Moonshot Cardio

Confidential Research Preview

Patent pending • Not for distribution
Moonshot Cardio

AI-Powered Cardiovascular Screening

Detecting cardiovascular risk from medical images that already exist — starting with the millions of DXA scans performed every year where aortic calcification is visible but never reported.

12 Patents Filed
95K UK Biobank DXA Dataset
15M+ US DXA Scans / Year
The Problem

A Screening Opportunity Hiding in Plain Sight

Abdominal aortic calcification is visible on virtually every lateral spine DXA image. It is a validated, independent predictor of cardiovascular events. Yet it is almost never reported.

15M+
DXA Scans / Year (US)
<1%
Report AAC Findings
AAC-24
Kauppila Score Standard

Why It's Missed

Manual Kauppila scoring is slow (5–10 min per image), subjective, and outside the workflow of DXA technologists. Radiologists reading bone density don't look for it. The finding exists in the image — no one extracts it.

Why It Matters

AAC severity independently predicts myocardial infarction, stroke, and cardiovascular mortality. Studies show AAC scores correlate with coronary artery calcium (CAC) and add prognostic value beyond Framingham risk factors. It's a free look at cardiovascular risk from a scan ordered for something else.

Immediate Project

Automated AAC Scoring from DXA

An AI model that automatically detects and quantifies abdominal aortic calcification from lateral spine DXA images — turning every VFA scan into a cardiovascular risk assessment.

Model Architecture

  • Binary classification: AAC present / absent (target AUC >0.85)
  • Regression: Kauppila AAC-24 score prediction (targeting inter-rater agreement level)
  • Risk stratification: Low / Moderate / High based on established AAC thresholds
  • Interpretability: Grad-CAM overlays with uncertainty quantification

Training Data

  • UK Biobank Field 20158 — ~95,000 lateral spine DXA images (GE Lunar iDXA)
  • Linked health outcomes: ICD-10 codes, death registry, hospital episodes
  • Demographics, labs, medications, imaging follow-up
  • Largest available DXA imaging dataset with longitudinal CV outcomes

Clinical Validation

  • External validation on prospectively collected Hologic Horizon images from Moonshot Medical
  • Cross-manufacturer generalization testing (GE Lunar iDXA → Hologic Horizon)
  • Inter-rater agreement vs. expert manual scoring
  • Subgroup analysis across age, sex, and BMI strata

Publication Pathway

  • Paper 1: Model development & internal validation (UK Biobank)
  • Paper 2: Cross-manufacturer external validation (prospective Hologic Horizon collection)
  • Paper 3: Prospective deployment study & clinical utility
  • Target journals: Radiology, European Radiology, Radiology: AI
Data Strategy

Training & Validation Datasets

We've identified datasets across every CardioScreen modality. Some we can access independently — others require an academic PI with institutional affiliation.

What We Can Access Now

Open-access and credentialed datasets — no PI required.

  • MIMIC-IV-ECG — 800K diagnostic ECGs, linkable to clinical outcomes
  • Icentia11k — 11K patients, single-lead continuous ECG (matches our hardware)
  • PTB-XL — 21.8K cardiologist-annotated 12-lead ECGs, 71 diagnostic classes
  • CirCor DigiScope — 5,272 heart sound recordings with murmur grading
  • PulseDB — 5.2M PPG + ECG + arterial BP segments from 5,361 subjects
  • CUBS — 1,088 carotid ultrasound images with expert IMT segmentation
  • NLST Imaging — 54K participants, chest CT with aortic calcification (transfer learning)

What Requires an Academic PI

Institutional access, BioLINCC applications, or IRB approval needed.

  • UK Biobank — 95K lateral spine DXA images (GE Lunar iDXA) with linked CV outcomes
  • Endeavor Health Epic DW — internal DXA, ECG, carotid US, ABI data with outcomes
  • UChicago CRDW + HIRO — UChicago imaging archive with de-identification pipeline
  • WHI — 11K Hologic DXA lateral spine images in postmenopausal women
  • MESA — 6,814 patients: CT calcium + carotid US + ABI + ECG (multi-modal gold standard)
  • ARIC — 16K patients, carotid IMT + plaque + ABI + ECG, 35-year outcomes
  • Manitoba Bone Density — 5K expert-annotated AAC-24 scored DXA images (gold-standard labels)

Institutional Leverage

Endeavor Health and UChicago are both partners in the Institute for Translational Medicine (ITM), which facilitates cross-institutional data sharing and offers pilot grants up to $40K. An Endeavor-affiliated PI can access internal patient data via Epic Data Warehouse and submit BioLINCC applications for NHLBI studies (MESA, ARIC, WHI, CHS, Framingham) simultaneously.

Long-Term Vision

CardioScreen

A point-of-care cardiovascular screening platform with tiered architecture — automated triage at the primary care level.

Tier 1 — Enhanced Vitals (3 Minutes)

Non-invasive, fully automated. Any MA or nurse can run it.

  • Single-lead ECG with AI rhythm & morphology analysis
  • PPG for pulse wave velocity & arterial stiffness
  • Digital auscultation with murmur/S3/S4 detection
  • 3-cuff BP (bilateral brachial + ankle) for automated ABI, SpO2

Automated Decision Gate

Rule-based orchestration (not ML) evaluates Tier 1 results. If cardiovascular risk flags are present, the system recommends Tier 2. Deterministic, auditable, explainable.

Tier 2 — AI-Guided Carotid Ultrasound

Only for flagged patients. 5–7 minutes. No sonographer required.

  • Spatial alignment guidance for non-specialist probe placement
  • B-mode ML vessel wall segmentation & IMT measurement
  • Plaque detection
  • Doppler angle correction & stenosis grading (v2 roadmap)
Tiered
Regulatory Strategy
Incremental
FDA Submissions per Modality
PCCP
Change Control Framework
Technical Architecture

Modality-Separated AI

Each sensor modality runs an independent, validated ML model. No black-box fusion. A rule-based orchestration layer combines outputs into a deterministic risk assessment — auditable, explainable, and aligned with FDA's Predetermined Change Control Plan framework.

01

Independent Models

Each modality (ECG, PPG, auscultation, ultrasound) has its own trained model with independent validation metrics. No cross-modality dependencies. Each component can pursue regulatory clearance independently.

02

Rule-Based Orchestration

Decision gating uses deterministic rules, not another ML layer. Clinicians and regulators can trace every decision. Update one modality's model within pre-specified performance bounds.

03

Device-Agnostic

Sensor hardware is interchangeable. The platform integrates any ECG, PPG, stethoscope, or ultrasound probe that meets spec. No vendor lock-in.

Intellectual Property

12 Provisional Patents Filed

Comprehensive IP coverage across the full system — device architecture, AI methods, screening workflows, and clinical applications. All provisional applications filed March 2026.

Core Device & Platform

Tiered multi-modal screening system, modular cascading architecture, software-defined orchestration platform, inter-module risk data communication

AI / ML Methods

DXA body composition with visceral fat estimation & aortic calcification scoring, federated learning with automated validation & rollback, incidental CV risk detection from non-cardiac images

Ultrasound Guidance

Spatial alignment guidance for non-specialist acquisition, automated Doppler angle correction using ML vessel wall segmentation

Screening & Clinical

Multi-organ point-of-care triage, population-level CV risk benchmarking, DVT screening via bilateral NIR spectroscopy, specialist consultation marketplace

Research Collaboration

What We're Looking For

We have the engineering, the clinical infrastructure, and the IP. We need an academic research partner who can shape the science and bring institutional credibility.

What We Bring

  • ML engineering — pipeline built, model architecture designed, team in place
  • Clinical DXA scanner — Hologic Horizon at our practice for prospective data collection & external validation
  • Pilot site — active clinical practice for real-world deployment testing
  • Co-authorship — on all publications from this collaboration
  • Clinical advisory role — ongoing involvement as the platform develops

What We Need

  • Academic PI — for UK Biobank application & IRB oversight
  • UK Biobank imaging access — Field 20158, ~95K lateral spine DXA images
  • Scientific review — study design input, endpoint selection, manuscript review

Let's Build This Together

A publishable research project, a clinical tool that changes screening, and a platform with real translational potential.

Tom Kashul
Moonshot Medical — Park Ridge, IL

Moonshot Cardio is in the research and development phase. No products have been cleared or approved by the FDA. Descriptions reflect the intended design and development roadmap. All patent applications are provisional.