Part 1: Wednesday, November 11, 2020 | 10:00 AM 1:00 PM (US ET)
Part 2: Wednesday, November 18, 2020 | 10:00 AM 1:00 PM (US ET)
Part 3: Wednesday, December 2, 2020 | 10:00 AM 1:00 PM (US ET)
Instructor:
Arvind Rao, PhD, Associate Professor, Department of Computational Medicine and Bioinformatics, Radiation Oncology, Biostatistics, Biomedical Engineering, The
University of Michigan, Ann Arbor
Healthcare with its rich variety and velocity of multi-modal data (from labs, clinic, pathology, wearables, and more) creates the need for a nuanced approach to the development, assessment and deployment of AI models, enabling transition from “data to knowledge” and from “knowledge to action”. These 3 workshops, geared to data scientists and healthcare professionals will provide an overview of AI methodologies, their application in healthcare domains, and some nuances to their utilization and deployment in your healthcare system.
The intended audience for these workshops is individuals in specification, development and decision-making responsibilities across the healthcare management and delivery landscape interested in the development and deployment of AI solutions within their healthcare system. AI engineers, data scientists, early adopter physicians & hospital leaders interested in AI for healthcare are invited to participate.
Part 1 Introduction to AI & ML in Medicine
Module 1 deals with basic principles in machine learning (ML). Concepts reviewed will include discussion of big data, and principles of grouping data (i.e. clustering) and predictive modeling. There will also be discussion of some common caveats in the development and application of ML algorithms.
Main topics covered will include:
- Introduction to AI/ML: What is AI, ML, data science?
- 5 Vs of big data and their meaning for AI
- Case studies of AI/ML in healthcare
- ML components:
-Preprocessing: Data cleaning, missing data, data transformations
-Data representation: Feature engineering (physics-based), dimension reduction (linear subspaces PCA vs. non-linear sub-spaces)
-Deriving meaning: Clustering & its anatomy, what choices underlie meaningful clustering, comparing various clustering algorithms
-Predictive modeling (regression/classification): Parametric vs. non-parametric, linear/logistic regression, LDA/QDA, SVM and random forests - Performance scoring from clustering: Internal and external validity, model selection
- Performance scoring of predictive models: Goodness-of-fit, classification metrics
- Machine learning caveats: Interpretability, GIGO, nested cross-validation, guarding against sample skew, train-test divergence, sample size
- Unsupervised learning: Introduction to deep learning; convolutional neural networks
Part 2 Advanced Concepts of AI & ML in Medicine
Continuing on the background built in Module 1, Module 2 delves into some advanced concepts around case studies in radiology, radiation oncology, and precision medicine. Specifically, ideas of multi-modal data integration, learning with structured data and quality aspects of data, models and inference will be discussed.
Main topics covered will include:
- Applications/case studies in radiology, and precision medicine
- Multimodal learning & data integration
- Distributed learning from large data repositories
- Learning from small amounts of data
- Ensuring data quality, model quality, and inferential quality/uncertainty
- Bias-variance tradeoffs
- Towards human-AI collaboration
- Concrete needs for successful AI deployment
Part 3 Towards Responsible AI & ML in Medicine
Module 3 will deal with concepts and rubrics towards the assessment and principled deployment of ML models in healthcare. This requires consideration of many key tradeoffs between model performance, quality and responsible deployment in practical decision making systems.
Main topics covered will include:
- Responsible AI practices
- Case study: Auditing of prostate segmentation & results
- Data quality: Quality of inputs, quality of outputs
- Model quality
- Inferential quality
- Bias assessment tools
- Reproducible ML
Instructor Biography:
Arvind Rao, PhD, Associate Professor, Department of Computational Medicine and Bioinformatics, Radiation Oncology, Biostatistics, Biomedical Engineering, The University of Michigan, Ann Arbor
Arvind Rao is an Associate Professor in the
Department of Computational Medicine and Bioinformatics, Radiation Oncology and Biomedical Engineering at the University of Michigan. Arvind received his PhD in Electrical Engineering and Bioinformatics from the University of Michigan, specializing
in transcriptional genomics, and was a Lane Postdoctoral Fellow at Carnegie Mellon University, specializing in image informatics. He was previously on the faculty in Bioinformatics and Computational Biology at the University of Texas MD Andersen
Cancer Center. His research group uses image analysis and AI/machine learning methods to mine and integrate different kinds of data sources for health applications like drug repurposing, genetic medicine, imaging diagnostics and health informatics.
He has taught AI courses in academia as well as for industry pharma and healthcare conferences.