https://github.com/ashwinpn/domain-expertise-medical-data-

Deep Learning and AI Domain Expertise: Medical Data - EHR [Electronic Health Records], Genomes Sequencing, Neuroscience, Biomedical.
https://github.com/ashwinpn/domain-expertise-medical-data-

Last synced: about 5 hours ago
JSON representation

Deep Learning and AI Domain Expertise: Medical Data - EHR [Electronic Health Records], Genomes Sequencing, Neuroscience, Biomedical.

• Host: GitHub
• URL: https://github.com/ashwinpn/domain-expertise-medical-data-
• Owner: ashwinpn
• Created: 2021-02-03T03:11:24.000Z (almost 4 years ago)
• Default Branch: main
• Last Pushed: 2021-02-04T17:13:16.000Z (almost 4 years ago)
• Last Synced: 2023-10-19T22:45:52.978Z (about 1 year ago)
• Size: 3.91 KB
• Stars: 1
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Domain-Expertise-Medical-Data-

Deep Learning and AI Domain Expertise: Medical Data - EHR [Electronic Health Records], Genomes Sequencing, Neuroscience, Biomedical.

# List of Research Papers

- 1] Mitchell, M., Wu, S., Zaldivar, A., Barnes, P., Vasserman, L., Hutchinson, B., Spitzer, E., Raji, I.D. and Gebru, T., 2019, January. Model cards for model reporting. In Proceedings of the conference on fairness, accountability, and transparency (pp. 220-229). Link: https://arxiv.org/abs/1810.03993.

# Deep Learning for Biomedical Data Course [Taught at NYU Langone]

## Week 1 Notes 

1] Reducing Medical Errors - Focus on model performance [how that translates to real world scenarios] rather than just model accuracy.

2] In the US, the most common causes of death [non-accidental, non-natural, caused due to a disease] are - Heart disease, Cancer [Both account for >30% of the deaths in the states].

3] Cancer cells burn sugar quite a lot [Thus, using this causal effect we can detect cancer cells by detecting parts of the body with high sugar depletion rate]. 

4] Alzheimer’s - have chemical trackers for processing.

5] Invasive - Surgery involved, Non-Invasive - No surgery, only (say) scanning.

6] EHR data needs to be processed as Time Series Data [Hence, Forward Chaining,  not K-fold cross validation].

Q1] Representation Learning - How do we go from these diverse modalities to an outcome. Also, Functional Forms - which form of the data would be the most optimal for the learning process.

Q2] Bias in data - think deeply about the data-collection criteria.