ISBI Daily - Friday

Poster Presentation: Bayesian Reconstruction of R-Fmri from K-T Undersampled Data Using a Robust, Subject- Invariant, Spatially-Regularized Dictionary Prior 20 Friday ISBI DAILY Prachi explains that the work is regarding resting state fMRI (R-fMRI). The R-fMRI is functional MRI of the brain taken when the subject is in a resting state – that is not performing any elaborate task. This is especially useful when the subjects are young children, elderly, or patients with some brain disorders, who may not be able to cooperate with tasks generally involved in task-based fMRI. The R-fMRI signal comprises of BOLD signal – which stands for blood- oxygenation-level-dependent signal. It means that the parts of the brain that are involved in a similar activity have a similar level of blood and oxygen supply and this shows up in this data. The problem with this data is that since the imaging is done in the resting state, the signal is typically weak and prone to large fluctuations of physiological noise. That is why even though the component frequencies of the time series in the R-fMRI data are less than 0.1Hz, higher temporal sampling rates are being used for the imaging to overcome this noise. This leads to a limitation on spatial resolution and thicker slices are being imaged – about 2-4mm thickness, but typically the thickness of the cerebral cortex is 3- 4mm, so this compromises the reliability with which studies of the cerebral cortex can be done. Faster per-slice scans through k-space subsampling are needed which can enable thinner slices. For k-space subsampling, some methods apply a complex pulse- sequence design through non- Cartesian k-space subsampling, but that may cause artefacts. Other methods use straightforward Cartesian k-space subsampling, but with priors on the signal. Prachi says their method falls in this second category, but they have a novel dictionary prior which is different from the other kinds of priors, and can produce better results: “ Our dictionary approach is robust to Prachi H. Kulkarni is a PhD student at the Indian Institute of Technology Bombay. She spoke to us before her poster session yesterday. Prachi H. Kulkarni “O ur method is able to support subsampling not only in k-space, but also in time (8X K-T subsampling), thereby potentially enabling 8X higher spatial resolution. ”