Retweeting John von Neumann’s words from the 1950s, “All stable processes we shall predict. All unstable processes we shall control.”, it becomes immediate how computational neuroscience might form a basis for novel engineering approaches in neural medicine. Brain modeling and neural engineering are, in footsteps of these words, aiming at understanding the brain and its emerging phenomena on the scientific side, and interacting with the brain—also with the perspective of medical treatment—from the engineering science side. Both sides of the theory-experiment coin have always been connected as, e.g., electrodes are used for data acquisition as well as to influence neural systems, be it only by well-defined input pulses of a measurement protocol (Siegelbaum and Kandel 1991; Bi and Poo 2001).

This roadmap, however, persistently is a difficult enterprise for various reasons. The brain is a highly nonlinear dynamical system and remains a challenge for data analysis, theoretical modeling, and large-scale computation (Singer 1999; Kantz and Schreiber 1997; Pikovsky et al. 2001; Lehnertz et al. 2000; Olbrich et al. 2011).

In review of these challenges, we would like to emphasize on four areas of research that mark out outstanding future potential, and are related to central keywords: Consciousness, Pathological Oscillations, Neuroprosthetics, and Neuroenhancement. In these four areas, computational models in a similar way connect natural sciences and neuroengineering research and pave a way towards medical applications.

FormalPara Understanding consciousness, anaesthesia and sleep

Large-scale synchronized oscillations are easily observed experimentally and can be both of physiological and pathological character. The mammal sleep-wake cycle is a remarkably robust oscillation, in whose regulation again various neural oscillations are involved, including the cortical slow oscillation in the so-called delta band, with frequencies around 1 Hz (Compte et al. 2003; Ngo et al. 2010; Mattia and Sanchez-Vives 2012). Electrical stimulation of the brain at this slow wave frequency enhanced the slow oscillations themselves as well as increased their memory consolidation effect (Marshall et al. 2006). The slow oscillations, comprised by the interplay between bursting activity and an activity-dependent self-inhibition, exhibit a specific anticorrelation in the durations of Up and Down states, as Mattia and Sanchez-Vives (2012) show in comparison of ferret brain slice data, mean-field models, and simulations. Mean-field models describe the gross activity of a neural subpopulation (at column level or below) but keep track of main types of neurons and their simplified connectivity. They are of great advantage in describing the consciousness transitions of sleep and general anaesthesia (Hutt 2012; Steyn-Ross et al. 2012). As modeled by Hutt (2012), GABAergic tonic inhibition influences the brain’s arousal system during general anaesthesia inducing a loss of consciousness. Including the effect of gap junctions on the dynamics elucidated the influence on the propensity of generalized seizures (Steyn-Ross et al. 2012).

FormalPara Medical treatment: pathological oscillations

It is only a recent development that medicine has spotted the essential importance of dynamical phenomena for understanding and treatment of certain diseases. These are primarily those where oscillations themselves comprise the disease as in movement disorders, namely essential tremor and Parkinson’s disease, for which electrical stimulation methods co-developed with theoretical work and computer simulations have found their way into clinical practice (Tass et al. 2006). But even when there is no observable mechanical or electrical oscillation, as in the cortical spreading depression which is comprised by slow (102–103s timescale) Ca2+ waves, control methods may become means of treatment (Dahlem et al. 2008). In epilepsy, and in more severe cases of mood disorders, deep brain stimulation is applied (Abelson et al. 2005) and sophisticated technical implementations as radio stimulation are developed (Delgado et al. 1968). Also, mechanical damages to neural pathways can result in pathological dynamical phenomena, as in the case of spinal cord injury which leads to a hyperexcitability of motorneurons. The modeling approach by Venugopal et al. (2012) goes down to the role of each ion channels and aims at a reduction of spasticity. Finding suitable parameter ranges for electrical stimulation is annother important issue. Krishnamurthi et al. (2012) investigate by measuring velocity reduction how an optimal amplitude of DBS can be found for Parkinson’s disease. Considering a Rempe-Terman based computational model of basal ganglia, in Njap et al. (2012) it is demonstrated that a high concentration of the inhibiting neurotransmitter GABA together with electrical stimulation reestablishes faithful thalamocortical relaying. A more general question is addressed in Schütt et al. (2012) by investigating low- to high-frequency stimulation in an Izhikevich-type cortical network model, with the observation that in a frequency range around 100 Hz, a dynamical desynchronization is observed.

FormalPara Neuroprosthetics and brain-computer interfaces

The consequent continuation of few-electrode stimulations and recordings are Human-Machine interfaces (HMIs) (Birbaumer 2006), the most apparent applications where theoretical brain science and engineering meet. In the loss of direct control through the natural pathways, prosthetic devices have to be controlled through an HMI which comprises a “thought-control” (Hochberg et al. 2006; Pfurtscheller et al. 2003). The consequent continuation of that idea—and most immediate demonstration of an HMI at work is that a blind patient can use the HMI for reading (Zrenner et al. 2011), cochlea implants improve hearing (Edgerton et al. 1982), and patients with locked-in syndrome can use an EEG-based BCI for expressing words (Wolpaw et al. 2002). Overall, prosthetic applications require a sound modeling approach and understanding of the neural processing, and, if possible, also coding, to design an effective information interfacing with the brain. The visual system is a part of the brain where experimental research and detailed modeling have made large progress. Here, Norheim et al. (2012) and Einevoll et al. (2012) investigate both a minimal and a feedback—extended model for temporal processing in the lateral geniculate nucleus (LGN). The final step of brain-machine interfacing goes towards dissolution of the border between computer and brain itself: In their beautiful experimental setup, Perez-Marcos et al. (2012) demonstrate virtual hand illusions and—at least for a part of the body—question the conscious awareness of our Self, and thereby connect two quite different fields.

FormalPara Understanding and influencing neural plasticity

Neural plasticity, the basis of all learning, can not only be influenced pharmacologically, but also by various means of electrical stimulation with remarkable effects on cognitive learning and consolidation (Marshall et al. 2006). Stimulation, as well as learning, can effect on both the dynamics (Mattia and Sanchez-Vives 2012; Schütt and Claussen 2012) and on the plasticity (Clopath 2012; Vogt and Hofmann 2012). Vogt and Hofmann (2012) demonstrate modulatory effects of dopamine based on an underlying STDP learning mechanism. Modulatory or multi-input based learning mechanisms are good candidates to explain memory consolidation. In this direction, Clopath (2012) compares two recently proposed mechanisms, namely tag-trigger consolidation, and a metastate tagging model, and provides a comprehensive comparison of both concepts. Finally, Weigenand et al. (2011, 2012) investigate the phase-dependence of stimulation of cortical slow waves. It would be highly interesting, based on improved understanding of neural coding and plasticity mechanisms, to design specific stimulation protocols that selectively strengthen desired acquired memories.

FormalPara Outlook

While it is tempting to go beyond the purpose of medical treatment by “enhancing the brain” (Farah et al. 2004), a deeper understanding of neural plasticity mechanisms by theoretical and computational models also is expected to offer pathways to the treatment of various neural disabilities and disorders, be them memory-related like Alzheimer, or mood-related disorders as depressive disorders. How emotion modulates learning, and how emotional processes dynamically regulate psychological states is an emerging field (Huber et al. 1998, Figueroa Helland et al. 2008) and can be expected to become the ‘neuroengineering’ extension of computational modeling for the treatment of brain disorders.