We infer voxel‑level ischaemic burden directly from standard CCTA. A GPU‑accelerated pipeline segments cardiac anatomy, computes node‑level FFR along the coronary tree, projects vessel signals to branch‑weighted myocardial territories, and localizes perfusion deficits with a novel 3‑D GAN trained on multimodal ground truth.
We’re preparing a $3M pre-Series A to fund clinical validation, MDR/FDA clearance, and first reimbursed deployments in cardiology centers.
→ Inputs: standard Coronary CT Angiography (CCTA).
→ Segmentation: neural networks identify heart chambers, artery centerlines, and detailed lumen geometry.
→ FFR: stacked networks predict node‑level FFR along the coronary tree.
→ Territories: branch‑weighted Voronoi assignment projects vessel signals to myocardium.
→ Ischaemia: a novel 3‑D network localizes & quantifies voxel‑level ischaemic burden.
A fully GPU‑accelerated inference stack that transforms CCTA anatomy into actionable functional insights. Node‑level FFR supports lesion‑specific assessment, while voxel‑level perfusion maps quantify the ischaemic core and penumbra across territories.
Total: 230 patients — 200 development, 30 validation. Stratified by FFR severity bins (<0.70, 0.70–0.80, >0.80). No cross‑patient leakage.
Preprocessing (including scaling) is fit on the development set and applied to validation only. Metrics are aggregated at the patient level to mirror clinical deployment.
Protected by international patent application PCT/IB2023/059985. Proprietary methods cover multi‑stage anatomical inference, physiological territory mapping, and voxel‑wise ischaemia estimation.
Current POC data (investigational)
Sub‑0.11 FFR RMSE on held‑out patients indicates clinically meaningful discrimination across the FFR spectrum. Correlation within ROI reflects fine‑grained perfusion structure beyond heuristic boundaries. Reported territory‑level AUC/sensitivity/specificity link directly to lesion‑centric decisions.
Outputs assist, not replace, clinical judgment — providing interpretable overlays and patient‑level summaries for efficient review.
Desktop – Arteries: coronary centerlines with FFR gradient overlay.
Desktop – Ischaemia: perfusion deficit highlighting ischaemic region.
Web application: browser‑based visualization and annotation tools.
Dataset juxtaposition: perfusion and FFR inputs vs ground‑truth imaging.
Inference analysis: radius, FFR, perfusion, and inferred perfusion heatmaps.
We invite clinical and research partners for pilot deployments and multi‑center validation. For collaboration inquiries, contact us at research@aihexeract.com, Piotr Tomasinski, petertom@aihexeract.eu
This is investigational software not yet cleared for clinical use. All studies comply with data protection and IRB requirements. Outputs are intended for professional decision support under qualified supervision.
AI Hexeract sp. z o.o. is implementing a project financed by the BTM Investments sp. z o.o. Fund under the Smart Growth Operational Programme 2014–2020, Measure 1.3: R&D projects financed with the participation of venture capital funds, Sub-measure 1.3.1: Support for R&D projects at the pre-seed stage through proof-of-concept funds – BRIdge Alfa. Project title: Ischaemia Project objective: To develop a diagnostic tool for detecting coronary artery disease based on anatomical data. Public co-funding amount: PLN 880,000.
