Predetermined Change Control for Adaptive Medical AI Systems
Adaptive medical AI should be governed as planned device change, not informal model maintenance.
Obuseff Journal
Publications
Research notes, essays, and technical observations on medical AI, clinical data architecture, and the infrastructure required to move biomedical work from prototype to dependable system.
Clinical Data Fabric as an Auditability Layer
Clinical data platforms should preserve provenance, consent, and transformation history as inspectable system evidence.
Why Medical AI Deployment Requires a Risk File
A deployable AI system should connect hazards, controls, validation evidence, and post-deployment monitoring in one traceable file.
Human Oversight Is a System Requirement, Not a UI Feature
Clinical oversight must be designed as authority, workflow, escalation, and evidence, not as a decorative approval button.
Dataset Lineage in Genomic and Clinical AI Pipelines
Genomic AI depends on defensible lineage from consent and accession through preprocessing, feature construction, and model use.
HIPAA-Aligned Architecture for Biomedical Compute Environments
Protected health information requires administrative, physical, and technical safeguards translated into practical compute controls.
Software Verification in Medical Device AI
Verification should prove that the software implements specified behavior before validation argues clinical usefulness.
Model Monitoring After Clinical Deployment
Clinical model monitoring should detect drift, missingness, misuse, and silent workflow change before they become patient risk.
Federated Learning for Hospitals: Governance Before Aggregation
Federated learning reduces raw data movement, but it does not remove consent, security, bias, or accountability requirements.
LIMS, PACS, EHR, and Research Data Interoperability
Interoperability is not message exchange alone; it is preservation of clinical meaning across systems with different duties.
Variant Interpretation Platforms and Evidence Traceability
Genetic interpretation systems should expose evidence version, rule basis, assertion confidence, and review history.
Gene Therapy Data Pipelines from Bench to Clinical Review
Cell and gene therapy programs need data pipelines that preserve assay context, batch identity, and CMC evidence.
MLOps Controls for Biomedical Foundation Models
Biomedical foundation models require model registries, evaluation gates, dataset restrictions, and deployment constraints.
Consent-Aware Data Routing in Biotech Platforms
Biomedical platforms should enforce consent at the data-plane level, not only through policy documents.
Knowledge Graphs for Drug Discovery Evidence Management
A discovery graph should preserve claim provenance, evidence grade, biological context, and contradictory findings.
Digital Twins in Systems Medicine: Validation Boundaries
A medical digital twin must define what it predicts, what it cannot infer, and which clinical decisions remain out of scope.
Synthetic Data for Medical Software Testing
Synthetic data can improve test coverage, but it cannot replace clinical validation or obscure provenance duties.
Secure Enclaves for Genomic Research Collaboration
Controlled-access genomic research requires environment-level controls for identity, export, logging, and incident response.
Incident Response for AI-Assisted Clinical Decision Support
AI incidents should be handled as clinical, software, data, and governance events at the same time.
From Research Prototype to Regulated Medical Software
The transition from research code to medical software is a change in evidence, accountability, and lifecycle control.