Closing the gap in the clinical adoption of computational pathology: a standardized, open-source framework to integrate deep-learning algorithms into the laboratory information system

Abstract

AbstractDigital pathology (DP) has revolutionized cancer diagnostics, allowing the development of deep-learning (DL) models supporting pathologists in their daily work and contributing to the improvement of patient care. However, the clinical adoption of such models remains challenging. Here we describe a proof-of-concept framework that, leveraging open-source DP software and Health Level 7 (HL7) standards, allows the integration of DL models in the clinical workflow.Development and testing of the workflow were carried out in a fully digitized Italian pathology department. A Python-based server-client architecture was implemented to interconnect the anatomic pathology laboratory information system (AP-LIS) with an external artificial intelligence decision support system (AI-DSS) containing 16 pre-trained DL models through HL7 messaging. Open-source toolboxes for DL model deployment, including WSInfer and WSInfer-MIL, were used to run DL model inference. Visualization of model predictions as colored heatmaps was performed in QuPath.As soon as a new slide is scanned, DL model inference is automatically run on the basis of the slide’s tissue type and staining. In addition, pathologists can initiate the analysis on-demand by selecting a specific DL model from the virtual slides tray. In both cases the AP-LIS transmits an HL7 message to the AI-DSS, which processes the message, runs DL model inference, and creates the appropriate type of colored heatmap on the basis of the employed classification model. The AI-DSS transmits model inference results to the AP-LIS, where pathologists can visualize the output in QuPath and/or directly from the virtual slides tray. The developed framework supports multiple DL toolboxes and it is thus suitable for a broad range of applications. In addition, this integration workflow is a key step to enable the future widespread adoption of DL models in pathology diagnostics.