TY - JOUR
T1 - Robust deep learning classification of adamantinomatous craniopharyngioma from limited preoperative radiographic images
AU - Prince, Eric W.
AU - Whelan, Ros
AU - Mirsky, David M.
AU - Stence, Nicholas
AU - Staulcup, Susan
AU - Klimo, Paul
AU - Anderson, Richard C.E.
AU - Niazi, Toba N.
AU - Grant, Gerald
AU - Souweidane, Mark
AU - Johnston, James M.
AU - Jackson, Eric M.
AU - Limbrick, David D.
AU - Smith, Amy
AU - Drapeau, Annie
AU - Chern, Joshua J.
AU - Kilburn, Lindsay
AU - Ginn, Kevin
AU - Naftel, Robert
AU - Dudley, Roy
AU - Tyler-Kabara, Elizabeth
AU - Jallo, George
AU - Handler, Michael H.
AU - Jones, Kenneth
AU - Donson, Andrew M.
AU - Foreman, Nicholas K.
AU - Hankinson, Todd C.
N1 - Funding Information:
The authors wish to express their gratitude for study coordinators Anastasia Arynchna (University of Alabama Birmingham), Hannah Goldstein (Columbia University), Stephen Gannon (Vanderbilt University), Cor-rine Gardner (Washington University St. Louis), Anthony Bet (Stanford University), Nassima Addour (McGill Univeristy), Kari Bollerman (Miami Children’s Hospital), Alyson Hignight (Cornell Univeristy), Robyn Ryans (Children’s Mercy Hospital), Kris Laurence (Children’s Mercy Hospital), Lisa Tetreault (Johns Hopkins All Children’s Hospital), Jennifer Spinelli (Orlando Health), Kaitlin Hardy (Children’s National Medical Center), Sabrina Malik (Children’s National Medical Center), and Brandy Vaughn (Lebonheur Children’s Research Hospital) for their assistance in making this study possible. The authors also wish to thank the University of Colorado Comprehensive Cancer Center for funding that supported this work (P30CA046934).
Publisher Copyright:
© 2020, The Author(s).
PY - 2020/12/1
Y1 - 2020/12/1
N2 - Deep learning (DL) is a widely applied mathematical modeling technique. Classically, DL models utilize large volumes of training data, which are not available in many healthcare contexts. For patients with brain tumors, non-invasive diagnosis would represent a substantial clinical advance, potentially sparing patients from the risks associated with surgical intervention on the brain. Such an approach will depend upon highly accurate models built using the limited datasets that are available. Herein, we present a novel genetic algorithm (GA) that identifies optimal architecture parameters using feature embeddings from state-of-the-art image classification networks to identify the pediatric brain tumor, adamantinomatous craniopharyngioma (ACP). We optimized classification models for preoperative Computed Tomography (CT), Magnetic Resonance Imaging (MRI), and combined CT and MRI datasets with demonstrated test accuracies of 85.3%, 83.3%, and 87.8%, respectively. Notably, our GA improved baseline model performance by up to 38%. This work advances DL and its applications within healthcare by identifying optimized networks in small-scale data contexts. The proposed system is easily implementable and scalable for non-invasive computer-aided diagnosis, even for uncommon diseases.
AB - Deep learning (DL) is a widely applied mathematical modeling technique. Classically, DL models utilize large volumes of training data, which are not available in many healthcare contexts. For patients with brain tumors, non-invasive diagnosis would represent a substantial clinical advance, potentially sparing patients from the risks associated with surgical intervention on the brain. Such an approach will depend upon highly accurate models built using the limited datasets that are available. Herein, we present a novel genetic algorithm (GA) that identifies optimal architecture parameters using feature embeddings from state-of-the-art image classification networks to identify the pediatric brain tumor, adamantinomatous craniopharyngioma (ACP). We optimized classification models for preoperative Computed Tomography (CT), Magnetic Resonance Imaging (MRI), and combined CT and MRI datasets with demonstrated test accuracies of 85.3%, 83.3%, and 87.8%, respectively. Notably, our GA improved baseline model performance by up to 38%. This work advances DL and its applications within healthcare by identifying optimized networks in small-scale data contexts. The proposed system is easily implementable and scalable for non-invasive computer-aided diagnosis, even for uncommon diseases.
UR - http://www.scopus.com/inward/record.url?scp=85092319026&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85092319026&partnerID=8YFLogxK
U2 - 10.1038/s41598-020-73278-8
DO - 10.1038/s41598-020-73278-8
M3 - Article
C2 - 33037266
AN - SCOPUS:85092319026
SN - 2045-2322
VL - 10
JO - Scientific Reports
JF - Scientific Reports
IS - 1
M1 - 16885
ER -