Deployment and Evaluation of Intelligent DICOM Viewers in Low-Resource Settings: Orthanc Plugin for Semi-automated Interpretation of Medical Images

Medical image interpretation is a crucial part of radiological workflows, providing valuable input to the final output of the process: medical image interpretation reports. Radiologists typically employ Digital Imaging and Communication in Medicine (DICOM) Viewers. While numerous types of DICOM Viewers have been implemented, there has been arguably little focus on how such software tools can be made more effective by integrating them with Artificial Intelligence (AI) services. This paper presents a study conducted to design and implement an Orthanc Web-based Picture Archiving and Communication System (PACS) plugin DICOM Viewer for facilitating the semi-automated interpretation of medical images using AI. An Orthanc DICOM Viewer plugin, interoperable with two (2) models—Pneumonia Classification and Detection models—was implemented using the Python programming language. The plugin was evaluated with Radiologist Residents at a large University Teaching Hospital in a controlled setting. TAM 2 instrument was used to assess its perceived usefulness and ease of use. The TAM 2 constructs were rated positively, with study participants expressing a desire to incorporate other pathologies to the plugin and, additionally, integrating similar tools in more widely used DICOM Viewers such as RadiAnt and Weasis. The integration of AI models has the potential to reduce the turnaround time and workload involved in interpreting medical images. More significantly, the positive responses related to perceived usefulness of the tool suggest a potential that such tools have in improving delivery of services at the point of clinical care.