Related Experiment Video
Updated: Dec 20, 2025
05:33
Introduction of an Integrated Pathology Image Management, Artificial Intelligence, and Reporting System
Published on: July 11, 2025
665
An Open-Source, Vender Agnostic Hardware and Software Pipeline for Integration of Artificial Intelligence in
Jae Ho Sohn1, Yeshwant Reddy Chillakuru2,3, Stanley Lee2
1Radiology and Biomedical Imaging, University of California San Francisco (UCSF), 505 Parnassus Ave, San Francisco, CA, 94143, USA. sohn87@gmail.com.
Journal of Digital Imaging
|May 30, 2020
Summary
This study introduces a simple system for integrating machine learning (ML) models into radiology workflows, enabling safe and accurate AI deployment. The developed framework facilitates ML model testing without disrupting existing Picture Archiving and Communication System (PACS) integration.
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Healthcare
- Clinical Workflow Optimization
Background:
- Machine learning (ML) shows promise in radiology, but clinical integration remains a significant challenge.
- Complex IT environments and Picture Archiving and Communication Systems (PACS) hinder practical ML implementation.
- Clinical validation is essential for ensuring the safety and accuracy of ML algorithms in healthcare.
Purpose of the Study:
- To propose, develop, and demonstrate a hardware and software system for seamless ML model integration into standard radiology workflows.
- To create a flexible framework for testing ML algorithms within existing hospital IT infrastructure.
- To address the challenges of integrating ML into clinical radiology practice.
Main Methods:
- Established a Digital Imaging and Communications in Medicine/Graphics Processing Unit (DICOM/GPU) server and software pipeline.
- Implemented a hospital IT-agnostic ML integration schema, tested with a breast density classification algorithm.
- Prospectively evaluated processing time delays using 100 digital 2D mammograms.
Main Results:
- Successfully implemented and demonstrated an open-source clinical ML integration schema.
- The ML pipeline processed 100 studies in an average of 26.52 seconds.
- Identified model load and study stability as key factors influencing processing time.
Conclusions:
- Demonstrated the feasibility of deploying and utilizing ML models in radiology without workflow disruption.
- The proposed system provides a simple, efficient, and understandable framework for ML integration and testing.
- The open-source schema supports the uploading of custom ML models, promoting wider adoption.
