A Plugin-Based Architecture for Integrating AI Services in an Open-Source PACS

Abstract

[Abstract]: Automated image analysis, supported by powerful artificial intelligence algorithms, promises significant workflow advantages in the screening of medical images. The ability to automatically detect and classify objects of interest can drastically reduce the screening time, reduce observer variability, and help doctors in the diagnosis and report formulation. These advancements have gathered interest from the medical community, and several imaging platforms have evolved and adapted to this new reality, developing new analysis algorithms or providing interfaces to develop and integrate new ones. These applications have grown and thrived in a non-standardized environment, which means reutilizing these algorithms in different applications or sharing the results they output is not always possible. Additionally, developing these algorithms takes time and needs labeled datasets, which are not easily acquired. These factors limit the reach and applicability of these algorithms. This paper presents a framework that intends to address the standardization issue in medical image analysis by facilitating the integration and development of new algorithms in a production-ready imaging archive. The work proposes a new open-source interface, based on standard industry protocols, to be integrated into the open-source vendor-neutral PACS Dicoogle.