Secondary Logo

Journal Logo

Institutional members access full text with Ovid®

The Use of Random Forests to Classify Amyloid Brain PET

Zukotynski, Katherine MD*,†; Gaudet, Vincent PhD; Kuo, Phillip H. MD, PhD§; Adamo, Sabrina HBSc; Goubran, Maged PhD; Scott, Christopher BScH; Bocti, Christian MD; Borrie, Michael MB ChB**; Chertkow, Howard MD††; Frayne, Richard PhD‡‡,§§; Hsiung, Robin MD∥∥; Laforce, Robert Jr MD, PhD¶¶; Noseworthy, Michael D. PhD; Prato, Frank S. PhD**; Sahlas, Demetrios J. MD; Smith, Eric E. MD, MPH‡‡; Sossi, Vesna PhD∥∥; Thiel, Alexander MD††; Soucy, Jean-Paul MD***; Tardif, Jean-Claude MD†††; Black, Sandra E. MD‡‡‡

doi: 10.1097/RLU.0000000000002747
Original Articles

Purpose To evaluate random forests (RFs) as a supervised machine learning algorithm to classify amyloid brain PET as positive or negative for amyloid deposition and identify key regions of interest for stratification.

Methods The data set included 57 baseline 18F-florbetapir (Amyvid; Lilly, Indianapolis, IN) brain PET scans in participants with severe white matter disease, presenting with either transient ischemic attack/lacunar stroke or mild cognitive impairment from early Alzheimer disease, enrolled in a multicenter prospective observational trial. Scans were processed using the MINC toolkit to generate SUV ratios, normalized to cerebellar gray matter, and clinically read by 2 nuclear medicine physicians with interpretation based on consensus (35 negative, 22 positive). SUV ratio data and clinical reads were used for supervised training of an RF classifier programmed in MATLAB.

Results A 10,000-tree RF, each tree using 15 randomly selected cases and 20 randomly selected features (SUV ratio per region of interest), with 37 cases for training and 20 cases for testing, had sensitivity = 86% (95% confidence interval [CI], 42%–100%), specificity = 92% (CI, 64%–100%), and classification accuracy = 90% (CI, 68%–99%). The most common features at the root node (key regions for stratification) were (1) left posterior cingulate (1039 trees), (2) left middle frontal gyrus (1038 trees), (3) left precuneus (857 trees), (4) right anterior cingulate gyrus (655 trees), and (5) right posterior cingulate (588 trees).

Conclusions Random forests can classify brain PET as positive or negative for amyloid deposition and suggest key clinically relevant, regional features for classification.

From the *University of Toronto, Toronto

McMaster University, Hamilton

University of Waterloo, Waterloo, ON, Canada

§University of Arizona, Tucson, AZ

Sunnybrook Research Institute, Toronto, ON

Université de Sherbrooke, Sherbrooke, QC

**Western University, London, ON

††Jewish General Hospital, Montreal, QC

‡‡Hotchkiss Brain Institute, University of Calgary, Calgary, AB

§§Seaman Family MR Research Centre, Foothills Medical Centre, Calgary, AB

∥∥University of British Columbia, Vancouver, BC

¶¶Université Laval, Québec, QC

***Montreal Neurological Institute, Montreal, QC

†††Montreal Heart Institute, Université de Montréal, Montreal, QC

‡‡‡Department of Medicine (Neurology), Sunnybrook Health Sciences Centre and Sunnybrook Research Institute, University of Toronto, Toronto, ON, Canada.

Received for publication April 24, 2019; revision accepted June 8, 2019.

Conflicts of interest and sources of funding: Funding was provided by CIHR MITNEC C6 (, from the Linda C Campbell Foundation and for in-kind support from Lilly-Avid Radiopharmaceuticals who generously provided the amyloid ligand. None declared to all authors.

The content of this article was presented at the 2019 ACNM Meeting in Palm Springs, CA.

Correspondence to: Katherine Zukotynski, MD, McMaster University, 1200 Main St W, Room 1P11, Hamilton, Ontario, Canada L8P 3Z5. E-mail:

Online date: May 25, 2019

Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.