KNOWLEDGE MANAGEMENT

AND DISCOVERY LAB

Tutorial KDD 2019

 

TUTORIAL - T3: Mining and model understanding on medical data

 

KDD 2019, Anchorage - from August 04 to August 08

Sunday August 4th, 2019 -  8:00 am - 12:00 pm (Tubughnenq 3, Level 2, Dena’ina)

 

Tutorialists: Panagiotis Papapetrou (Stockholm) and Myra Spiliopoulou (Magdeburg)

Abstract

Medical research and patient caretaking are increasingly benefiting from advances in machine learning. The penetration of smart technologies and the Internet of Things give a further boost to initiatives for patient self-management and empowerment: new forms of health-relevant data become  available  and  require  new  data  acquisition  and  analytics’  workflows.  As  data complexity and model sophistication increase, model interpretability becomes mission-critical. But what constitutes model interpretation in the context of medical machine learning: what are the questions for which KDD should provide interpretable answers?

In  this  tutorial,  we  discuss  basic  forms  of  health-related  data Electronic  Health  Records, cohort data from population-based studies and clinical studies, mHealth recordings and data from  internet-based  studies.  We  elaborate  on  the  questions  that  medical  researchers  and clinicians pose on those data, and on the instruments they use giving some emphasis to the instruments “population-based study” and“Randomized Clinical Trial”.We elaborate on what questions are asked with those instruments, on whatquestions can be answered from those data,  on  ML  advances  and  achievements  on  such  data,  and  on  ways  of  responding  to  the medical experts’ questions about the derived models.

PART 1: Introduction (BOTH)

1. The scope of the tutorial: tutorialists, structure, main topics

2. Introductory terms: What are patient data? Electronic Health Records (EHRs), social data, data collected in cohort studies

PART 2: EHRs and temporal abstractions (PANOS)

1.  Definition and examples of EHRs and EHR systems

2. Overview of the usage of EHRs globally

3. Predictive models on EHR data

4. Dealing with missing values in EHR variables

5. Survival analysis in EHRs

Slides here

PART 3: Learning from cohorts (MYRA)

1. Definition and examples of cohorts

2. Cohorts for clinical and population-based studies

3. Randomized clinical trials (RCTs)

4. Expert driven cohort refinement on EHR data

5. Cohort alignment for model validation

6. Expert inputs and what-if questions to models on cohorts

Slides here

PART 4: Deep learning and interpretability (PANOS)

1. Deep learning architectures for EHRs

2. Recurrent Neural Networks for diagnosis prediction

3. Deep learning with attention mechanisms

4. Interpretable model-specific methods for EHRs

5. Interpretable model-agnostic methods for EHRs

Slides here

PART 5: Learning from eHealth and mHealth data (MYRA)

1.  Using the internet for therapy, the example of iCBT

2.  Potential challenges and pitfalls in mHealth

3.  Momentary assessments and the promise of smart devices

4.  Learning from the data of mobile devices

5.  Monitoring the momentary assessments of patients

Slides here

PART 6: Conclusions (BOTH)

1.  Summary and challenges in learning

2.  Challenges of small data

3.  Challenges on reliability

4.  Challenges in involving the expert

5.  Challenges in model explainability

 

Target audience and prerequisites

This tutorial  is targeted to all  KDD  participants, with  particular  focus  group  being junior researchers  interested in machine for  health-related  data and on  how  to  convey  models  to experts. The main prerequisites for the participants concerns basic knowledge within the areas of data mining, machine learning, and databases. The audience is expected to be familiar with standard concepts and methods of machine learning. Such knowledge can be expected from KDD participants, including students.

  

Contact info of the tutors

Prof. Myra Spiliopoulou

Research Group on Knowledge Management and Discovery (KMD),

Faculty of Computer Science, Otto-von-Guericke-University Magdeburg,

PO Box 4120, 39016 Magdeburg, Germany

Email: myra _at_ ovgu [dot] de

URL: http://www.kmd.ovgu.de/Team/Academic+Staff/Myra+Spiliopoulou.html

 

 

Prof. Panagiotis Papapetrou

Data Science group

Department of Computer and Systems Sciences

PO Box 7003, 164 07, Stockholm, Sweden

Email: panagiotis _at_ dsv [dot] su [dot] se

URL: http://people.dsv.su.se/~panagiotis/