publications

2019

1. MICCAI

   Deep Learning vs. Classical Machine Learning: A Comparison of Methods for Fluid Intelligence Prediction Guerdan, L*, Sun, P*, Rowland, C, Harrison, L, Tang, Z, Wergeles, N, and Shang, Y In Adolescent Brain Cognitive Development Neurocognitive Prediction 2019 [Abs] [PDF]

   Predicting fluid intelligence based on T1-weighted magnetic resonance imaging (MRI) scans poses several challenges, including developing an adequate data representation of three dimensional voxel data, extracting predictive information from this data representation, and devising a model that is able to leverage the predictive information. We evaluate two strategies for prediction of fluid intelligence given structural MRI scans acquired through the Adolescent Brain Cognitive Development (ABCD) Study: deep learning models trained on raw imagery and classical machine learning models trained on extracted features. Our best-performing solution consists of a classical machine learning model trained on a combination of provided brain volume estimates and extracted features. Specifically, a Gradient Boosting Regressor (GBR) trained on a PCA-reduced feature space produced the best performance (train MSE = 66.29, validation MSE = 70.16), surpassing regression models trained on the provided volume data alone, and 2D/3D Convolutional Neural Networks trained on various representations of imagery data. Nonetheless, these results remain slightly better than a mean prediction, suggesting that neither approach is capturing a high degree of variance in the data.

2. NER

   Extracting Motion-Related Subspaces from EEG in Mobile Brain/Body Imaging Studies Gehrke, L*., Guerdan, L*., and Gramman, K 9th International IEEE/EMBS Conference on Neural Engineering, 2019 [Abs] [PDF] [Poster]

   Mobile Brain/Body Imaging (MoBI) is an emerging non-invasive approach to investigate human brain activity and motor behavior associated with cognitive processes in natural conditions. MoBI studies and analyses pipelines com- bine brain measurements, e.g. Electroencephalography (EEG), with motion data as participants conduct tasks with near- natural behavior. Within the field, standard source decomposition and reconstruction pipelines largely rely on unsupervised blind source separation (BSS) approaches and do not consider movement information to guide the decomposition of oscillatory brain sources. We propose the use of a super- vised spatial filtering method, Source Power Co-modulation (SPoC), for extracting source components that co-modulate with body motion. Further, we introduce a method to validate the quality of oscillatory sources in MoBI studies. We illustrate the approach to investigate the link between hand and head movement kinematics and power dynamics of EEG sources while participants explore an invisible maze in virtual reality. Stable oscillatory source envelopes correlating with hand and head motion were isolated in all subjects, with median ρ = .13 for all sources and median ρ = .16 for sources passing the selection criteria. The results indicate that it is possible to improve movement related source separation to further guide our understanding of how movement and brain dynamics interact.

2018

1. DSC

   TigerAware: An Innovative Mobile Survey and Sensor Data Collection and Analytics System Morrison, M., Guerdan, L., Kanugo, J., Trull, T., and Shang, Y. IEEE DSC, 2018

2017

1. MobileCloud

   Augmented Resource Allocation Framework for Disaster Response Coordination in Mobile Cloud Environments Guerdan, L., Apperson, O., and Calyam, P 5th IEEE International Conference on Mobile Cloud Computing, Services, and Engineering 2017 [PDF] [Poster]

2016

1. SmartComp

   ADA - Automatic Detection of Alcohol Usage for Mobile Ambulatory Assessment, Sun, P., Wergeles, N., Zhang, C., Guerdan, L., Trull, T., and Shang, Y IEEE International Conference on Smart Computing 2016