RIPE: Radiology Information and PACS Extractor
RIPE: Break the bottleneck of AI-driven radiology
In the new era of AI-driven radiology, data is the new gold. Still, data extraction and generation of high-quality labeled cohorts is extremely tedious and work intensive.
Data extraction can consume up to 90% of the total effort of most machine learning (ML) projects. Ultimately, the quality of these input data defines the usability of the final ML model.
Here, we present an open-source, customizable data extraction and preprocessing pipeline (Radiology Information and PACS Extractor: RIPE) that enables automated bulk information retrieval tailored for consequent modular deployment of downstream prognostic and predictive models using computer assisted feedback for translational research in the daily praxis.
RIPE has received the poster prize in poster session IX (AI & neurooncology) of the Annual Meeting of the German Society of Neuroradiology (DGNR 2019, Frankfurt am Main, 10/12/2019).
We developed (project lead Andreas Junge) a customizable, light-weight, open-source software solution on top of the local RIS/PACS system, that
- can be set up flexibly at any institution capable of csv export of RIS data
- can not only diminish the need for manual data retrieval and preprocessing but also
- significantly speed up the end to end development and deployment process of AI algorithms
Hence, supporting radiologists to provide precision medicine and to improve patient care at early stages of diagnosis.
RIPE - Dashboard View
RANO criteria vs. Patient age
Proportions of RANO criteria vs. Tumor type
Density of age distributions vs. Tumor type