harmonycvi

Cardiovascular Health Assessment for Healthcare Providers

By /

The Challenge

The traditional process of analysing cardiac MRI scans is:

  • Time-consuming: A cardiac MRI scan (study) has between 3500 and 5000 slices, and manual analysis often takes up to 60 minutes per scan.
  • Subjective & inconsistent: Accuracy varies based on radiologist experience and fatigue, often resulting in diagnostic errors or delays.
  • Resource-intensive: Many centres lack specialised personnel or infrastructure to process cardiac MRIs effectively.
  • Disconnected: Existing solutions are vendor-specific, lack interoperability, and are often incompatible across platforms.

These limitations contribute to decreased accuracy, delays in reporting, and ultimately suboptimal output.

Our Solution

HarmonyCVI is an AI-powered cardiac MRI post-processing platform that automates and accelerates diagnostic interpretation using deep learning.

  • Fully automated segmentation and quantification of cardiac anatomy, including volumetric and Q-flow analysis, with clinical-grade accuracy.
  • Empowers clinicians with real-time, vendor-neutral, device-agnostic access to insights, reports, and structured data anytime, anywhere.

Key Features

    • Volumetric Analysis: Automated AI-based contour detection for left and right ventricles (EDV, ESV, etc.)
    • Q‑Flow Analysis: Automated segmentation and blood flow/velocity calculation without manual contouring
    • Multi‑View Layouts: Customizable composite layouts for syncing functional views for enhanced interpretation
    • Advanced Imaging: Supports Delayed Enhancement, Myocardial mapping (T1, T2 & T2*), Extracellular Volume Calculation (ECV), Global Longitudinal Strain, Atrial Volumes Analysis and bookmark-sharing across sessions

    Integrated Reporting: Real-time, interactive, customisable reports aligned to clinician needs.

The Impact

  1. Clinical Outcomes
    • <2-minute for processing all the slices and giving contours and parameters, making a quicker turnaround for the radiologist to report compared to traditional 60+ minutes from a manual process to about a 10-minute turnaround
    • Reduced radiologist reporting time from 60+ minutes to ~10 minutes, processing all slices, contours, and parameters in under 2 minutes, streamlining workflow and boosting efficiency
    • Accurate analysis for cardiac MRI workflows
    • Enables faster triage and better therapeutic decisions
  2. Clinician Efficiency
    • Seamless integration into clinical workflows
    • Reduces radiologist burden and minimizes inter-reader variability
    • Supports reimbursable, billable diagnostic workflows
  1. Health System Benefits
    • Adaptable for large hospitals, private clinics, and remote settings
    • Reduces time-to-treatment, lowers risk of missed diagnoses

Technologies Used

  1. Front-End Technologies
    • AngularJS: A component-based web framework for dynamic and responsive UI.
    • Weasis Viewer: Embedded DICOM viewer for rich, interactive medical image visualisation directly in the browser.
  2. Back-End Technologies
    • Java: Core service architecture, handling application logic, orchestration, and system integration.
    • AI: Multiple pre-trained deep learning and transformer models like SAM3, MedSAM2, Streamlined MLOps with CubeFlow.
  3. Databases & Imaging Infrastructure
    • PostgreSQL: Relational database used for secure storage of clinical metadata, user data, and analysis results.
    • Redis: In-memory data store for real-time caching and task queuing, enabling low-latency responses.
    • PACS: To retrieve and store DICOM files, enabling seamless integration into clinical imaging workflows.
  4. Cloud Infrastructure & Services (AWS)
    • Amazon S3: Secure, scalable storage for all incoming and processed imaging studies.
    • AWS Lambda: Manages the lifecycle of the AI inference server, auto-starting/stopping for efficient resource usage.
    • Amazon CloudWatch: Real-time logging and system monitoring for platform health and diagnostics.
    • EC2 Instances:
      1. AI Server: Dedicated GPU-enabled EC2 instance for executing AI inference tasks.
      2. Backend Server: Handles API requests, authentication, data processing, and user management
  5. On-premise: Entire platform is installed on Mac Mini and a few other GPU-based Edge devices
    1. GPU-enabled machine
      1. AI Server: Docker instance for executing AI inference tasks
      2. Backend Server: A Docker instance handles API requests, authentication, data processing and user management and connects to the client’s PACS server.

Conclusion

HarmonyCVI redefines how cardiovascular imaging is delivered. By combining AI & deep learning, cloud-native infrastructure, and clinician-first design, it enables rapid, accurate, and accessible cardiac MRI interpretation. With scan-to-report times under 2 minutes and diagnostic accuracy above 99%, HarmonyCVI transforms the cardiac imaging workflow from a manual bottleneck into a scalable, intelligent diagnostic tool.
It’s more than automation; it’s intelligence, delivered.

Scroll to Top