README.md
In fernandoPalluzzi/tmsClassifier: Random Forest Classifiers for Transcranial Magnetic Stimulation Data
tmsClassifier
Random Forest Classifiers for Transcranial Magnetic Stimulation Data
The tmsClassifier package predicts subject diagnosis (Frontotemporal Dementia, Alzheimer's Disease, Dementia with Lewy Bodies, or Healthy) based on Transcranial Magnetic Stimulation (TMS) data, through the integration of linear modeling and Random Forest Classifiers.
Installation
The development version of tmsClassifier can be installed from GitHub:
```{r, eval = FALSE}
devtools::install_github("fernandoPalluzzi/tmsClassifier")
## Unknown subject classification
With the term of "unknown subject" we define a patient with no ascertained diagnosis. Package **tmsClassifier** provides a default TMS coefficient matrix, that can be used to predict unknown subject diagnosis.
Default TMS signatures include: Alzheimer's Disease (AD), Frontotemporal Dementia (FTD), Dementia with Lewy Bodies (DLB), and healthy controls (HC).
The prediction can be simply done with the `tmsClassify` command:
```{r, eval = FALSE}
library(tmsClassifier)
# Sample 10 random subjects from default TMS data
n.subjects <- 10
TMS <- tms.coef[sample(nrow(tms.coef), n.subjects), ]
# Subjects' diagnosis prediction
predicted <- tmsClassify(TMS)
The input TMS object must be a data.frame containing TMS regression coefficients as columns and subjects as rows. In this example, we used estimated TMS regression coefficients from default data (object tms.coef). The user may generate a custom TMS coefficients matrix using tmsClassifier functions (see below and package documentation for details).
Input TMS data and imputation
The TMS regression coefficient matrix can be directly generated from TMS data and, optionally, subject covariates (e.g., age and sex). Package tmsClassifier uses a default TMS data format, including three blocks of attributes (columns): (i) TMS measures, (ii) diagnosis field (optional), and (iii) covariates (optional).
The default dataset includes 4 TMS measures:
- SICI-ICF (SICI: short-interval intracortical inhibition; ICF: intracortical facilitation) as the first 7 columns, taken at times (interstimulus intervals, ISI): 1, 2, 3, 5 ms (SICI) and 7, 10, 15 ms (ICF);
- SAI (short-latency afferent inhibition). By default, they should be the 4 columns following SICI-ICF, taken at time steps (ISI): -4, 0, 4, 8 ms;
- LICI (long-interval intracortical inhibition). By default, they should be the 3 columns following SAI, taken at time
steps (ISI): 50, 100, 150 ms.
Optionally, the user may provide a column specifying subjects' diagnosis (e.g., the diagnosis field in the example below). This field is not required for diagnosis prediction, but only to test RFC performances (see tmsClassifier documentation). In the TMS sample data below, the last three columns are common covariates: center (i.e., where data were generated), sex (i.e., subject's sex), and age (i.e., subject's age at which TMS was done). If required by the user, TMS regression parameters can be adjusted to remove covariates influence.
```{r, eval = FALSE}
SICI_1 SICI_2 SICI_3 SICI_5 ICF_7 ICF_10 ICF_15 SAI_m4 SAI_0
1 0.405 0.63600 0.797 1.510 1.9640000 2.057 1.390000 0.7800000 0.92500
2 0.061 0.05800 0.100 1.159 1.1810000 1.745 1.694000 0.8500000 0.87400
3 0.570 0.72500 0.699 1.085 1.8420000 1.875 2.418000 1.1320000 0.47900
4 0.000 0.00000 0.444 0.556 0.8847778 1.111 1.262250 0.9270000 0.54225
5 0.080 0.19425 0.150 0.170 1.1300000 1.300 1.301875 0.8518571 0.42175
6 1.031 0.95900 0.791 1.082 1.1480000 0.843 0.780000 0.9420000 0.38600
7 0.406 0.80800 1.492 1.649 1.5080000 1.001 2.126000 0.9640000 1.07400
8 1.142 1.87200 1.872 1.121 1.7080000 1.738 1.766000 0.9255000 1.06100
9 0.097 0.60300 0.851 0.837 0.9810000 1.002 0.920000 0.8690000 0.60000
10 0.380 0.78100 1.081 1.404 1.5190000 1.311 1.387000 0.9630000 1.28600
SAI_4 SAI_8 LICI_50 LICI_100 LICI_150 diagnosis center sex age
1 0.772 0.557 0.1590000 0.4190000 0.4250000 AD C1 M 76
2 0.779 0.899 0.2730000 0.3872857 0.4587143 AD C1 F 68
3 0.283 0.592 0.3397143 0.6042857 0.5590000 HC C3 F 63
4 0.673 0.577 0.2573333 0.1993333 0.5286667 HC C2 M 76
5 0.450 0.450 0.2380000 0.1784000 0.3358000 DLB C4 F 77
6 0.563 1.070 0.0300000 0.1560000 0.6860000 FTD C1 F 63
7 0.663 0.888 0.4985000 0.7678750 0.5547500 AD C1 M 57
8 1.053 1.038 0.2921429 0.4844286 0.4631429 AD C2 F 68
9 0.489 0.858 0.5810000 0.6040000 0.4580000 FTD C1 M 52
10 0.860 0.890 0.5860000 0.5800000 0.5900000 DLB C1 M 78
If TMS data contain missing values (`NA`), they can be inputed with the `tmsImpute` command:
```{r, eval = FALSE}
tms <- tmsImpute(tms.withNA)
yielding the input dataset with imputed missing values.
TMS regression parameters
If TMS data do not contain missing values, it can be used to generate TMS regression coefficients. This matrix is used as input for subject diagnosis prediction and other tmsClassifier functions. For each subject, TMS indicators (SICI-ICF, SAI, LICI) are modeled as polynomial functions of time (t), in the form y ~ poly(t). This function estimates two parameters for SICI (SICI = bs0 + bs*t), two parameters for ICF (ICF = bi0 + bi*t), three parameters for SAI (SAI = b0 + b1*t + b2*t^2), and two parameters for LICI (LICI = a0 + a1*t). Regression coefficients can be generated using the command:
```{r, eval = FALSE}
No covariates adjustment
B <- tmsRegression(tms)
Adjustment by center (column 16), sex (column 17), and age (column 18)
B <- tmsRegression(tms, adjust = 16:18)
Matrix B contains the 9 TMS regression coefficients, as shown in the example below:
```{r, eval = FALSE}
bs0 bs bi0 bi b0
1 0.44605611 -0.172286425 -0.28845631 -0.0103719530 -0.079070482
2 -0.63479541 0.309619620 1.32187388 -0.0846722830 0.247493127
3 -0.24740092 -0.049733826 0.64390434 -0.0071768698 -0.070655046
4 -0.35290457 0.028751748 0.24985075 -0.0202048839 -0.170287969
5 -0.01846587 0.006138788 -0.23161357 0.0043496782 0.136867922
6 0.06317109 0.103245351 0.43792815 -0.0026170343 0.001647889
7 0.07581085 -0.057192403 -0.05528611 -0.0066774232 0.215741779
8 -0.01910764 -0.074350541 0.47464356 -0.0002787369 -0.154575367
9 0.13769556 0.068909513 0.83590337 0.0290674695 0.086760707
10 0.23425097 -0.050166576 -0.29430881 -0.0002431529 -0.167022626
b1 b2 a0 a1
1 -0.0070619753 -0.0034500884 -0.9269894 0.011129428
2 0.0285864074 -0.0065340234 0.1173041 0.061815564
3 0.0082017802 -0.0001614682 -0.2955762 -0.004567065
4 0.0335833733 -0.0050643560 -1.3435844 0.008719332
5 0.0477466022 -0.0086142874 -0.4132057 -0.002285809
6 0.0201640174 -0.0005187694 2.3338259 0.089213463
7 0.0310105467 -0.0035992202 -9.8801634 -0.164766244
8 -0.0424625002 0.0058572147 0.1134623 -0.007114505
9 0.0140306025 -0.0083004128 -0.7459555 0.000597424
10 0.0008075851 0.0019799213 0.5456834 -0.007858925
The default TMS regression coefficients matrix in tmsClassifier (object tms.coef) is generated from 694 subjects: 273 Alzheimer's Disease patients (AD), 207 Frontotemporal Dementia patients (FTD), 67 Dementia with Lewy Bodies patients (DLB), and 147 healthy controls (HC).
References
Benussi A, Grassi M, Palluzzi F, Cantoni V, Cotelli MS, Premi E, Di Lorenzo F, Pellicciari MC, Ranieri F, Musumeci G, Marra C, Manganotti P, Nardone R, Di Lazzaro V, Koch G, Borroni B. Classification Accuracy of TMS for the Diagnosis of Mild Cognitive Impairment. Brain Stimulation 2021 Mar; 14(2):
241-249. PMID: 33453454. DOI: 10.1016/j.brs.2021.01.004. * These authors contributed equally*.
Benussi A, Grassi M, Palluzzi F, Koch G, Di Lazzaro V, Nardone R, Cantoni V, Dell'Era V, Premi E, Martorana A, Di Lorenzo F, Bonnì S, Ranieri F, Capone F, Musumeci G, Cotelli MS, Padovani A, Borroni B. Classification Accuracy of Transcranial Magnetic Stimulation for the Diagnosis of
Neurodegenerative Dementias. Annals of Neurology 2020 Mar; 87(3): 394-404. PMID: 1925823. DOI: 10.1002/ana.25677. * These authors contributed equally*.
fernandoPalluzzi/tmsClassifier documentation built on Feb. 3, 2021, 12:31 p.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
tmsClassifier
Random Forest Classifiers for Transcranial Magnetic Stimulation Data
The tmsClassifier package predicts subject diagnosis (Frontotemporal Dementia, Alzheimer's Disease, Dementia with Lewy Bodies, or Healthy) based on Transcranial Magnetic Stimulation (TMS) data, through the integration of linear modeling and Random Forest Classifiers.
Installation
The development version of tmsClassifier can be installed from GitHub:
```{r, eval = FALSE} devtools::install_github("fernandoPalluzzi/tmsClassifier")
## Unknown subject classification
With the term of "unknown subject" we define a patient with no ascertained diagnosis. Package **tmsClassifier** provides a default TMS coefficient matrix, that can be used to predict unknown subject diagnosis.
Default TMS signatures include: Alzheimer's Disease (AD), Frontotemporal Dementia (FTD), Dementia with Lewy Bodies (DLB), and healthy controls (HC).
The prediction can be simply done with the `tmsClassify` command:
```{r, eval = FALSE}
library(tmsClassifier)
# Sample 10 random subjects from default TMS data
n.subjects <- 10
TMS <- tms.coef[sample(nrow(tms.coef), n.subjects), ]
# Subjects' diagnosis prediction
predicted <- tmsClassify(TMS)
The input TMS object must be a data.frame containing TMS regression coefficients as columns and subjects as rows. In this example, we used estimated TMS regression coefficients from default data (object tms.coef). The user may generate a custom TMS coefficients matrix using tmsClassifier functions (see below and package documentation for details).
Input TMS data and imputation
The TMS regression coefficient matrix can be directly generated from TMS data and, optionally, subject covariates (e.g., age and sex). Package tmsClassifier uses a default TMS data format, including three blocks of attributes (columns): (i) TMS measures, (ii) diagnosis field (optional), and (iii) covariates (optional). The default dataset includes 4 TMS measures: - SICI-ICF (SICI: short-interval intracortical inhibition; ICF: intracortical facilitation) as the first 7 columns, taken at times (interstimulus intervals, ISI): 1, 2, 3, 5 ms (SICI) and 7, 10, 15 ms (ICF); - SAI (short-latency afferent inhibition). By default, they should be the 4 columns following SICI-ICF, taken at time steps (ISI): -4, 0, 4, 8 ms; - LICI (long-interval intracortical inhibition). By default, they should be the 3 columns following SAI, taken at time steps (ISI): 50, 100, 150 ms.
Optionally, the user may provide a column specifying subjects' diagnosis (e.g., the diagnosis field in the example below). This field is not required for diagnosis prediction, but only to test RFC performances (see tmsClassifier documentation). In the TMS sample data below, the last three columns are common covariates: center (i.e., where data were generated), sex (i.e., subject's sex), and age (i.e., subject's age at which TMS was done). If required by the user, TMS regression parameters can be adjusted to remove covariates influence.
```{r, eval = FALSE} SICI_1 SICI_2 SICI_3 SICI_5 ICF_7 ICF_10 ICF_15 SAI_m4 SAI_0 1 0.405 0.63600 0.797 1.510 1.9640000 2.057 1.390000 0.7800000 0.92500 2 0.061 0.05800 0.100 1.159 1.1810000 1.745 1.694000 0.8500000 0.87400 3 0.570 0.72500 0.699 1.085 1.8420000 1.875 2.418000 1.1320000 0.47900 4 0.000 0.00000 0.444 0.556 0.8847778 1.111 1.262250 0.9270000 0.54225 5 0.080 0.19425 0.150 0.170 1.1300000 1.300 1.301875 0.8518571 0.42175 6 1.031 0.95900 0.791 1.082 1.1480000 0.843 0.780000 0.9420000 0.38600 7 0.406 0.80800 1.492 1.649 1.5080000 1.001 2.126000 0.9640000 1.07400 8 1.142 1.87200 1.872 1.121 1.7080000 1.738 1.766000 0.9255000 1.06100 9 0.097 0.60300 0.851 0.837 0.9810000 1.002 0.920000 0.8690000 0.60000 10 0.380 0.78100 1.081 1.404 1.5190000 1.311 1.387000 0.9630000 1.28600 SAI_4 SAI_8 LICI_50 LICI_100 LICI_150 diagnosis center sex age 1 0.772 0.557 0.1590000 0.4190000 0.4250000 AD C1 M 76 2 0.779 0.899 0.2730000 0.3872857 0.4587143 AD C1 F 68 3 0.283 0.592 0.3397143 0.6042857 0.5590000 HC C3 F 63 4 0.673 0.577 0.2573333 0.1993333 0.5286667 HC C2 M 76 5 0.450 0.450 0.2380000 0.1784000 0.3358000 DLB C4 F 77 6 0.563 1.070 0.0300000 0.1560000 0.6860000 FTD C1 F 63 7 0.663 0.888 0.4985000 0.7678750 0.5547500 AD C1 M 57 8 1.053 1.038 0.2921429 0.4844286 0.4631429 AD C2 F 68 9 0.489 0.858 0.5810000 0.6040000 0.4580000 FTD C1 M 52 10 0.860 0.890 0.5860000 0.5800000 0.5900000 DLB C1 M 78
If TMS data contain missing values (`NA`), they can be inputed with the `tmsImpute` command:
```{r, eval = FALSE}
tms <- tmsImpute(tms.withNA)
yielding the input dataset with imputed missing values.
TMS regression parameters
If TMS data do not contain missing values, it can be used to generate TMS regression coefficients. This matrix is used as input for subject diagnosis prediction and other tmsClassifier functions. For each subject, TMS indicators (SICI-ICF, SAI, LICI) are modeled as polynomial functions of time (t), in the form y ~ poly(t). This function estimates two parameters for SICI (SICI = bs0 + bs*t), two parameters for ICF (ICF = bi0 + bi*t), three parameters for SAI (SAI = b0 + b1*t + b2*t^2), and two parameters for LICI (LICI = a0 + a1*t). Regression coefficients can be generated using the command:
```{r, eval = FALSE}
No covariates adjustment
B <- tmsRegression(tms)
Adjustment by center (column 16), sex (column 17), and age (column 18)
B <- tmsRegression(tms, adjust = 16:18)
Matrix B contains the 9 TMS regression coefficients, as shown in the example below:
```{r, eval = FALSE}
bs0 bs bi0 bi b0
1 0.44605611 -0.172286425 -0.28845631 -0.0103719530 -0.079070482
2 -0.63479541 0.309619620 1.32187388 -0.0846722830 0.247493127
3 -0.24740092 -0.049733826 0.64390434 -0.0071768698 -0.070655046
4 -0.35290457 0.028751748 0.24985075 -0.0202048839 -0.170287969
5 -0.01846587 0.006138788 -0.23161357 0.0043496782 0.136867922
6 0.06317109 0.103245351 0.43792815 -0.0026170343 0.001647889
7 0.07581085 -0.057192403 -0.05528611 -0.0066774232 0.215741779
8 -0.01910764 -0.074350541 0.47464356 -0.0002787369 -0.154575367
9 0.13769556 0.068909513 0.83590337 0.0290674695 0.086760707
10 0.23425097 -0.050166576 -0.29430881 -0.0002431529 -0.167022626
b1 b2 a0 a1
1 -0.0070619753 -0.0034500884 -0.9269894 0.011129428
2 0.0285864074 -0.0065340234 0.1173041 0.061815564
3 0.0082017802 -0.0001614682 -0.2955762 -0.004567065
4 0.0335833733 -0.0050643560 -1.3435844 0.008719332
5 0.0477466022 -0.0086142874 -0.4132057 -0.002285809
6 0.0201640174 -0.0005187694 2.3338259 0.089213463
7 0.0310105467 -0.0035992202 -9.8801634 -0.164766244
8 -0.0424625002 0.0058572147 0.1134623 -0.007114505
9 0.0140306025 -0.0083004128 -0.7459555 0.000597424
10 0.0008075851 0.0019799213 0.5456834 -0.007858925
The default TMS regression coefficients matrix in tmsClassifier (object tms.coef) is generated from 694 subjects: 273 Alzheimer's Disease patients (AD), 207 Frontotemporal Dementia patients (FTD), 67 Dementia with Lewy Bodies patients (DLB), and 147 healthy controls (HC).
References
Benussi A, Grassi M, Palluzzi F, Cantoni V, Cotelli MS, Premi E, Di Lorenzo F, Pellicciari MC, Ranieri F, Musumeci G, Marra C, Manganotti P, Nardone R, Di Lazzaro V, Koch G, Borroni B. Classification Accuracy of TMS for the Diagnosis of Mild Cognitive Impairment. Brain Stimulation 2021 Mar; 14(2): 241-249. PMID: 33453454. DOI: 10.1016/j.brs.2021.01.004. * These authors contributed equally*.
Benussi A, Grassi M, Palluzzi F, Koch G, Di Lazzaro V, Nardone R, Cantoni V, Dell'Era V, Premi E, Martorana A, Di Lorenzo F, Bonnì S, Ranieri F, Capone F, Musumeci G, Cotelli MS, Padovani A, Borroni B. Classification Accuracy of Transcranial Magnetic Stimulation for the Diagnosis of Neurodegenerative Dementias. Annals of Neurology 2020 Mar; 87(3): 394-404. PMID: 1925823. DOI: 10.1002/ana.25677. * These authors contributed equally*.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.