Advanced

Monte Carlo feature selection and rule-based models to predict Alzheimer's disease in mild cognitive impairment

Kruczyk, Marcin; Zetterberg, Henrik; Hansson, Oskar LU ; Rolstad, Sindre; Minthon, Lennart LU ; Wallin, Anders; Blennow, Kaj; Komorowski, Jan and Andersson, Mats Gunnar (2012) In Journal of Neural Transmission 119(7). p.821-831
Abstract
The objective of the present study was to evaluate a Monte Carlo feature selection (MCFS) and rough set Rosetta pipeline for generating rule-based models as a tool for comprehensive risk estimates for future Alzheimer's disease (AD) in individual patients with mild cognitive impairment (MCI). Risk estimates were generated on the basis of age, gender, Mini-Mental State Examination scores, apolipoprotein E (APOE) genotype and the cerebrospinal fluid (CSF) biomarkers total tau (T-tau), phospho-tau(181) (P-tau) and the 42 amino acid form of amyloid beta (A beta 42) in two sets of longitudinally followed MCI patients (n = 217 in total). The predictive model was created in Rosetta, evaluated with the standard tenfold cross-validation approach... (More)
The objective of the present study was to evaluate a Monte Carlo feature selection (MCFS) and rough set Rosetta pipeline for generating rule-based models as a tool for comprehensive risk estimates for future Alzheimer's disease (AD) in individual patients with mild cognitive impairment (MCI). Risk estimates were generated on the basis of age, gender, Mini-Mental State Examination scores, apolipoprotein E (APOE) genotype and the cerebrospinal fluid (CSF) biomarkers total tau (T-tau), phospho-tau(181) (P-tau) and the 42 amino acid form of amyloid beta (A beta 42) in two sets of longitudinally followed MCI patients (n = 217 in total). The predictive model was created in Rosetta, evaluated with the standard tenfold cross-validation approach and tested on an external set. Features were ranked and selected by the MCFS algorithm. Using the combined pipeline of MCFS and Rosetta, it was possible to predict AD among patients with MCI with an area under the receiver operating characteristics curve of 0.92. Risk estimates were produced for the individual patients and showed good correlation with actual diagnosis in cross validation, and on an external dataset from a new study. Analysis of the importance of attributes showed that the biochemical CSF markers contributed the most to the predictions, and that added value was gained by combining several biochemical markers. Despite a correlation with the biochemical markers, the genetic marker APOE epsilon 4 did not contribute to the predictive power of the model. (Less)
Please use this url to cite or link to this publication:
author
organization
publishing date
type
Contribution to journal
publication status
published
subject
keywords
Alzheimer's disease, Decision support, Monte Carlo feature selection, Rosetta, Rough sets, Biomarkers, Cerebrospinal fluid
in
Journal of Neural Transmission
volume
119
issue
7
pages
821 - 831
publisher
Springer
external identifiers
  • wos:000305525800012
  • scopus:84863465731
ISSN
0300-9564
DOI
10.1007/s00702-012-0812-0
language
English
LU publication?
yes
id
da35ccae-e95f-49a4-b768-8bb35d1986bd (old id 2883915)
date added to LUP
2012-08-01 09:41:10
date last changed
2017-07-30 04:08:15
@article{da35ccae-e95f-49a4-b768-8bb35d1986bd,
  abstract     = {The objective of the present study was to evaluate a Monte Carlo feature selection (MCFS) and rough set Rosetta pipeline for generating rule-based models as a tool for comprehensive risk estimates for future Alzheimer's disease (AD) in individual patients with mild cognitive impairment (MCI). Risk estimates were generated on the basis of age, gender, Mini-Mental State Examination scores, apolipoprotein E (APOE) genotype and the cerebrospinal fluid (CSF) biomarkers total tau (T-tau), phospho-tau(181) (P-tau) and the 42 amino acid form of amyloid beta (A beta 42) in two sets of longitudinally followed MCI patients (n = 217 in total). The predictive model was created in Rosetta, evaluated with the standard tenfold cross-validation approach and tested on an external set. Features were ranked and selected by the MCFS algorithm. Using the combined pipeline of MCFS and Rosetta, it was possible to predict AD among patients with MCI with an area under the receiver operating characteristics curve of 0.92. Risk estimates were produced for the individual patients and showed good correlation with actual diagnosis in cross validation, and on an external dataset from a new study. Analysis of the importance of attributes showed that the biochemical CSF markers contributed the most to the predictions, and that added value was gained by combining several biochemical markers. Despite a correlation with the biochemical markers, the genetic marker APOE epsilon 4 did not contribute to the predictive power of the model.},
  author       = {Kruczyk, Marcin and Zetterberg, Henrik and Hansson, Oskar and Rolstad, Sindre and Minthon, Lennart and Wallin, Anders and Blennow, Kaj and Komorowski, Jan and Andersson, Mats Gunnar},
  issn         = {0300-9564},
  keyword      = {Alzheimer's disease,Decision support,Monte Carlo feature selection,Rosetta,Rough sets,Biomarkers,Cerebrospinal fluid},
  language     = {eng},
  number       = {7},
  pages        = {821--831},
  publisher    = {Springer},
  series       = {Journal of Neural Transmission},
  title        = {Monte Carlo feature selection and rule-based models to predict Alzheimer's disease in mild cognitive impairment},
  url          = {http://dx.doi.org/10.1007/s00702-012-0812-0},
  volume       = {119},
  year         = {2012},
}