The paper, "Hardware Implementation of a Brain Inspired Filter for Image Processing," published in IEEE Transactions on Nuclear Science (vol. 64, no. 6, part 1, 2017, pp. 1374-1381) has been selected to receive this year Transactions on Nuclear Science Best Paper Award. It is customary to present the award at the Nuclear and Plasma Sciences Society (NPSS)-sponsored conference most relevant to the topic, that is the Real Time conference that will be held in Vietnam in October 2020.
The study presented in the paper was developed within the European-FP7 project “Fast Tracker for Hadron Colliders” and exploit analogies between fast reconstruction of two types of big data images: particle interaction events and Magnetic Resonance brain images (MRI).
From the left side: Alessandra Retico and Paola Giannetti (INFN Pisa), Chiara Roda and Mauro Dell’Orso (University of Pisa).
The ability of the human brain to select only relevant data for a given task has inspired our collaboration to develop new processing technologies for the world of Big Data. The awarded paper shows that these technologies can lead to much quicker medical diagnoses.
The project team built “accelerators” for algorithms that usually take up a large amount of processing time and resources. This technology works by filtering significant information for further processing out of images that are too complex to be processed directly by standard computers. This is how the brain processes images. For higher-level processing and long-term storage, it only selects data that matches a particular set of memorized patterns. The used technology emulates this low-level brain function.
The awarded paper presents the software and hardware implementation of a pattern-matching algorithm that emulates the pattern matching process of the human brain and demonstrates its potential use in biomedical applications, more specifically MRI research.
The hardware implementation of this algorithm combines Filed Programmable Gate Arrays and the Associative Memory integrated circuits in a powerful combination to execute both the training and the data acquisition phase of the algorithm in real-time. We have demonstrated that for 2-D B/W images, this system accelerate the training and acquisition phases by about 1000 and 100 respectively compared to a last generation i7 CPU. The system is generic and can be easily adapted to process the much more demanding 3-D MRI images.
Full details on the research subject of the prized paper is described in: http://ftk-iapp.physics.auth.gr/Pisa/image.html.