ISeeU: Visually interpretable deep learning for mortality prediction inside the ICU

To improve the performance of Intensive Care Units (ICUs), the field of bio-statistics has developed scores which try to predict the likelihood of negative outcomes. These help evaluate the effectiveness of treatments and clinical practice, and also help to identify patients with unexpected outcomes. However, they have been shown by several studies to offer sub-optimal performance. Alternatively, Deep Learning offers state of the art capabilities in certain prediction tasks and research suggests deep neural networks are able to outperform traditional techniques. Nevertheless, a main impediment for the adoption of Deep Learning in healthcare is its reduced interpretability, for in this field it is crucial to gain insight into the why of predictions, to assure that models are actually learning relevant features instead of spurious correlations. To address this, we propose a deep multi-scale convolutional architecture trained on the Medical Information Mart for Intensive Care III (MIMIC-III) for mortality prediction, and the use of concepts from coalitional game theory to construct visual explanations aimed to show how important these inputs are deemed by the network. Results show our model attains a ROC AUC of 0.8735 (± 0.0025) which is competitive with the state of the art of Deep Learning mortality models trained on MIMIC-III data, while remaining interpretable. Supporting code can be found at https://github.com/williamcaicedo/ISeeU.

[1]  Jason H. Moore,et al.  Mapping Patient Trajectories using Longitudinal Extraction and Deep Learning in the MIMIC-III Critical Care Database , 2017, bioRxiv.

[2]  Jürgen Schmidhuber,et al.  Long Short-Term Memory , 1997, Neural Computation.

[3]  Syed Muhammad Anwar,et al.  Deep Learning in Medical Image Analysis , 2017 .

[4]  Matthew Richardson,et al.  Do Deep Convolutional Nets Really Need to be Deep (Or Even Convolutional)? , 2016, ArXiv.

[5]  Rob Fergus,et al.  Visualizing and Understanding Convolutional Networks , 2013, ECCV.

[6]  Scott Lundberg,et al.  A Unified Approach to Interpreting Model Predictions , 2017, NIPS.

[7]  David C. Kale,et al.  Directly Modeling Missing Data in Sequences with RNNs: Improved Classification of Clinical Time Series , 2016, MLHC.

[8]  Charles Elkan,et al.  Learning to Diagnose with LSTM Recurrent Neural Networks , 2015, ICLR.

[9]  David Sontag,et al.  Temporal Convolutional Neural Networks for Diagnosis from Lab Tests , 2015, ArXiv.

[10]  Jian Sun,et al.  Deep Residual Learning for Image Recognition , 2015, 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[11]  G. G. Stokes.  "J." , 1890, The New Yale Book of Quotations.

[12]  L. S. Shapley,et al.  17. A Value for n-Person Games , 1953 .

[13]  Peter Szolovits,et al.  MIMIC-III, a freely accessible critical care database , 2016, Scientific Data.

[14]  Yan Liu,et al.  Benchmarking deep learning models on large healthcare datasets , 2018, J. Biomed. Informatics.

[15]  Linda G. Shapiro,et al.  Computer Vision , 2001 .

[16]  Azriel Rosenfeld,et al.  Computer Vision , 1988, Adv. Comput..

[17]  Shamim Nemati,et al.  Machine Learning and Decision Support in Critical Care , 2016, Proceedings of the IEEE.

[18]  Constantin F. Aliferis,et al.  An evaluation of machine-learning methods for predicting pneumonia mortality , 1997, Artif. Intell. Medicine.

[19]  Dumitru Erhan,et al.  Going deeper with convolutions , 2014, 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[20]  T. Lasko,et al.  Computational Phenotype Discovery Using Unsupervised Feature Learning over Noisy, Sparse, and Irregular Clinical Data , 2013, PloS one.

[21]  S. Lemeshow,et al.  A new Simplified Acute Physiology Score (SAPS II) based on a European/North American multicenter study. , 1993, JAMA.

[22]  Yan Liu,et al.  Causal Phenotype Discovery via Deep Networks , 2015, AMIA.

[23]  Zachary Chase Lipton.  The mythos of model interpretability , 2016, ACM Queue.

[24]  A. Rapsang,et al.  Scoring systems in the intensive care unit: A compendium , 2014, Indian journal of critical care medicine : peer-reviewed, official publication of Indian Society of Critical Care Medicine.

[25]  Klaus-Robert Müller,et al.  Investigating the influence of noise and distractors on the interpretation of neural networks , 2016, ArXiv.

[26]  Nitish Srivastava,et al.  Dropout: a simple way to prevent neural networks from overfitting , 2014, J. Mach. Learn. Res..

[27]  Erik Strumbelj,et al.  An Efficient Explanation of Individual Classifications using Game Theory , 2010, J. Mach. Learn. Res..

[28]  Michael Bailey,et al.  Risk prediction of hospital mortality for adult patients admitted to Australian and New Zealand intensive care units: development and validation of the Australian and New Zealand Risk of Death model. , 2013, Journal of critical care.

[29]  Alexander Binder,et al.  On Pixel-Wise Explanations for Non-Linear Classifier Decisions by Layer-Wise Relevance Propagation , 2015, PloS one.

[30]  Yan Liu,et al.  Deep Computational Phenotyping , 2015, KDD.

[31]  Sergey Ioffe,et al.  Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift , 2015, ICML.

[32]  Geoffrey E. Hinton,et al.  Deep Learning , 2015, Nature.

[33]  Andrew Zisserman,et al.  Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps , 2013, ICLR.

[34]  Kung Chen,et al.  Development of a daily mortality probability prediction model from Intensive Care Unit patients using a discrete-time event history analysis , 2013, Comput. Methods Programs Biomed..

[35]  Yan Liu,et al.  Interpretable Deep Models for ICU Outcome Prediction , 2016, AMIA.

[36]  Geoffrey E. Hinton,et al.  Learning representations by back-propagating errors , 1986, Nature.

[37]  G. Moody,et al.  Predicting in-hospital mortality of ICU patients: The PhysioNet/Computing in cardiology challenge 2012 , 2012, 2012 Computing in Cardiology.

[38]  Rich Caruana,et al.  Do Deep Nets Really Need to be Deep? , 2013, NIPS.