OncoSphere AI is built on a parallel reasoning framework. Rather than a single generic model, it runs a coordinated ensemble of eight domain-specialized agents. Each agent independently analyzes the patient’s structured and unstructured data against its own specialty-specific guidelines and contemporary peer-reviewed literature. Their outputs are then synthesized into one integrated, guideline-anchored recommendation for the physician to review.
The workflow
- 1. Ingest. Paste a case summary or upload records (pathology, imaging, genomics). OncoSphere AI structures the heterogeneous, longitudinal data into a clean case.
- 2. Reason in parallel. Eight specialty agents analyze the case simultaneously, each against its guidelines and the latest literature.
- 3. Synthesize. The platform assembles the agents’ findings into an integrated MDT recommendation, with alternatives and areas of uncertainty made explicit.
- 4. Verify. Every section surfaces its citations and notes on uncertainty, so the physician can trace each point to its source and decide.
The eight agents
Medical Oncology
Systemic therapy options, NCCN-anchored regimen selection, and rationale tied to the patient’s stage, biomarkers, and prior lines.
Radiation Oncology
Indications, target/technique considerations, and follow-up, aligned to guideline criteria.
Surgical Oncology
Resectability assessment and procedure considerations against NCCN/AJCC criteria.
Pathology
Histology, IHC, and molecular findings — reported exactly as documented, including negatives that rule out targeted therapy.
Radiology
Imaging interpretation, measurements, and AJCC/RECIST-aligned staging.
Pharmacy
Dosing, preparation/administration, interactions, and supportive-care guidance.
Clinical Research
Trial concepts and eligibility considerations matched to the case.
Nurse Navigation
Care-coordination checklist and intake worksheet to keep the plan moving.
Adaptive by design
System telemetry from the ASCO 2026 validation showed adaptive compute behavior: the platform autonomously allocated 1.73× more reasoning to complex, refractory cases than to standard scenarios — spending effort where the case warrants it.