Objectives
VISCERAL will define and execute a targeted benchmark framework to speed up progress towards:
- Automated anatomy identification and pathology identification in 3D (MRI, CT) and 4D (MRI with a time component) radiology images.
- Similar case retrieval for these images and the reports associated with them.
The following objectives will be met in carrying out the VISCERAL project:
- Create an evaluation infrastructure and software to allow the benchmarks to be carried out efficiently and effectively, but that also allows continuous evaluation to take place beyond the benchmarks;
- Innovate through the use of a cloud infrastructure for the evaluation of algorithms on huge amounts of data;
- Run two benchmarks and two workshops at which benchmark results will be discussed;
- Fuse a large number of automated benchmark entries to create a very large silver corpus;
- Create a small but sufficiently large gold corpus by manual annotation of the radiology images (used to evaluate the quality of the evaluation by the silver corpus);
- Release the radiology data and the silver and gold corpora as research collections at the end of the project.
Two benchmarks will be organised during the project:
- Benchmark 1: Identification, localization and segmentation
- The Benchmark tasks will be organ identification and the segmentation of bones, inner organ and relevant substructures.
- Benchmark Dates: August to November 2013
- Benchmark 2: Retrieval
- The Benchmark task will be the retrieval of similar cases based on both visual information and radiology reports.
- Benchmark Dates: May to August 2014