2.4 - Addressing Gaps and Challenges for Successful BDT Implementation - Part 2

Back to 2024 Agenda

This session will dive deeper into BDT components and present tools to address specific BDT development challenges.

Charge to Speakers: Present your topic with these BDT components in mind:

  1. Fit-for-Purpose BDT problems
  2. Modularity/Interoperability/Systems of Systems mindset
  3. Sustainability, regulatory issues
  4. Data/Knowledge systems

Deeper Dive into BDT Pitfalls:

Each speaker will present a 10-minute talk, followed by five minutes for questions

Describe how your tools/resources...:

  • fit biomedical problems at the population level versus at the individual level. What math/stats/computational methods can be used to couple population to individual levels?
  • incorporate mechanistic modeling versus data-driven modeling. What math/stats/computational methods will be used to reflect the biological mechanisms of the system?
  • integrate mathematical, statistical, and computational methods that represent unique biomedical assumptions, factors and features. Are these assumptions realistic/feasible?

Speaker Bios:

Natalia Trayanova

Ralph Smith

Maria Eduarda Montezzo Coelho

Barbara Evans

Moderator Bio:

Julia Berzhanskaya

Reference: NASEM DT components from Day 1:

  • Fit-for-Purpose BDT
  • Verification, Validation and Uncertainty Quantification (VVUQ)
  • Physical assets/data collection/sensors
  • Mathematical and statistical foundations for BDT
  • Virtual to Physical Control Algorithms/ Expert in the loop
  • Ethical, security issues
  • Team Science Approach, Governance

Presentations:

BDT_Smith_Final (nih.gov)

https://www.imagwiki.nibib.nih.gov/sites/default/files/2024-10/d2_1_1045am_berzhanskaya_-_bdt_deep_dive_nt.pptx 

https://www.imagwiki.nibib.nih.gov/sites/default/files/2024-10/d2_4_1045am_evans_teaming4bdt.pptx 

https://www.imagwiki.nibib.nih.gov/system/temporary?file=d2_3_1045am_eduarda_coelho_inl.pptx 

Materials:

Handout2_BDT Components.pdf292.29 KB

Comment

Your name
Mitchel Colebank
Comment

Fantastic talk, Natalia!

Could you comment on how incorporate possible uncertainties in the geometry reconstruction and measurement noise in the ECG signals? I imagine these influence your predicted interventions?

- Mitchel

Submitted by Mitchel Colebank (not verified) on Tue, 10/01/2024 - 11:02

Your name
Emek
Comment

What are fundamental differences between a DT for an engineered system vs an evolved system? How does these differences happen inform you that we can and can't do for but?

Submitted by Emek Demir (not verified) on Tue, 10/01/2024 - 11:34

Your name
Jerry Myers
Comment

Great topic. Is this interpretation of data use also pertain to "models and simulation" in general and not just digital twins specifically?

Submitted by Jerry Myers (not verified) on Tue, 10/01/2024 - 11:39

Your name
Carlos F Lopez
Comment

I really like the use of Bayesian inference for the parameter searches, but as these systems get larger and multidimensional, the choice of priors and the parameter covariance becomes important. How would you address that in the context of a digital twin?

Submitted by Question for Ralph (not verified) on Tue, 10/01/2024 - 11:48

Your name
Elena Sizikova (OSEL/FDA)
Comment

Do there exist VVUQ methods that evaluate the ability of the BDT to extrapolate outside of the information that was used to create the BDT?

Submitted by Elena Sizikova… (not verified) on Tue, 10/01/2024 - 12:03

Your name
Tina Morrison (FDA)
Comment

Thank you Ralph for your talk on UQ.  You made a statement, AI can be used to support the parts of the model we dont trust?  I would be skeptical to use AI/ML to "augment/support" a part of the model that you dont understand.  Can you elaborate more precisely what you mean?

Submitted by morrisontm on Tue, 10/01/2024 - 14:36

Table sorting checkbox
Off