Deep Learning for Intervention Prediction (August 2016 - present)
MIT CSAIL Clinical Decision Making Group
With the wealth of medical data now available, we have the capability to build automated systems that can take advantage of the experiences of a large cohort of doctors and patients. During my Master's thesis, I used Long Short Term Memory networks (or LSTMs, a type of recurrent neural network) and Convolutional Neural Networks (CNNs) to implement a machine learning system to predict onset and weaning from invasive interventions, like dialysis or ventilation, in a way that is precise and personalized for individual patients. I also focused on making these predictions interpretable.
Feature Representation for ICU Mortality (August 2015 - May 2016)
MIT RLE Computational Biophysics Group
Quantitatively compared existing and new feature representations to optimize mortality prediction tasks with clinical event data. Experiments compared several models of ICU mortality, including regularized logistic regression and random forests.
© 2017 Harini Suresh. All rights reserved.