Over the course of the project, THOR evolved to a clinically validated, market-ready AI-based Clinical Decision Support System (CDSS) for thoracic CT imaging. Work focused on three pillars: development of a robust AI model suite, creation of scalable infrastructure for clinical integration, and validation through real-world pilot studies.
The AI model suite was enhanced to deliver comprehensive quantification for major lung diseases, including COPD, bronchiectasis, cystic fibrosis, and asthma. Early prototypes showed promising results, and subsequent iterations incorporated advanced segmentation algorithms for airways, lobes, and fissures, plus mucus detection and ventilation estimation. These improvements were supported by expanded training datasets to ensure robustness. A dedicated quality control system safeguarded reliability, integrating artifact detection and confidence scoring. By the end of the project, the models achieved strong performance metrics within clinical usability thresholds.
Parallel to algorithmic development, significant effort went into building infrastructure integrating into hospital workflows. A modular containerised architecture guaranteed portability, scalability, and simplified updates. A cloud-based platform orchestrated data transfer, automated processing, and generated structured reports, while integration endpoints were developed to support both cloud-to-cloud and on-premise deployments. Multiple report formats were finalised, including DICOM encapsulated PDFs and structured reports, ensuring compatibility with diverse clinical environments.
Pilot studies in major hospitals validated the system’s technical performance and usability. Clinicians confirmed THOR’s outputs were accurate, interpretable, and aligned with workflow requirements. Feedback informed refinements in report design and integration processes, resulting in a solution that maximises clinical relevance. Security and compliance assessments confirmed adherence to GDPR and ISO 27001 standards, while preparatory steps for CE-marking were completed. By the end of the reporting period, THOR had reached Technology Readiness Level 8, positioning it for imminent commercial deployment.
