RadiologyNet

RadiologyNet: Machine Learning for Knowledge Transfer



DESCRIPTION:

The goal of this research effort is building deep predictive image annotation models for the purpose of learning with knowledge transfer in data-lacking domains. Models will be trained using the entire database of labelled anonymised radiology images from the Picture Archiving and Communication System (PACS) of Clinical Hospital Centre Rijeka, Croatia. The database consists mostly of CT and MR volumes. The project involves building a meaningful ontology of all colected images (approx. 30 million) based on database records and DICOM tags. The purpose of the learned models will be using them in computer-aided diagnosis (CAD) for solving various complex tasks related to medical radiology imaging, e.g. detection, classification or segmentation. Learned models will be made available online.


ONTOLOGY:

To be added...


MODELS:

To be added...


NEWS:

June 2017: Our application to the NVIDIA GPU Grant Program was succesfull. We will soon be receiving a Titan Xp GPU to be used for this research.

March 2017: We finished collecting the data. Work has started on examining rules for building an ontology and assigning tags to images.

May 2018: We developed a fast in-memory distributed service for querying radiology images by DICOM metadata.


CONTACT:

If you have any questions, please contact Ivan Ċ tajduhar:

University of Rijeka - Faculty of Engineering
Department of Computer Engineering

Vukovarska 58, 51000 Rijeka, Croatia

+385 51 651 448, istajduh@riteh.hr


This research is supported in part by the University of Rijeka under the project number 16.09.2.2.05

UNIRI KBCRI MEDRI RITEH