public

Views 96

Implement Machine Learning in Medicine

published by MaHarrisoncheo. on. ca

Want to create a visual like this?

Get Started

Infrastructure and resources that can be difficult to access



Challenges in ensuring data security and privacy

Poorly performing mathematical models

SUPPORT COMPLEX DECISION MAKING

AUTOMATE MUNDANE TASKS

IMPROVE WORKFLOWS, PROCESSES & QUALITY IMPROVEMENT INITIATIVES

BETTER RESOURCE ALLOCATION & GUIDE STAFFING DECISIONS

Considerations

How to...

IMPLEMENT MACHINE LEARNING IN MEDICINE


The process of developing systems that learn from data to recognize patterns and make accurate predictions of future events. ML can also be used for explanation of phenomena.


STEPS TO FOLLOW TO IMPLEMENT ML

There are 3 phases to ML development:
       a. Exploration
       b. Solution Design
       c. Implementation and Evaluation

3. Understand and decompose the problem and how it can be solved

2. Is ML a feasible solution to this problem?

  • Availability of data



4. Decide on how to measure success, set goals



1. Identify the problem:

  • Based on impact and priority (how solving it will impact “outcomes”)

  • Compose team





MACHINE LEARNING DEVELOPMENT

Develop an easy to use tool




Consider RCT for assessing the impact of the ML solution as intervention.


Time series and matched cohort designs are acceptable options.

IMPLEMENTATION CONSIDERATIONS

EVALUATION


Design a ML system with implementation in mind

(easy to implement)


Consider integration with existing work-flow

Consider phased implementation


Engage end-users



Implement iterative evaluation

(training-assessment-retraining-reassessment)

Use a “silent trial”


HOW CAN IT HELP IN HEALTHCARE?


WHAT ARE POSSIBLE BARRIERS TO IMPLEMENTATION?

WHAT IS MACHINE LEARNING (ML)?


5. Consider generalizability:

  • Your institution vs general

  • Different aspects and data needs and considerations

Consider sustainability, drift over time, possible retraining

Difficulty integrating tools into existing workflows

Low acceptance of ML solutions by clinician users

Uncertainty about how to evaluate them



(i.e. large, real-time clinical data sets, technical skills in data science, computing power and clinical informatics infrastructure)