Thursday, May 18, 2017, 12:20PM, NAC 6/113

Basilis Gidas (Brown University), Finding Genes and Towards a Mathematical Framework for Artificial Intelligence and Biological Systems 

The first half of the lecture will be on a statistical model for finding genes in the human genome. The model contains two parts: (a) A finite network (graph) which represents the overall architecture of a gene. The vertices in the network represent DNA signals (small patterns) associated with a gene and which are recognized by proteins and enzymes involved in the transcription and translation of genes. The edges of the network correspond to interactions among these signals and represent statistical variability in the architecture across genes; (b) each signal and each part of a gene is a piece of DNA with a random length as well as a random variability of its nucleotide sequence. The second part of the model articulates these variabilities.

The above gene finding procedure is conceptually similar to what is believed to underlie speech recognition whereby recognition involves two types of information: The acoustic signal represented by a concatenation of phonemes, and global regularities articulated by grammars (or syntax). The underpinning process in visual recognition is undoubtedly similar. And so is – many practitioners believe – the functioning of biological processes whereby two principles are at work: physics (biochemistry) and evolution. Physics controls the biochemical interaction of macromolecules, but it is evolution that produced the perfect “code” or “syntactic language” for the collective behavior of genes (Gene Regulatory Networks), or the collective behavior of proteins in Signal Transduction Pathways in cell growth, cell division or immunology. While specific questions and application in speech, vision, and biology have seen impressive advances and have lead to a great deal of mathematical innovation (e.g. modern statistical learning), an underpinning mathematical framework is missing. Though we do not have the framework, we know quite a bit of some of the problems the framework needs to articulate and some of the properties it needs to have. Building on the gene finding process, the second part of the talk will aim at identifying some key sources that makes information processing in cognition and biology difficult, and hint towards a coherent hierarchical/grammatical framework.

Renowned neural engineer Risto Ilmoniemi will be speaking on Friday 4/21 at 3 pm in CDI 3rd floor conference room (3.352)

Professor Risto Ilmoniemi is a physicist and neuroscientist at Aalto University, Finland; he is the Head of the Department of Neuroscience and Biomedical Engineering. He built and designed multichannel MEG instruments in the 1980’s and invented for MEG use the minimum-norm estimate (together with Matti Hämäläinen), the signal-space projection, formulas for the forward problem, the channel-capacity measure for comparing sensor arrays, the triangle phantom, and several TMS techniques. He is the founder and former CEO of Nexstim Ltd., a company where he introduced navigated transcranial magnetic stimulation (TMS) and the combined use of TMS and EEG. As a professor of Applied Physics since 2006, he has led the development of new technologies for MEG–MRI and for a new generation of TMS. He is currently the Chair of the Biodesign Finland Program, which started in 2016.

See the complete Details here: Download Full Bio and Research Focus

Prof. Luca Parra (CCNY Biomedical Engineering), On Brainwaves and Videos and Video Games
Thursday, February 09, 2017, 03:30 PM, NAC 4/156
What are the immediate neural response of the brain to natural stimuli, in particular audiovisual narratives and video games? To answer this question we record EEG while subjects are exposed to the identical audiovisual narratives and measure inter-subject correlation, which captures how similarly and reliably different people respond to the same natural stimulus. We find that inter-subject correlation of EEG is strongly modulated by attention, correlates with long term memory, and provides a quantitative estimate for “audience engagement”. In children and adolescents watching videos we find changes with age and gender that are consistent with an increase in diversity of brain responses as they mature. During video game play, which are unique experiences that preclude correlation across subjects, we measure the strength of stimulus-response correlations instead. We found that correlation with both auditory and visual responses drive the correlation observed between subjects for video and that they are are modulated by attention in video game play. Importantly, the strongest response to visual and auditory features had nearly identical neural origin suggesting that the dominant response of the brain to natural stimuli is supramodal.

New Paper: Tolerability of Repeated Application of Transcranial Electrical Stimulation with Limited Outputs to Healthy Subjects

tolerability_limited_output_paper

Brain Stimulation 2016 May 24. pii: S1935-861X(16)30104-8. doi: 10.1016/j.brs.2016.05.008. [Epub ahead of print]

Abstract: The safety and tolerability of limited output tES in clinical populations support a non-significant risk designation. The tolerability of long-term use in a healthy population had remained untested. We tested the tolerability and compliance of two tES waveforms, tDCS and modulated high frequency transcranial pulsed current stimulation (MHF-tPCS) compared to sham-tDCS, applied to healthy subjects for three to five days (17–20 minutes per day) per week for up to six weeks in a communal setting. MHF-tPCS consisted of asymmetric high-frequency pulses (7–11 kHz) having a peak amplitude of 10–20 mA peak, adjusted by subject, resulting in an average current of 5–7 mA. A total of 100 treatment blind healthy subjects were randomly assigned to one of three treatment groups: tDCS (n = 33), MHF-tPCS (n = 30), or sham-tDCS (n = 37). In order to test the role of waveform, electrode type and montage were fixed across tES and sham-tDCS arms: high-capacity self-adhering electrodes on the right lateral forehead and back of the neck. We conducted 1905 sessions (636 sham-tDCS, 623 tDCS, and 646 MHF-tPCS sessions) on study volunteers over a period of six weeks. Common adverse events were primarily restricted to influences upon the skin and included skin tingling, itching, and mild burning sensations. The incidence of these events in the active tES treatment arms (MHF-tPCS, tDCS) was equivalent or significantly lower than their incidence in the sham-tDCS treatment arm. Other adverse events had a rarity (<5% incidence) that could not be significantly distinguished across the treatment groups. Some subjects were withdrawn from the study due to atypical headache (sham-tDCS n = 2, tDCS n = 2, and MHF-tPCS n = 3), atypical discomfort (sham-tDCS n = 0, tDCS n = 1, and MHF-tPCS n = 1), or atypical skin irritation (sham-tDCS n = 2, tDCS n = 8, and MHF-tPCS n = 1). The rate of compliance, elected sessions completed, for the MHF-tPCS group was significantly greater than the sham-tDCS group’s compliance (p = 0.007). There were no serious adverse events in any treatment condition. We conclude that repeated application of limited output tES across extended periods, limited to the hardware, electrodes, and protocols tested here, is well tolerated in healthy subjects, as previously observed in clinical populations.

Electrode configurations and montages. Identical electrodes and montages were ...

This page is no longer maintained. All data has been transferred to parralab.org.

 

marsEqCurrently available automated segmentation tools only provide results for brain tissues (gray matter, white matter, cerebrospinal fluid (CSF)), have a limited FOV, and do not guarantee continuity and smoothness of tissues, which is crucially important for accurate modeling of transcranial electrical or magnetic stimulation. MARS is a tool that addresses these needs.

MARS is an extended toolbox for SPM software. It is developed based on the SPM8 toolbox “New Segment”, which is an implementation of the Unified Segmentation algorithm (Ashburner & Friston 2005), where the image intensity model and anatomical prior (atlas) are combined into a Bayesian inference framework. MARS further added the morphological constraints on neighboring tissue voxels to the framework, which is encoded by Markov Random Field (MRF). This MRF information is added into the updating equation of the E-step of the algorithm, not just as a post-processing step. Compared to the New Segment, the resulting segmentation is shown to be more smooth and with less discontinuities, while at the same time the accuracy is not significantly changed. For more details on the algorithm, see Yu Huang, Lucas C. Parra, Fully Automated Whole-Head Segmentation with Improved Smoothness and Continuity, with Theory Reviewed. PLoS ONE 10(5): e0125477.

MARS is always initialized by the New Segment. In other words, this toolbox will run New Segment in the first step to get the optimal estimates on the bias field, the registration parameters between atlas and the image(s) to be segmented, and the initial estimates of segmentation posteriors and parameters of Gaussian mixture model. MARS will then continues to update the segmentation posteriors and Gaussian mixture model by using both atlas (tissue probability map, TPM) and tissue correlation map (TCM), while keep the bias field and registration parameters untouched. The toolbox is backwards compatible with the New Segment toolbox, as users can disable the MARS update so that only New Segment will be run.

The TCM is a way of representing MRF constraint. It indicates the probability of co-occurrence of specific two tissues in two neighboring voxels. For example, white matter cannot be adjacent to the air, and thus the probability of observing white matter as a neighbor of air voxel is 0, and this is encoded in the TCM. We developed three categories of TCM: the local, global and regional. The local TCM gives different co-occurrence probabilities depending on the location of the voxels, while the global version uses the same value for all the voxels in the image. The local TCM has nice theoretical properties, but not practical for implementation, as it needs too much memory to store and compute; while the global one is too rough to include any local information. The regional TCM is a trade-off between the two: it encodes locality information such as the interior of white matter and the boundary between grey and white matters, while at the same time uses the same co-occurrence value for one specific locality. Therefore, the regional TCM is highly recommended, and local TCM is not recommended unless you have a 64-bit computer with at least 50 GB memory.

Download the Matlab code and TPM/TCM here.