile.

Experimental setup conditions with porcine skin.

Table 3:

Experimental conditions

Configuration parameter

Measurement parameter

Injection pattern φ_s

Electrode length ratio to skin thicknessδ_b

Conductivity-changed layery_p

ConcentrationC_Nacl [mM]

Frequency pair

f_{r}^{l h}

[kHz]

I. Bipolar

II. Tetrapolar

D/H < 1, D/H = 1, D/H > 1

Dermis (D) only

c₁ = 15

c₂ = 20;

c₃ = 25;

c₄ = 30;

c₅ = 35;

\begin{array}{l} f_{1}^{l h} = 2 & 10; \\ f_{2}^{l h} = 2 & 35; \\ f_{3}^{l h} = 2 & 100; \\ f_{4}^{l h} = 2 & 225; \\ f_{5}^{l h} = 10 & 35; \\ f_{6}^{l h} = 10 & 100; \\ f_{7}^{l h} = 10 & 225; \\ f_{8}^{l h} = 35 & 100; \\ f_{9}^{l h} = 35 & 225; \\ f_{10}^{l h} = 100 & 225 \end{array}

DRT results based on experiment with *φ_s* = Bipolar in all frequencies.

DRT results based on experiment with *φ_s* = Tetrapolar in all frequencies.

The confusion matrix shows the highest validation accuracy of experiments of porcine skin with FNN and four impedance inputs *α_ξ*.

Table 4:

Comparison of FNN accuracy Acc in terms of variation of frequency pair selection from experiment results.

Frequency pair $f_{r}^{l h}$ [kHz]	Bipolar Acc [%]	Tetrapolar Acc [%]
$f_{1}^{l h} = 2 & 10$	44.1	1.2
$f_{2}^{l h} = 2 & 35$	80.6	17.6
$f_{3}^{l h} = 2 & 100$	56.5	16.5
$f_{4}^{l h} = 2 & 225$	40.6	10.0
$f_{5}^{l h} = 10 & 35$	88.8	90.0
$f_{6}^{l h} = 10 & 100$	90.6	84.1
$f_{7}^{l h} = 10 & 225$	61.2	32.4
$f_{8}^{l h} = 35 & 100$	60.6	90.6
$f_{9}^{l h} = 35 & 225$	55.3	25.3
$f_{10}^{l h} = 100 & 225$	68.2	23.5

Comparison of sensitivity map distribution: bipolar and tetrapolar.

Table 5:

Comparison studies of skin condition detection using only BIS or combined with machine learning algorithm.

No	Author	Skin Condition	Machine Learning Algorithm & Data Features	Frequency Pair Selection & Reason	Injection Patterns
1	Stig Ollmar [18]	Oral mucosa	NA & Ź, $\overset{´}{Θ}$ , Ŕ, and $X^{'}$	20 [kHz] & 50 [kHz]: NA	Bipolar
2	Nicander et al. [7]	Irritant dermatitis	NA & Ź	20 [kHz] & 1 [MHz]: NA	Bipolar
3	Ramos & Bertemes-Filho [36]	Skin cancer	NA & Electrical equivalent circuit	100 [Hz] – 1 [MHz]: To analyze the tetrapolar probe’s sensitivities’ frequency response. It was discovered that frequency has little effect on sensitivity.	Tetrapolar
4	Ferreira et al. [37]	Skin irritation	NA & Resistance ratio from two depth locations	20 [kHz] & 50 [kHz]: The ratio of total skin impedance obtained at low (20 [kHz]) and high frequencies (500 [kHz]) is the basis of the irritation indices.	Bipolar
5	Gessert et al. [12]	Skin melanoma	CNN and SVM & Ź and Images	1 [kHz] − 2.5 [MHz]: Following the device’s frequency range.	Bipolar
6	Luo et al. [38]	Skin cancer	NA & Transversal and longitudinal of relative permittivity and conductivity	8 – 256 [kHz]: For measurements of lesional and normal skin, intraclass correlation coefficient (ICC) conductivity values were low at 8 and 16 [kHz]. In contrast, relative permittivity ICC results displayed excellent repeatability at 16 [kHz].	Tetrapolar
7	Sarac et al. [39]	Skin cancer	NA & EIS Score	1 [kHz] – 2.5 [MHz]: The extracellular environment affects resistance to low frequencies, whereas both the intracellular and extracellular environments influence readings at higher frequencies.	Bipolar
8	This study	Skin layer classification	FNN & α_\|Z\|, α_θ, α_R, and α_X	$f_{r}^{l h}$ are selected based on DRT results.	Bipolar, Tetrapolar

1 Abbreviations: CNN, convolutional neural network; SVM, support vector machine; FNN. feedforward neural network.

References

Authors

Metrics

Articles in this issue

DOI: https://doi.org/10.2478/joeb-2023-0004 | Journal eISSN: 1891-5469

Journal RSS Feed

Language: English

Page range: 19 - 31

Submitted on: Jul 4, 2023

Accepted on: Aug 5, 2023

Published on: Aug 10, 2023

Published by: University of Oslo

In partnership with: Paradigm Publishing Services

Keywords:

Bioelectrical impedance spectroscopy,

Conductivity change,

Distribution of relaxation times,

Feedforward neural network

Related subjects:

Engineering,

Bioengineering and biomedical engineering,

Biomedical electronics,

Life sciences,

Biophysics,

Medicine,

Biomedical engineering,

Physics,

Spectroscopy and metrology

© 2023 Kiagus Aufa Ibrahim, Marlin Ramadhan Baidillah, Ridwan Wicaksono, Masahiro Takei, published by University of Oslo
This work is licensed under the Creative Commons Attribution 4.0 License.

Volume 14 (2023): Issue 1 (January 2023)

Skin layer classification by feedforward neural network in bioelectrical impedance spectroscopy

Figures & Tables

Fig. 1:

Table 1:

Fig. 2:

Fig. 3:

Fig. 4:

Table 2:

Fig. 5:

Fig. 6:

Fig. 7:

Table 3:

Fig. 8:

Fig. 9:

Fig. 10:

Table 4:

Comparison of FNN accuracy Acc in terms of variation of frequency pair selection from experiment results.

Fig. 11:

Table 5:

Comparison studies of skin condition detection using only BIS or combined with machine learning algorithm.

Paradigm

My account