PUMA

The project aims to transfer the technologies developed for high-energy physics to Magnetic Resonance Fingerprinting (MRF), a methodology for the quantitative analysis of magnetic resonance images. In MRF, the acquired data are used to create a unique “fingerprint” of the properties of the different types of biological tissue. These fingerprints are then compared, through a special recognition algorithm, with a “dictionary,” to produce a quantitative map showing the properties of each tissue.

The process, therefore, takes place in three steps:

• • data acquisition;
  • recognition (Pattern Matching);
  • display of the tissue properties.

The slow stage is step 2, because the processing time increases exponentially with the size of the dictionary. For each pixel acquired, it is necessary to compare the measured signal with the one recorded in the dictionary; each comparison is performed sequentially on the CPU or GPU, and the complete reconstruction of a typical dataset takes days.

The proposed technology is instead based on parallel processing, executed through FPGAs and a specially developed Associative Memory, integrated on a dedicated ASIC. This can speed up the process by thousands of times compared to traditional GPU processing.