MD-SpeakerDeck.pdf

Transcript

1. Anthony Herr Product Manager Lifesaving Technology in an Open Source Framework Using AI/ML to Assist with Medical Diagnosis
2. 2 1 2 3 4 What is Edge Computing Business Problem Considerations Solution Agenda
3. STRICTLY INTERNAL ONLY End-user Premises Edge Provider Edge Provider or Enterprise Core “Last mile” Footprint Scale Edge Gateway & Endpoint Regional Data Center Infrastructure Edge Provider Far Edge Provider Access Edge Provider Aggregation Edge Core Data Center 3 Many different edges * Edge computing = Fog computing (there is no real difference other than marketing) Device or Sensor Centralized Management and Control Telco Services, CDNs, Regional Cloud Data Analytics Data Collection Data Aggregation
4. Accelerating medical diagnosis using condition detection in medical imagery with AI/ML at medical facilities
5. Tackling the challenges ▸ Faster diagnosis ▸ Give more time to practitioners ▸ Increase throughput of patients ▸ Increase efficiency of the medical facility
6. Pipeline Architecture Pneumonia risk detection ML model S3 Anonymize Image New object event kafka ‘producer’ xxx ODF S3 Knative Eventing kafka ‘consumer’ Knative Serving OpenShift Serverless Function (on-demand) <80% confidence diagnosis? Y Notify Medical Staff Predict: pneumonia probability (0-100%) OpenShift Serverless OpenShift Data Foundation Object Bucket Notifications Anonymized images sent to central data center for model re-training Medical facility Central Lab 1 2 3 4 5
7. 1. Automatically categorize x-rays into high/low/unknown risk of pneumonia. 2. Notify medical practitioners in real-time about high-risk pneumonia patients. 3. Submit all x-ray categorization results for human review and approval. 4. Anonymize x-rays with low-confidence ML inference results and send to data center for model retraining, with no PII. At each medical facility (on-demand): At central lab (batch): 1. Retrain pneumonia risk detection ML model for improved accuracy. 2. Report on trend analysis of ML pneumonia risk prediction accuracy. Solution - Data Pipeline 1. High-risk of pneumonia 2. Low-risk of pneumonia 3. Low-confidence in ML inference Pneumonia risk inference from ML model
8. Model training - Shared data science platform Demo - http://youtube.com/watch?v=BDgYgi24jXo
9. Model in production - Real time dashboard
10. Edge Site …..n Edge Site 2 Edge Site 1 Data Streaming Data ingestion and routing Edge AI Application ML trained application for intelligent edge Local Dashboard Display real time info to line operators Edge Management Apply cluster and application config changes Core Data Center Data Streaming Sensor data ingestion DevOps Automation Continuous integration/ continuous deployment Edge Management Cluster and application lifecycle management Data Lake Data store for model training ML Model Training Data science and ML model training Consistent Distributed Architecture Code Management Repository for code, config and deployment blueprints Image Store Container registry for images and ML models
11. Edge Site …..n Edge Site 2 Edge Site 1 Data Streaming Edge AI Application Local Dashboard Edge Management Core Data Center Data Streaming DevOps Automation Edge Management Data Lake ML Model Training Consistent Distributed Architecture Code Management Image Store
12. GitOps: What is it exactly? A developer-centric approach to Continuous Delivery and infrastructure operation Treat everything as code Git is the single source of truth Operations through Git workflows
13. Validated Patterns: Core Technologies Work Together to Orchestrate Workloads https://validatedpatterns.io/
14. 1 - http://youtube.com/watch?v=BDgYgi24jXo 2 - https://www.redhat.com/architect/portfolio/detail/6 3 - https://hybrid-cloud-patterns.io/medical-diagnosis/ Where can I find more information? 14 Review the use case architecture 2 Learn more about the Medical Diagnosis Validated Pattern3 Watch a demo of this use case 1
15. linkedin.com/company/red-hat youtube.com/user/RedHatVideos facebook.com/redhatinc twitter.com/RedHat 15 Red Hat is the world’s leading provider of enterprise open source software solutions. Award-winning support, training, and consulting services make Red Hat a trusted adviser to the Fortune 500. Thank you
16. 16 Let’s look at the Medical Diagnosis pattern today https://validatedpatterns.io/ https://validatedpatterns.io/patterns/medical-diagnosis/