Alcidion Group Limited · ASX:ALC
Alcidion — Health Informatics (July 2016)
17 slides from Alcidion's conference presentation, Jul 26, 2016
Slide 1: Deep Learning for Patient Flow
For personal use only DEEP LEARNING FOR PATIENT FLOW MALCOLM PRADHAN, CMO
Slide 2: Overview
For personal use only OVERVIEW ▶ Why are smart machines are important for health care ▶ The emergence of deep learning ▶ Deep learning vs existing methods ▶ Some early results ▶ Practical tips on getting started ▶ Future directions
Slide 3: About Alcidion
For personal use only ABOUT ALCIDION ▶ Health informatics company with products in ▶ Patient flow & bed management ▶ Emergency Department ▶ Outpatient and referrals management ▶ SmartForms, clinical decision support (CDS) ▶ A health informatics approach ▶ Computers should play a more active role in
Slide 4: Background
BACKGROUND ▶ I started research into AI in the early 1990’s For personal use only ▶ Focused on decision theory and complex models (uncertainty in AI) ▶ Probabilistic networks based on knowledge and data ▶ 448 nodes, > 900 connections, > 90m probabilities
Slide 5: Smart Machines in Health Care
inflates the costs for seniors. SMART MACHINES Other studies have shown that a largeIN HEALTH portion CARE of health care costs is incurred once a person reaches an advanced age. As shown in Figure 1, public health care spending per person tends to be relatively high during the first year of life, a
Slide 6: Challenges in Patient Flow
For personal use only CHALLENGES IN PATIENT FLOW ▶ Increasingly complex patients ▶ Increased referrals to allied health and other specialties ▶ Higher resource utilization, difficult to predict ahead of time ▶ Current models ▶ Predict ED admissions and future admissions ▶ Challenges ▶ Predicting det
Slide 7: A (Very) Brief History of AI in Health
For personal use only A (VERY) BRIEF HISTORY OF AI IN HEALTH ‣ Expert level performance since 1970’s ‣ Difficult to integrate into workflows ‣ Clinical Decision Support for ‣ Not integrated with IT systems ‣ Diagnosis ‣ Brittle ‣ Management ‣ Difficult to maintain over time ‣ Safety ‣ Not easy to lo
Slide 8: The 3rd Wave of Neural Networks
For personal use only THE 3 RD WAVE OF NEURAL NETWORKS 1. Technical improvements to deal with deep networks 2. Large data sets e.g. ImageNet has >14m annotated images 3. GPU
Slide 9: CPU vs GPU
CPU VS GPU Titan X v2 For personal use only 1996 2013 GeForce GTX Titan X GeForce 980 Ti
Slide 10: What Can They Do?
For personal use only WHAT CAN THEY DO? ▶ Image description ▶ Text generation ▶ Self-driving cars ▶ Playing complex games ▶ Early work in demonstrating ▶ Radiology ▶ Histopathology ▶ Risk detection
Slide 11: Deep Network Intuition
For personal use only DEEP NETWORK INTUITION Lee, H. et. al. Convolutional Deep Belief Networks for Scalable Unsupervised Learning of Hierarchical Representations. ICML 09 www.rsipvision.com/exploring-deep-learning/
Slide 12: Why Do They Work?
For personal use only WHY DO THEY WORK? Latysheva N., Ravarani C. Misleading modelling: overfitting, cross-validation, and the bias-variance trade-off.
Slide 13: Methods
For personal use only METHODS ▶ Data generated from anonymised HL7 feeds ▶ Admissions, readmissions, ICDs, DRGs ▶ Later adding labs, radiology reports, OPD ▶ Multilayer perceptrons (MLP) ▶ Tested 2-5 layers, many variations ▶ Significant work in encoding categorical variables such as DRGs and ICDs ▶
Slide 14: Early Results
EARLY RESULTS Readmission For personal use only ▶ ▶ Detection of readmission < 30 days against ▶ Usually for COPD, Heart failure, Pneumonia, AMI, Total hip/knee arthroplasty ▶ On par with best published algorithms based on ICDs and LACE tool approx. 0.65 AUC (better than traditional models on this d
Slide 15: DL Issues for Health Care
University of Wyoming Cornell University University of Wyoming anguyen8@uwyo.edu DL ISSUES FOR yosinski@cs.cornell.edu HEALTH CARE jeffclune@uwyo.edu Abstract For personal use only ▶ Very data hungry eep neural networks (DNNs) have recently been ▶ Can’tperformance ving state-of-the-art provide hints
Slide 16: Practicalities
For personal use only PRACTICALITIES ▶ Large volumes of data required ▶ 100,000+ cases ▶ Non-image methods are still being developed ▶ Representation of categorical data ▶ Research moves ahead fast using a direct publish model (arxiv) ▶ Many experiments required over hours – days ▶ Configurations, e
Slide 17: Future Directions
For personal use only FUTURE DIRECTIONS ▶ Collaboration ▶ Aim to create Health Informatics Machine Learning group, currently discussing with UniSA, University of Adelaide ▶ Working to integrate more data sources e.g. community ▶ Improved models ▶ Further incorporation of clinical data such as labs ▶