RAPIER – Radiology Pathology Information Exchange Resource

Project RAPIER aims to create a RadPath datalake on liver lesions and myriad deployable AI-driven applications which can detect, describe features and diagnose liver abnormalities, catalysing next generation AI-enabled clinical decision support, predictive analytics, and precision medical care.

Target Sector: Healthcare

Lead PI: Dr Liu Yong (A*STAR)

Co-PIs

  • Xu Xinxing (IHPC, A*STAR)
  • Fu Huazhu (IHPC, A*STAR)
  • Zhen Liangli (IHPC, A*STAR)
  • Li Bing (IHPC, A*STAR)
  • Rick Goh Siow Mong (IHPC, A*STAR)
  • Li Shaohua (IHPC, A*STAR)

Host Institution: Institute of High Performance Computing (IHPC), A*STAR

Industry Partners:

  • Singapore General Hospital (SGH)
  • National Cancer Centre Singapore (NCCS)

Medical Principal Investigators

  • Dr Lionel Cheng Tim-Ee, Department of Diagnostic Radiology, SGH
  • Dr Tony Lim Kiat Hon, Department of Anatomical Pathology, SGH

Medical Investigators

  • Dr Gideon Ooi Su Kai, Division of Oncologic Imaging, NCCS
  • Dr Thng Choon Hua, Division of Oncologic Imaging, NCCS
  • Dr Khor Li Yan, Department of Anatomical Pathology, SGH
  • Dr Tran Nguyen Tuan Anh, Diagnostic Radiology, SGH
  • Dr Leow Wei Qiang, Department of Anatomical Pathology, SGH
  • Dr Shi Ruoyu, Department of Anatomical Pathology, SGH

Problem Scope

Radiology and pathology archives are huge treasure troves of clinical images and text reports. However, as such data are mostly in silos, the potential for AI to tap such information is limited. 

Project RAPIER aims to create a datalake hosting the RadPath archives of images and reports for liver pathology in SGH campus. This datalake will be used to develop AI algorithms that can detect, describe lesion features and diagnose liver abnormalities on RadPath images.  

The key objectives of this project are to develop AI-assisted data labeling tools, liver lesion analysis, reporting of CT/MRI and digital pathology images, and liver lesion progression prediction. 

Design Approach

An AI-assisted data labelling solution will first be developed to assist clinicians in labelling medical RadPath images. The final labels applied by clinicians would be used for training to iteratively improve the system. Longitudinal RadPath data of patients over time will also allow AI algorithms to develop predictive analytic capability for disease progression.

Potential Impact/Benefits to Target Sector

The tools from this project will unlock previously untapped medical data across the RadPath domains to facilitate AI model training and validation. The suite of well-trained AI algorithms can augment specialists for early detection, accurate diagnosis, and prompt treatment of liver disease, thus improving accuracy and report turnaround, and healthcare outcomes. In addition, the capabilities developed in RAPIER can be applied to many other diseases beyond the liver.