Boosting 1H-MRS Alzheimer Diagnosis with Boosted Trees

Fengmao Lv, Yuchen Li, Guowu Yang

Abstract

Nowdays, more and more attention has been attached to early-onset alzheimer's disease (EOAD). It is well known that the Presenilin-1 gene (PSEN1) penetrant mutation can cause EOAD, and subjects with the PSEN1 p.Glu280Ala (E280A) mutation are at high risk of EOAD. Sensitive to the the progress of dementia, the brain metabolites measured by proton magnetic resonance spectroscopy ( <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> H-MRS) can be used as informative biomarkers of EOAD. In this work, we focus on designing an effective diagnosis framework for EOAD caused by E280A mutation through the combination of <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> H-MRS biomarkers and machine learning techniques. Specifically, we utilize gradient boosting decision tree (GBDT), which is an advanced machine learning tool, to analyze the physiological characteristics of EOAD patients and E280A mutation carriers. According to the relative importance of each <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> H-MRS variable provided by GBDT, informative brain metabolites are selected out as biomarkers of AD symptoms. Combining the <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> H-MRS biomarkers with GBDT, our method can achieve desirable predictive performance. <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> MCI is the transitional state of AD, which involves early cognitive impairments [4].
BibTeX
@inproceedings{lv2018boosting,
  author = {Fengmao Lv and Yuchen Li and Guowu Yang},
  title = {{Boosting <sup>1</sup>H-MRS Alzheimer Diagnosis with Boosted Trees}},
  booktitle = {BIBM},
  year = {2018},
  url = {https://openreview.net/forum?id=z1FtVWeNVv},
}

Loading related content...

Figure