Shiva Amiri, CEO, BioSymetrics at MLconf Seattle 2017
summary
A deployable pipeline iterates hundreds of preprocessing and normalization parameter sets and uses real-time model building to reveal and minimize site effects across scanners and protocols, which initially produce opposite connectivity results on the ABIDE cohort of ~1,000 subjects and a 400–500-subject subset. Iterating parameters yields larger gains than swapping classifiers, and selecting parameter sets that reduce site bias improves autism prediction. Adding genomics to MRI (combined MRI+genetics improved accuracy) further raises classification performance.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 0:45Biomedical data challenges
- 3:14Pre-processing in ML workflows
- 7:10Precision medicine questions
- 7:32Autism imaging study
- 12:29Bias in autism diagnosis
- 17:56Removing bias and integrating data
- 20:34Summary
related talks
20mDiagnostic Medical application of ML
20mAdvancing Epigenetic Predictors with Scalable Machine Learning
03h 51mMachine Learning Summit
42mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
30mMachine Learning: A New Approach to Drug Discovery
17mShiva Amiri, Chief Product Officer, RTDS Inc. @ MLconf SEA
49mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
20mFrom promise to impact with gen AI in healthcare & life sciences