Figure 1.
Overview of the modelling approach to predict AN firing rates for different hearing prosthesis options in a personalized way, as well as to obtain the LAP scores for each prosthesis option. The stimulus s(t) is put into models of acoustical stimulation, acoustical amplification, and electrical stimulation to obtain place–time representations of AN firing rates. Firing rates are obtained for the normal hearing reference (nh), unaided hearing (ua), electrical stimulation including residual acoustical hearing (el), sole acoustical amplification (ac), and electrical stimulation combined with acoustical amplification (co). The loss of action potentials (LAP) is obtained for ua, el, ac, and co, by comparing the corresponding AN firing rates with nh. * For modelling low-frequency residual hearing with a cochlear implant, a 20 dB attenuation is applied to the stimulus.
Figure 1.
Overview of the modelling approach to predict AN firing rates for different hearing prosthesis options in a personalized way, as well as to obtain the LAP scores for each prosthesis option. The stimulus s(t) is put into models of acoustical stimulation, acoustical amplification, and electrical stimulation to obtain place–time representations of AN firing rates. Firing rates are obtained for the normal hearing reference (nh), unaided hearing (ua), electrical stimulation including residual acoustical hearing (el), sole acoustical amplification (ac), and electrical stimulation combined with acoustical amplification (co). The loss of action potentials (LAP) is obtained for ua, el, ac, and co, by comparing the corresponding AN firing rates with nh. * For modelling low-frequency residual hearing with a cochlear implant, a 20 dB attenuation is applied to the stimulus.
Figure 2.
Numbers of subjects with regard to the individually predicted best hearing prosthesis, i.e., the prosthesis corresponding to the minimum cost-adjusted loss of action potentials (LAP) score. The types of hearing loss are as follows: (a) normal hearing (NH), (b) cochlear gain loss (CGL), (c) loss of IHCs (IHCL), (d) IHCL + CGL. The types of hearing prosthesis are as follows: unaided hearing (ua), acoustical amplification (ac), electrical stimulation (el), and el combined with ac (co).
Figure 2.
Numbers of subjects with regard to the individually predicted best hearing prosthesis, i.e., the prosthesis corresponding to the minimum cost-adjusted loss of action potentials (LAP) score. The types of hearing loss are as follows: (a) normal hearing (NH), (b) cochlear gain loss (CGL), (c) loss of IHCs (IHCL), (d) IHCL + CGL. The types of hearing prosthesis are as follows: unaided hearing (ua), acoustical amplification (ac), electrical stimulation (el), and el combined with ac (co).
Figure 3.
Numbers of subjects with regard to the individually predicted best hearing prosthesis, i.e., the prosthesis corresponding to the minimum cost-adjusted loss of action potentials (LAP) score. The types of hearing loss are as follows: (a) cochlear synaptopathy (CS), (b) cochlear gain loss (CGL) + CS, (c) loss of IHCs (IHCL) + CGL + CS, (d) IHCL + CGL + CS. The types of hearing prosthesis are as follows: unaided hearing (ua), acoustical amplification (ac), electrical stimulation (el), and el combined with ac (co).
Figure 3.
Numbers of subjects with regard to the individually predicted best hearing prosthesis, i.e., the prosthesis corresponding to the minimum cost-adjusted loss of action potentials (LAP) score. The types of hearing loss are as follows: (a) cochlear synaptopathy (CS), (b) cochlear gain loss (CGL) + CS, (c) loss of IHCs (IHCL) + CGL + CS, (d) IHCL + CGL + CS. The types of hearing prosthesis are as follows: unaided hearing (ua), acoustical amplification (ac), electrical stimulation (el), and el combined with ac (co).
Figure 4.
AN firing rates with regard to fiber index and time. The subject has severe IHCL and CGL. The stimulus was an audio recording of a speaker uttering the word ‘nest’, which was scaled to an equivalent sound pressure level of 65 dB (20 µPa). (a) Normal hearing reference. (b) Unaided hearing, with hearing loss parameters (, , , , and ). (c) Sole acoustical amplification, with amplification parameters ( and ). (d) Sole electrical stimulation, with insertion depth of the CI. (e) Electrical stimulation combined with acoustic amplification, with insertion depth of the CI and amplification parameter . LAP refers to the cost-adjusted loss of action potentials.
Figure 4.
AN firing rates with regard to fiber index and time. The subject has severe IHCL and CGL. The stimulus was an audio recording of a speaker uttering the word ‘nest’, which was scaled to an equivalent sound pressure level of 65 dB (20 µPa). (a) Normal hearing reference. (b) Unaided hearing, with hearing loss parameters (, , , , and ). (c) Sole acoustical amplification, with amplification parameters ( and ). (d) Sole electrical stimulation, with insertion depth of the CI. (e) Electrical stimulation combined with acoustic amplification, with insertion depth of the CI and amplification parameter . LAP refers to the cost-adjusted loss of action potentials.
Figure 5.
AN firing rates with regard to fiber index and time. The subject showed moderate IHCL at low frequencies and severe IHCL at high frequencies. (a) Normal hearing reference. (b) Unaided hearing, with hearing loss parameters (, , , , and ). (c) Sole acoustical amplification, with amplification parameters ( and ). (d) Sole electrical stimulation, with insertion depth of the CI. (e) Electrical stimulation combined with acoustic amplification, with insertion depth of the CI and amplification parameter . LAP refers to the cost-adjusted loss of action potentials.
Figure 5.
AN firing rates with regard to fiber index and time. The subject showed moderate IHCL at low frequencies and severe IHCL at high frequencies. (a) Normal hearing reference. (b) Unaided hearing, with hearing loss parameters (, , , , and ). (c) Sole acoustical amplification, with amplification parameters ( and ). (d) Sole electrical stimulation, with insertion depth of the CI. (e) Electrical stimulation combined with acoustic amplification, with insertion depth of the CI and amplification parameter . LAP refers to the cost-adjusted loss of action potentials.
Figure 6.
AN firing rates with regard to fiber index and time. The subject showed moderate CGL. See (a) Normal hearing reference. (b) Unaided hearing, with hearing loss parameters (, , , , and ). (c) Sole acoustical amplification, with amplification parameters ( and ). (d) Sole electrical stimulation, with insertion depth of the CI. (e) Electrical stimulation combined with acoustic amplification, with insertion depth of the CI and amplification parameter . LAP refers to the cost-adjusted loss of action potentials.
Figure 6.
AN firing rates with regard to fiber index and time. The subject showed moderate CGL. See (a) Normal hearing reference. (b) Unaided hearing, with hearing loss parameters (, , , , and ). (c) Sole acoustical amplification, with amplification parameters ( and ). (d) Sole electrical stimulation, with insertion depth of the CI. (e) Electrical stimulation combined with acoustic amplification, with insertion depth of the CI and amplification parameter . LAP refers to the cost-adjusted loss of action potentials.
Figure 7.
AN firing rates with regard to fiber index and time. The subject showed moderate to severe CS. (a) Normal hearing reference. (b) Unaided hearing, with hearing loss parameters (, , , , and ). (c) Sole acoustical amplification, with amplification parameters ( and ). (d) Sole electrical stimulation, with insertion depth of the CI. (e) Electrical stimulation combined with acoustic amplification, with insertion depth of the CI and amplification parameter . LAP refers to the cost-adjusted loss of action potentials.
Figure 7.
AN firing rates with regard to fiber index and time. The subject showed moderate to severe CS. (a) Normal hearing reference. (b) Unaided hearing, with hearing loss parameters (, , , , and ). (c) Sole acoustical amplification, with amplification parameters ( and ). (d) Sole electrical stimulation, with insertion depth of the CI. (e) Electrical stimulation combined with acoustic amplification, with insertion depth of the CI and amplification parameter . LAP refers to the cost-adjusted loss of action potentials.
Figure 8.
Bar charts of the cost-adjusted loss of action potentials (LAP). Four cases selected for detailed analysis, i.e., the cases reported in
Figure 4,
Figure 5,
Figure 6 and
Figure 7. The bar charts (
a–
d) reflect for each of the four subjects personalized predictions of LAP with regard to the hearing prosthesis options. The shortest bars in each of (
a–
d) reflect the recommended hearing prosthesis option. Their colors are made distinct. The bar charts may be displayed to clinicians as personalized recommendation charts. They provide a concise overview of the prosthesis options, enabling quick judgement of absolute and relative bar heights. Unaided hearing (ua), sole acoustical amplification (ac), sole electrical stimulation (el), and electric stimulation combined with acoustical amplification (co).
Figure 8.
Bar charts of the cost-adjusted loss of action potentials (LAP). Four cases selected for detailed analysis, i.e., the cases reported in
Figure 4,
Figure 5,
Figure 6 and
Figure 7. The bar charts (
a–
d) reflect for each of the four subjects personalized predictions of LAP with regard to the hearing prosthesis options. The shortest bars in each of (
a–
d) reflect the recommended hearing prosthesis option. Their colors are made distinct. The bar charts may be displayed to clinicians as personalized recommendation charts. They provide a concise overview of the prosthesis options, enabling quick judgement of absolute and relative bar heights. Unaided hearing (ua), sole acoustical amplification (ac), sole electrical stimulation (el), and electric stimulation combined with acoustical amplification (co).
Figure 9.
Hearing loss parameters with regard to implantation recommendation. Shown are (a) the flat and (b) sloping components of the loss of IHCs, i.e., and ; (c) the flat and (d) sloping components of the cochlear gain loss, i.e., and ; and (e) the average number of active fibers per IHC , reflecting cochlear synaptopathy.
Figure 9.
Hearing loss parameters with regard to implantation recommendation. Shown are (a) the flat and (b) sloping components of the loss of IHCs, i.e., and ; (c) the flat and (d) sloping components of the cochlear gain loss, i.e., and ; and (e) the average number of active fibers per IHC , reflecting cochlear synaptopathy.
Figure 10.
ROC curves regarding binary prediction of implantation recommendation. The used single predictors are as follows: (a) the flat component of the IHC loss (), (b) the sloping component of the IHC loss (), (c) the flat component of the cochlear gain loss (), (d) the sloping component of the cochlear gain loss (), and (e) the number of active fibers attached to each IHC (). Yellow asterisks and cut-off thresholds THR were obtained at maximal perpendicular distances of the ROC curves to the lines of equality.
Figure 10.
ROC curves regarding binary prediction of implantation recommendation. The used single predictors are as follows: (a) the flat component of the IHC loss (), (b) the sloping component of the IHC loss (), (c) the flat component of the cochlear gain loss (), (d) the sloping component of the cochlear gain loss (), and (e) the number of active fibers attached to each IHC (). Yellow asterisks and cut-off thresholds THR were obtained at maximal perpendicular distances of the ROC curves to the lines of equality.
Figure 11.
Cut-off threshold THR of the flat component of the loss of inner hair cells (dB), with respect to the amplification cost and implantation cost . One operating point (yellow asterisks) was at , and , the other one was at .
Figure 11.
Cut-off threshold THR of the flat component of the loss of inner hair cells (dB), with respect to the amplification cost and implantation cost . One operating point (yellow asterisks) was at , and , the other one was at .
Figure 12.
Minimum LAP achieved with (y-axes) and without (x-axes) implantation. The dashed lines are the lines of equality. The flat component of IHCL is color-coded. The shown types of hearing losses are as follows: (a) IHCL only, (b) CGL and IHCL, (c) IHCL + CS, and (d) CGL + IHCL + CS.
Figure 12.
Minimum LAP achieved with (y-axes) and without (x-axes) implantation. The dashed lines are the lines of equality. The flat component of IHCL is color-coded. The shown types of hearing losses are as follows: (a) IHCL only, (b) CGL and IHCL, (c) IHCL + CS, and (d) CGL + IHCL + CS.
Figure 13.
Minimum LAP achieved with (y-axes) and without (x-axes) implantation. The dashed lines are the lines of equality. The number of functional synapses per IHC is color-coded. The shown types of hearing losses are as follows: (a) CS only, (b) CGL + CS, (c) IHCL + CS, and (d) CGL + IHCL + CS.
Figure 13.
Minimum LAP achieved with (y-axes) and without (x-axes) implantation. The dashed lines are the lines of equality. The number of functional synapses per IHC is color-coded. The shown types of hearing losses are as follows: (a) CS only, (b) CGL + CS, (c) IHCL + CS, and (d) CGL + IHCL + CS.
Table 1.
Distributions of hearing loss parameters with regard to subject groups.
Table 1.
Distributions of hearing loss parameters with regard to subject groups.
| | Hearing Loss Parameters |
---|
Cochlear Gain Loss | Loss of Inner Hair Cells | Cochlear Synaptopathy |
---|
Flat Component | Sloping Component | Flat Component | Sloping Component | Number of Active Fibers |
---|
| | (dB) | (dB) | (dB) | (dB) | |
Subject groups | NH | 0 | 0 | 0 | 0 | 19 |
CS | 0 | 0 | 0 | 0 | |
IHCL | 0 | 0 | | | 19 |
IHCL + CS | 0 | 0 | | | |
CGL | | | 0 | 0 | 19 |
CGL + CS | | | 0 | 0 | |
CGL + IHCL | | | | | 19 |
CGL + IHCL + CS | | | | | |
Table 2.
Logistic regression coefficients for predicting cochlear implantation recommendation from hearing loss parameters .
Table 2.
Logistic regression coefficients for predicting cochlear implantation recommendation from hearing loss parameters .
| | Model Coefficients | p-Value |
---|
| Intercept | 4.875 | <0.001 |
Hearing loss parameters | (dB) | 2.417 | <0.001 |
(dB) | −0.053 | 0.079 |
(dB) | −0.098 | 0.161 |
(dB) | 0.175 | 0.007 |
| −1.754 | <0.001 |
Table 3.
Confusion table of the logistic regression.
Table 3.
Confusion table of the logistic regression.
| | True Class |
---|
| | Implant | No Implant |
Predicted class | Implant | 436 | 6 |
No Implant | 7 | 575 |
Table 4.
Test statistics and confidence intervals regarding the logistic regression.
Table 4.
Test statistics and confidence intervals regarding the logistic regression.
Statistic | Value | 95% Confidence Interval |
---|
Sensitivity | 98.19% | 96.47% to 99.22% |
Specificity | 98.62% | 97.30% to 99.40% |
Positive likelihood ratio | 71.31 | 35.83 to 141.94 |
Negative likelihood ratio | 0.02 | 0.01 to 0.04 |
Positive predictive value | 98.19% | 96.47% to 99.08% |
Negative predictive value | 98.62% | 97.30% to 99.30% |
Accuracy | 98.44% | 97.47% to 99.10% |
Table 5.
Confusion table of the support vector machine classification using a quadratic kernel.
Table 5.
Confusion table of the support vector machine classification using a quadratic kernel.
| | True Class |
---|
| | Implant | No Implant |
Predicted class | Implant | 576 | 5 |
No Implant | 3 | 440 |
Table 6.
Test statistics and confidence intervals regarding the support vector machine classification using a quadratic kernel.
Table 6.
Test statistics and confidence intervals regarding the support vector machine classification using a quadratic kernel.
Statistic | Value | 95% Confidence Interval |
---|
Sensitivity | 99.48% | 98.49% to 99.89% |
Specificity | 98.88% | 97.40% to 99.63% |
Positive likelihood ratio | 88.54 | 37.03 to 211.68 |
Negative likelihood ratio | 0.01 | 0.00 to 0.02 |
Positive predictive value | 99.14% | 97.97% to 99.64% |
Negative predictive value | 99.32% | 97.94% to 99.78% |
Accuracy | 99.22% | 98.47% to 99.66% |