1 of 38

Application of AI in Data Fusion for Improving Clinical Decision Making

Rohit Jain, nference

Michael Chan, NDI

​

​

1

Information Classification: General

#DiscoverEngineerBuild #IMEWest

2 of 38

Rohit Jain

Head of Data Science, nference

​

rjain@nference.net | nference.com

​

Rohit Jain is an Applied Mathematician and Data Scientist who holds a PhD in Mathematics from the University of Texas, Austin. His experience includes research in Partial Differential equations, Image Processing at NASA and building ML Applications in E-Commerce. Rohit is currently at nference, a company focused on AI and ML for biomedical applications.

Michael Chan

​

Director Sales and Marketing, NDI

​

mchan@ndigital.com| ndigital.com

Michael Chan is an Engineer and MBA with over 25 years of experience in the technology field ranging from software, to communications, to big data and applications with Artificial Intelligence and Machine Learning. Michael currently is the Director of Sales and Marketing at NDI, a company focused on applying its navigation technology in the medical device market.

2

Information Classification: General

#DiscoverEngineerBuild #IMEWest

3 of 38

  • Introduction to AI
  • AI in Healthcare
  • AI Use in a Clinical Setting
  • Clinical Examples
  • Considerations for AI Adoption in a Clinical Setting
  • Conclusions

3

Information Classification: General

#DiscoverEngineerBuild #IMEWest

4 of 38

Introduction to AI

4

Information Classification: General

#DiscoverEngineerBuild #IMEWest

5 of 38

Breaking down the terminology

5

Information Classification: General

#DiscoverEngineerBuild #IMEWest

6 of 38

Breaking down AI

http://ai.stanford.edu/~latombe/cs121/2011/slides/A-introduction.pdf

6

Information Classification: General

#DiscoverEngineerBuild #IMEWest

7 of 38

Breaking down Machine Learning

https://towardsdatascience.com/what-are-the-types-of-machine-learning-e2b9e5d1756f

7

Information Classification: General

#DiscoverEngineerBuild #IMEWest

8 of 38

Supervised Learning

https://medium.com/@metehankozan/supervised-and-unsupervised-learning-an-intuitive-approach-cd8f8f64b644

Typically for:

​

Classification

​

Regression

8

Information Classification: General

#DiscoverEngineerBuild #IMEWest

9 of 38

Unsupervised Learning

https://medium.com/@metehankozan/supervised-and-unsupervised-learning-an-intuitive-approach-cd8f8f64b644

Typically for:

​

Clustering

​

Dimensionality Reduction

​

Generative Models

9

Information Classification: General

#DiscoverEngineerBuild #IMEWest

10 of 38

Reinforcement Learning

https://www.sigmoid.com/blogs/introduction-to-reinforcement-learning/

Typically for:

​

Navigation

​

Skill Acquisition

​

Game AI

10

Information Classification: General

#DiscoverEngineerBuild #IMEWest

11 of 38

General ML Development Workflow

11

Information Classification: General

#DiscoverEngineerBuild #IMEWest

12 of 38

Examples of AI Use in Industry

Recommendation System

Route Optimization

Autonomous Vehicles

Spam Filtering

Autonomous Rovers

12

Information Classification: General

#DiscoverEngineerBuild #IMEWest

13 of 38

Why AI use in Industry?

The AI-Industrial Complex

https://blog.ml.cmu.edu/2020/08/31/1-domain-knowledge/

13

Information Classification: General

#DiscoverEngineerBuild #IMEWest

14 of 38

Amazon

The Recommendation Engine is based on various data signals collected throughout a shopper’s experience.

Combining these two mainstream solutions in E-Commerce:

​

Item-to-Item Collaborative Filtering

​

Rather than matching user to similar customers, item-to-item collaborative filtering matches each of the user’s purchased and rated items to similar items and combines them into a recommendation list.

https://www.linkedin.com/pulse/amazons-recommendation-engine-secret-sauce-mario-gavira/

​

http://www.cs.umd.edu/~samir/498/Amazon-Recommendations.pdf

14

Information Classification: General

#DiscoverEngineerBuild #IMEWest

15 of 38

Uber

Uber’s ML System updates the app with conditions in every route and suggests the fastest route to the driver.

Routing Algorithm: Using the correct data structure for the problem

https://www.uber.com/blog/engineering-routing-engine/

15

Information Classification: General

#DiscoverEngineerBuild #IMEWest

16 of 38

AI in Healthcare

16

Information Classification: General

#DiscoverEngineerBuild #IMEWest

17 of 38

The Case for AI Use in Healthcare

​

  • Data intense - collection and analysis resulting in long lead times for research
  • Multiple data formats, sources and siloed information leading to difficult data integration and consolidation
  • Accurate updated clinical and patient information needed to facilitate learning, training, and diagnostics

​

​

17

Information Classification: General

#DiscoverEngineerBuild #IMEWest

18 of 38

Examples of AI Use in Healthcare

Medical Diagnosis

Drug Discovery

Healthcare Data Management

Robotic Surgery

18

Information Classification: General

#DiscoverEngineerBuild #IMEWest

19 of 38

AI Use in a Clinical Setting

19

Information Classification: General

#DiscoverEngineerBuild #IMEWest

20 of 38

Issues with Data Consolidation In Surgical Procedures

  • Data Creation - comprised of heterogeneous types, formats, and electronic languages with no “Universal Language”
    • Narrative, textural data (charts, histories, clinician notes)
    • Numerical measurements (lab results, vital signs, physical measurements)
    • Recorded signals (ECG, anesthesia, system pressures)
    • Image / visualization (Fluoroscopy, ultrasound, CT, MRI, Xray, EP Mapping)
  • Clinician Fatigue - Without consolidation or “fusion”, data must be displayed on separate outputs for the physician to decipher in real- time
  • Scalability—unstructured data such as MRIs, CT scans, etc. grows by massive amounts requiring varied storage locations and techniques

20

Information Classification: General

#DiscoverEngineerBuild #IMEWest

21 of 38

Benefits of AI Use in Surgical Procedures

  • Enhancement of precursory diagnosis-based on data from an entire patient library to select optimal treatments
  • Pre-procedure planning – Helping operators’ understanding to predict device interaction with patient anatomy
  • Image acquisition-integrating current imaging and past patient data from remote locations
  • Data fusion and data integration from diverse sources within the operating suite
  • Intraoperative body system guidance for navigation with targeting and risk avoidance

21

Information Classification: General

#DiscoverEngineerBuild #IMEWest

22 of 38

Clinical Example: Real Time Detection of Arrhythmia

22

Information Classification: General

#DiscoverEngineerBuild #IMEWest

23 of 38

3D Electroanatomical Mapping

  1. 3D Electroanatomical Mapping (EAM) Systems were first proposed in the 1990s to assist ablation procedures to correct cardiac rhythm disturbances.
  2. Mapping Systems display the position of catheters in real-time along with electrophysiological information of the cardiac chamber
  3. EAM use is consistently associated with reduced fluoroscopy time, radiation dose, and procedure time.

23

Information Classification: General

#DiscoverEngineerBuild #IMEWest

24 of 38

Clinical example: �Real-Time Detection of Patient Rhythm

24

Information Classification: General

#DiscoverEngineerBuild #IMEWest

25 of 38

Clinical example: �Real-Time Detection of Patient Rhythm: VT Map

Activation Mapping

​

  1. Entire path of activation during re-entry circuits or focal sources may be seen on the anatomical shell in 3D.
  2. Critical Assumption:�Every point used to build the activation map was acquired when the patient was in the same rhythm!
  3. Accuracy of maps is dependent on the expertise of the Clinical/Tech Team
  4. AI provides a data driven approach to building accurate maps

https://www.sciencedirect.com/science/article/pii/S1875213618300901

25

Information Classification: General

#DiscoverEngineerBuild #IMEWest

26 of 38

Clinical Example: Navigation and Data Fusion

26

Information Classification: General

#DiscoverEngineerBuild #IMEWest

27 of 38

EP Sample Data – EM Position Sensors

  • Measurement technology that uses electromagnetic fields to track miniature sensors in 3D space
  • Sensors integrated into EP instruments (e.g. catheters) to track location in real time for mapping and ablation
  • Information provided
    • Location
    • Orientation
    • Timing
  • Affected by metal distortion

​

27

Information Classification: General

#DiscoverEngineerBuild #IMEWest

28 of 38

Demo

28

Information Classification: General

#DiscoverEngineerBuild #IMEWest

29 of 38

EP Sample Data – Impedance Sensors

  • Measurement volume based on voltage and distance
  • Any piece of metal can be used
    • Size is not a factor
  • Not affected by Metal Distortion
  • Potential shift/drift over time
    • Fluid loading
    • Chemicals
    • Body size
    • Cabling
    • Temperature

​

29

Information Classification: General

#DiscoverEngineerBuild #IMEWest

30 of 38

Consolidating Data

30

Information Classification: General

#DiscoverEngineerBuild #IMEWest

31 of 38

Considerations for AI Adoption in a Clinical Setting

31

Information Classification: General

#DiscoverEngineerBuild #IMEWest

32 of 38

Use of AI in the operating room

32

Information Classification: General

#DiscoverEngineerBuild #IMEWest

33 of 38

AI/ML Platform

  • Connect N Data streams
  • Time Synchronize Streams
  • Based on EP Application in procedure identify data streams needed
  • Apply Analytics Engine
  • Integrated Visualization in Procedure

​

33

Information Classification: General

#DiscoverEngineerBuild #IMEWest

34 of 38

AI/ML Platform

  • Connect N Data streams
  • Time Synchronize Streams
  • Based on EP Application in procedure identify data streams needed
  • Apply Analytics Engine
  • Integrated Visualization in Procedure

​

Data Integration and Standardization

34

Information Classification: General

#DiscoverEngineerBuild #IMEWest

35 of 38

AI/ML Platform

  • Connect N Data streams
  • Time Synchronize Streams
  • Based on EP Application in procedure identify data streams needed
  • Apply Analytics Engine
  • Integrated Visualization in Procedure

​

Data Integration and Standardization

AI and Analytics Engine

35

Information Classification: General

#DiscoverEngineerBuild #IMEWest

36 of 38

AI/ML Platform

  • Connect N Data streams
  • Time Synchronize Streams
  • Based on EP Application in procedure identify data streams needed
  • Apply Analytics Engine
  • Integrated Visualization in Procedure

​

Data Integration and Standardization

AI and Analytics Engine

Application and Output

36

Information Classification: General

#DiscoverEngineerBuild #IMEWest

37 of 38

Conclusions

  • AI successfully implemented in various industries
  • With the explosion of data in the Healthcare industry, AI can be a valuable tool to help with:
    • Turning data into useful information for speed and accuracy for decision making (e.g. Surgical procedures)
    • Shortening the time to uncover new innovations and discoveries
    • Providing physicians with more confidence in their diagnosis
    • Democratizing patient care and improving outcomes
  • An AI platform provides an important foundation in providing a systematic approach to addressing various data streams to fuse and provide valuable information

​

37

Information Classification: General

#DiscoverEngineerBuild #IMEWest

38 of 38

—

​

QUESTIONS?

​

Thank you!

​

38

Information Classification: General

#DiscoverEngineerBuild #IMEWest