Summary
Dyania Health says its Synapsis AI system analysed records from more than 7,800 aortic valve replacement patients to help build a model of valve deterioration and reintervention risk. NVIDIA’s account describes the model’s task but gives no performance results.
Dyania Health says its Synapsis AI system reviewed medical records from more than 7,800 aortic valve replacement patients to help build a model predicting valve deterioration and the risk of needing another intervention. The study covered patients between 2005 and 2021, and assessed risk at three, five and ten years after surgery, according to an account published by NVIDIA.
The work illustrates how AI can be used to search long-running patient records for evidence relevant to a specific clinical question. The account describes a model-building effort; it does not report prediction accuracy or patient outcomes.
Finding signals across years of records
Hospital records contain information in many forms, including doctors’ notes, operative reports, imaging and pathology reports, laboratory results and discharge summaries. Synapsis AI is designed to connect findings across those records, including information recorded years apart. Dyania describes its outputs as clinical signals linked back to the source material, so care teams can examine the supporting evidence.
For the valve study, the system analysed each patient’s record, including echocardiograms added in the years after surgery. An echocardiogram uses ultrasound to show the heart’s structure and function; here, those later scans formed part of the information used to develop the deterioration and reintervention-risk model. NVIDIA says the platform completed chart review across the full cohort, a task that would have taken clinicians years to do manually.
The workflow is presented as support for clinicians, not a substitute for their judgement. The account says clinicians review the evidence and make decisions, and that signals surfaced by Synapsis AI must be confirmed and acted on by care teams.
Processing records inside a health system
Dyania says Synapsis AI runs on NVIDIA-accelerated computers within a health system’s own firewall, allowing records to be processed onsite. The company also says it uses NVIDIA NeMo libraries to build and train its clinical models, then NVIDIA TensorRT and TensorRT-LLM to optimise them for its GPU servers.
The described study establishes the intended use of the system and the size and time span of the patient cohort. NVIDIA’s account provides no model-performance measures or patient outcome results, so the predictive value of the model cannot be assessed from the information reported.