Healthcare IT · PhilipsR&D Intern · Philips Innovation CampusAug 2018 – Jul 2019
Predictive PACS Latency Monitoring Pipeline
Built an ELK Stack-powered predictive monitoring pipeline for PACS (Picture Archiving and Communication System) that forecasts medical image transfer latency using decision-tree models — enabling proactive IT triage before clinicians feel impact.
Proactive
Monitoring
Predictive PACS telemetry
Data-Driven
Decision-Making
ETL + visualization
// THE PROBLEM
Problem formulation
Medical image transfer latency degrades clinician workflow and patient outcomes. Reactive alerting only fires after users complain; the team needed predictive visibility into latency trends.
// DATA FOUNDATIONS
Data engineering & ingestion
- Developed ETL pipelines ingesting PACS network telemetry logs
- Elasticsearch indexing of transfer events, file sizes, and network metrics
- Logstash transformations normalizing heterogeneous log formats
- Kibana dashboards for operational visibility
// MODELING STRATEGY
Mathematical & algorithmic approach
- Decision Tree models predicting image transmission latency
- Feature selection from payload size, modality, time-of-day, and network path
- Regression evaluation against holdout transfer events
// EXPLAINABILITY
Interpretability & governance
- Tree-importance ranking of latency drivers across modalities
- Rule subsets identifying high-risk transfer patterns
- Kibana visualizations making model outputs legible to ops teams
// PRODUCTION
Deployment & integration
- ELK stack monitoring deployed for PACS infrastructure
- Proactive alerting thresholds derived from model predictions
- Extraction, preprocessing, and visualization pipelines supporting data-driven decisions
// TECH STACK
ELK StackDecision TreesETL PipelinesPythonElasticsearchKibana