Download The Living Network Lab focuses its group is

A cultured human neural network drives a robotic actuator
R.M.R. Pizzi°, D. Rossetti°, G. Cino°, D. Marino* and A.L.Vescovi*
°Living Networks Lab
Department of Information Technologies – University of Milan
Via Bramante 65, 26013 Crema (CR)
Tel +39 02 503 30072- Fax +39 02 503 30010
[email protected]
*Stem Cells Research Institute, DIBIT S. Raffaele
Via Olgettina 58 – 20132 Milano (Italy) – tel +39 02 2643 4952
Abstract
Since 2002 the Living Networks Lab (Department of Information Sciences, University of
Milan) has experimented the growth of networks of human neural stem cells on a MEA
(Microelectrode Array) support. The group is composed by physicists, electronic designers,
computer scientists and biotechnologists, with the support of external biological labs.
The group is concerned with the research in the field of computational biology, bionics and
Artificial Intelligence.
Many experiments have been performed developing and analyzing organized structures of
biological neural networks.
The neurons were stimulated by simulated perceptions in form of digital patterns and the
output signals were analysed .
In previous experiments, the neurons replied selectively to different patterns and showed
similar reactions in front of the presentation of identical or similar patterns.
Analyses performed with a novel Artificial Neural Network called ITSOM showed the
possibility to decode the neural responses to different patterns.
In the described experiment, the neurons are connected to a robotic actuator: simulated
perceptions stimulate the neurons, that react with organized electric signals. The signals are
decoded by the Artificial Neural Network that drives a minirobot. The hybrid system shows
interesting performances, that are quantitatively evaluated and discussed in the paper.
Keywords: MEA, neurons, stem cells, hybrid, bionics, robot, Artificial Neural Networks
1. Introduction
The interest towards the development of a direct interface between electronics and nervous cells
began in the early nineties with the Fromherz’s (Max Planck Institute of Biochemistry) works
(Fromherz et al (1991), Fromherz et al (1993), Fromherz and Schaden (1994), Weis et al (1996),
Jenkner and Fromherz (1997), Schatzhauer and Fromherz (1998)).
After these pioneering experiences, researches in this field were carried out in many other
laboratories with different approaches and with the aim to develop future bioelectronic prostheses,
hybrid human/electronics devices and bionic robots.
Advances in neurophysiology were reached by means of experiments based on the
multielectrode stimulation of animal neurons (Akin et al (1994), Wilson (1994), Bove et al (1996),
Canepari et al (1997), Borkholder (1997), Maher et al (1999), Jimbo and Robinson (2000), Egert
(2002)).
Chaos-based neural computation was accomplished by a Georgia Tech group (Schiff et al
(1994), Lindner and Ditto (1996), Garcia et al (2003)).
In 2005 the SISSA group (Ruaro et al (2005)) experimented the possibility to use neurons on
MEAs (Micro Electrode Arrays) as “neurocomputers” able to filter digital images.
While the Fromherz’s group carried on its work on more and more evolved chip-neuron
junctions (Jenkner et al (2001), Zeck and Fromherz (2001), Bels and Fromherz (2002), Bonifazi and
Fromherz ( 2002), Fromherz (2002)), first attempts were accomplished in developing hybrid
creatures formed by animal neurons connected to robotic arms or virtual animals (Chapin et al
(1999), Reger et al (2000), Taylor et al (2002)).
In 2003 the Duke University’s group (Carmena et al (2003)) inserted 320 microelectrodes into
a monkey brain, allowing the monkey to learn to move a robotic arm.
The Potter’s group at Georgia Tech (Potter (2001), DeMarse et al (2001), Wagenaar et al (2001),
DeMarse et al (2002)) created a hybrid creature made by neurons from rat cortex able to learn from
the environment.
Lately, this and other groups designed closed-loop studies with hybrid natural/artificial systems
(hybrots-neurobots): Kositsky et al (2003), Wagenaar and Potter (2004), Martinoia et al (2004),
Potter et al (2005). In Bakkum et al (2004), under the control of the neural network a Koala 6wheeled rover was commanded to approach another randomly driven robot.
Nonetheless, the dynamics of a network of neurons that receives sensory inputs, stores memories
and controls movement and behaviour is not fully understood yet. The current methods are based on
the detection and analysis of spike trains (Rieke et al (1997), Borst and Theunissen(1999), Paninski
et al (2004), Pillow et al (2005)), with interesting but not decisive results.
Our group used MEAs to culture human neural stem cells and an Artificial Neural Network to
interpret the neural signals in its entirety, analyzing not only the spike trains but the whole signals.
As we will describe in the following, this allowed our hybrid neurons/electronics system to learn
simulated perceptions and to react correctly to a following presentation of the learnt patterns.
2. Materials
2.1 The MEAs
Our system is based on the MEA (Microelectrode Array) technology.
A MEA is a glass Petri dish where small electrodes are inserted. The cultured neural cells
adhere directly on these electrodes. Each electrode is connected by means of an isolated track to a
pad suitable for the external connection. MEAs allow to record the activity of cells simultaneously
on different channels for a long time without damaging the cultures.. Our Panasonic MEAs have 64
ITO (Indium Tin Oxide)- Platinum microelectrodes. The microelectrode size is 50µ, the interpolar
distance 150µ. Their very low impedance (10kΩ) is critical to achieve a good signal-to-noise ratio .
In particular, MEAs are suitable for our experiments that study the dynamical behavior of a
whole neural network.
2.2 The neural cells
In our experiments we used human neural stem cells isolated from the diencephal/telencephal
area of human fetuses naturally aborted at the tenth week of gestation (Vescovi (1999)).
The stem cells can be isolated and cultured in-vitro as they have the ability to proliferate
forming at first neurospheres that, as multipotent progenitors, can differentiate if suitably stimulated
and become neurons, astrocytes and oligodendrocytes (McKay (1997), Gritti et al (1999), Gritti et
al (2000), Gritti et al (2001), Galli et al (2003))
It is important to settle if differentiated stem-derived cells have appropriate electrophysiological
properties.
It has been proved (Hiroki et al (2000), Song et al (2002)) that adult neural stem cells isolated
from a rat hippocampus, after differentiating by means of BDNF (Brain Derived Neurotrophic
Factor), showed to have developed functional synapses in-vitro.
Besides, the existence of in-vitro synapses has been demonstrated also on the basis of the
morphological appearance (see Fig. 1). The neural maturation takes 18 days and the increase of the
currents amplitude in the Na channels and the appearance of action potentials suggest that the
channel functionalities grow gradually during their maturation.
In conclusion we can say that the existence of functional synapses in cultures of neural cells
derived by neural stem cells has been experimentally proved.
The use of stem cells was justified by the fact that this kind of cells shows a high adaptation
ability that allow them to differentiate and adhere on the MEAs. Besides they have the capability to
proliferate exponentially without requiring the use of multiple dissections.
Moreover, their well-known adaptability to grow in any kind of tissue make them optimal
candidates to future possible bionic implants.
The stem cells in form of neurospheres have been plated at a density of 3500 cells/cm2 in a
chemical medium containing EGF (Epidermal Growth Factor) and FGF-2 (Fibroblast Growth
Factor 2).
The MEAs must be accurately washed before the use because they are assembled using silicon,
that is neurotoxic, and the electrodes must be perfectly clean in order to conduct correctly the
electrical signals. The washing protocol includes 48 hours of rinsing with distilled water and 70%
methylic alcohol, then an overnight exposure to UV rays and sterile cap drying.
Fig. 1. Stem cells on the MEA electrodes
Once the washing procedure is finished, it is possible to lay down the matrigel on the MEAs.
This is a delicate step, as the matrigel (an adhesion molecule derived by a mouse sarcoma) is at
a gel state at 4 °C but solidifies at 37 °C after 20 minutes. However we experimented a detachment
of the matrigel after 2-3 days (see Fig. 2), that caused the cells’ death.
Fig. 2. Matrigel detachment
After many trials we solved the problem by leaving the matrigel overnight on the MEAs before
adding the cells, and this techniques allowed a good adhesion of the matrigel during the 25 days
needed to cell maturation.
Cells were mechanically dissociated and plated in a number of 50000 on the MEAs with FGF2,
BDNF and other components for 15 days before our learning experiment.
Observing the cells after 15 days it is possible to see that in some areas the matrigel is missing
and the cells adhere directly on the electrodes.
The MEAs are put inside a thermically controlled box at 37 °C., whose resistors are carbonbased and are DC supplied.
In order to ensure a good electromagnetical shielding, the plexy-glass box is coated by a brass
thick-meshed (1 mm) net.
However, the thermocontrolled box does not ensure the suitable athmosphere ( Oxygen 20%,
CO2 5%) for the cells’ survival. Thus we adopted a Tyrode buffer that allows to maintain the
correct ph and glucose levels ( containing NaCl, KCl, CaCl2 , MgCl2 , glucose and 7.4 ph).
Adding the Tyrode allowed 2 hours of survival , enough to complete or experiments.
It is worth mentioning that at the end of the experiment we performed a last recording on cells
treated with TTX (tetrodotoxyn). TTX is a potent neurotoxin extracted from the Tetraodontidae fish
(blowfish) which blocks selectively the sodium ion transport through the membrane, modifying
excitability and inhibiting the action potentials.
After adding TTX to the cell culture, the voltage values attenuated significantly up to a noiselevel signal. In this way we could show that the recorded signals contained a functional neuronal
activity and that it was active up to the end of the recording activity.
2.3 The hardware
As described in previous works (Pizzi et al (2004), Pizzi et al (2007)), our technique consists of
stimulating the cells by means of simulated perceptions in form of digital patterns constituted by
organized bursts of multichannel electrical stimulations.
These can be preset by means of a custom preamplifier circuit, that is shown in Fig. 3 .
Fig. 3. The 8 separated circuits of the controller
All the circuits are isolated and contained into a thick metallic box connected to the ground. In
order to avoid the presence of spurious signal, input and output signals are completely isolated by
means of special Texas Instruments chips that avoid any kind of coupling. Even the digital signals
are uncoupled from the internal circuit by means of photocouplers. In this way the MEA electrodes
are never in contact with the outside.
Four LiON batteries supply a “clean” voltage. All the system is connected to the ground.
The cells receive electrical stimulations thanks to eight shielded cables connected to the MEA,
and the same cables collect the electrical reactions of the cells after the simulated perceptions. In
Fig. 4 the controller is depicted with the eight cables connected to the MEA inside the brass box.
Fig. 4 . Thermocontrolled Faraday cage containing the MEAs is connected to the
preamplifier/controller
.
After the amplifier, the signals generated by the cells are acquired by a NI6052E National
Instruments DAQ with the following specifications: 333 kS/s , 16 bit, 16 analog inputs, 2 analog
output, 8 digital I/O lines.
3. Methods
3.1 The past experiments
In the past experiments the cells were stimulated by electrical pulses with different frequencies
and voltages (30-100 mV, suitable for neuron stimulation). On the basis of the experiments referred
in the first section of this paper and of the daily experience of clinical and research laboratories, the
stimulations are usually delivered by means of square wave pulses.
The duration of one pulse was set from 1.25 ms to 25 ms in different experiments
The pulses were delivered simultaneously on all the electrodes in the form of patterns and
delivered to the network as trains of electrical pulses in such a way as to represent black (bit 1) or
white (bit 0) squares of a bitmap, as in the example showed in Fig. 5.
Fig. 5. Example of bitmaps submitted to the biological network
In the past experiments our group verified that the cells reply selectively to this kind of patterns
(Pizzi et al (2004), Pizzi et al (2007)).
We implemented connection schemata on the MEAs resembling an artificial neural network
(ANN) architecture.
In particular, we arranged eight input channels picked from eight electrodes, on which living
cells were attached (Fig.2). The cells were cultured on the connection sites of the MEAs and
connected each others as in the case of the Hopfield network (Hopfield (1984), Tank and Hopfield
(1989))
As in the Hopfield model, the output channels coincide with the input channels, thus after
disconnecting the stimulation circuit and after a short relaxation time (around 10 ms), the output
signals were collected from the same electrodes.
We iterated the experiment with different stimulation lengths, ranging from 1.25 to 25 ms.
The experiments showed that the network reacts with different electrical behaviors depending on
the delivered pattern.
We used Recurrence quantification analysis (RQA) (Takens (1981), Zbilut and Webber (1992),
Kononov (1996), Zbilut et al (2002)) to evaluate the organization state of the biological networks
before and after the training procedure.
RQA is a non-linear statistics procedure suitable for physiological time series. The resulting
plots show how the vectors of the dynamical system represented by the time series are near or
distant each others.
This analysis shows (Pizzi et al (2004), Pizzi et al (2007)) that introduction of organized stimuli
modifies the network structure and increases and maintains the information content even after the
end of stimulation, suggesting a form of learning and memorization as in the case of ANNs.
Moreover, the RQA analysis shows that the network reacts in an organized fashion, and
behaves differently depending on the input signal and on the different channels. We observed that
signals coming from similar bitmaps gave rise to similar recurrent plots. The self-organization is
maintained after the end of stimulations.
The biological network show sto be able to answer selectively to different patterns. The signal
behavior changes depending on the network channels, and similar patterns give rise to similar
answers.
3.2 The new experiment
On the basis of the past experiments we tried a step forward, aiming to decode the signals
emitted by the neurons and to use the ability of the biological neural network to retain information
in order to respond to commands and move an actuator.
For this purpose we developed a hybrid biological-electronic system that we schematized in Fig.
6.
Fig. 6. Block diagram of the hybrid system
We arranged on the MEA eight input channels picked from eight electrodes, on which living
cells were attached. The cells were cultured on the connection sites of the MEAs and connected
each others as in the case of a Hopfield (Hopfield, 1980) artificial neural network.
The stimulation occurs with a 100 mV positive voltage followed by a brief -100 mV
depolarization pulse to avid electrolysis. The stimulation frequency is 433 Hz, the sampling rate is
10 kHz.
Each pattern is constituted by a matrix of 8 x 8 bits. Every bit lasts 300 ms. The cells are
stimulated 2.4 sec for each pattern. Each stimulation is followed by a 1 sec pause and is repeated 10
times for each pattern, in order to allow the neurons to learn.
The first phase of the experiment consisted of stimulating the neurons with a set of simulated
perceptions in form of four digital patterns, that are depicted in Fig. 7.
Rows represent the 8 simultaneously activated channels, columns represent time. Each square
represents one bit: if it contains a dot, this means that a stimulation has been activated, otherwise
there is no stimulation. At the end of the tenth stimulation, the reactions of the cells, collected after
disconnecting the stimulation circuit and after a 10 ms relaxation time, have been recorded and
sent to an Artificial Neural Network to classifies them.
The Artificial Neural Networks supplies, in form of an integer number, the direction that will
drive the robot. This number is converted into a TTL signal that is directed to the parallel port of the
PC. A suitable circuit converts it into an infrared signal that drives the robot.
It must be stressed that the ANN works just on the signals emitted by the biological neurons
10 ms after the end of the stimulation, thus the ANN’s task is not to classify the digital patterns, but
to decode the biological reactions to different patterns.
Fig. 7. The four patterns : Forward, Backwards , Left, Right
We developed a Labview graphical interface that allows to choose the commands both on an
orderly basis (for training) and on a random basis (for testing).
3.3 The ITSOM Artificial Neural Network
The model of ANN, a novel architecture called ITSOM (Inductive Tracing Self Organizing
Map), was selected considering that a self-organizing architecture was necessary, as we had not a set
of examples to train it.
The ITSOM network model is useful to highlight structures in the temporal series of a signal.
The ITSOM was tested in the past with electrophysiological signals (Pizzi et al (2002)), correctly
showing their organized structures.
The Self-Organizing Map (SOM) (Saarinen and Kohonen (1985), Kohonen (1990)) features are
well-known, as well as its limits in classifying topologically entangled input structures.
In order to overcome the SOM’s limits we developed a novel architecture based on the evidence
that, even if the SOM’s winning weights may vary at any presentation epoch, their temporal
sequence tends to repeat itself. The dynamical properties of the SOM are well known (Ritter and
Schulten (1986), Ritter and Schulten (1988), Ermentrout(1992)) and show periodic oscillations and
limit cycles.
In particular we observed that the sequence of winning weights constitutes chaotic attractors
that characterize univocally the input element that has determined them .
In this way, due to the countless variety of possible combinations among winning neurons, these
sequences allow to finely classify the corresponding input value.
A detailed description of the ITSOM’s architecture is reported in (Pizzi (1997), Pizzi et al
(2007)).
After interrupting the network processing cycle, an algorithm is needed that codifies the
obtained chaotic configurations of winning weights into a small set of outputs.
To this purpose the cumulative scores related to each input have been normalized following the
distribution of the standardized variable z given by:
z = (x − μ) σ
where μ is the average of the scores on all the competitive layer weights and σ is the root mean
squared deviation. Once fixed a threshold 0 ≤ τ ≤1, we have put
z=1
z=0
for z > τ,
for z ≤ τ .
In this way every winning configuration is represented by a binary number with as many 1’s and
0’s as many the competitive layer weights. Due to the existence of a threshold, these binary numbers
coincide every time the series of winning neurons are approximately similar.
Then the task of comparing these binary numbers is straightforward and allows to identify
similar or identical inputs.
It should be stressed that the ITSOM's crucial feature is that the network does not need to be
brought to convergence, as the cyclic configurations stabilize their structure within a small number
of epochs. The extremely low processing time makes this model very effective in case of real-time
applications.
In our experiment the ITSOM must memorize the four states representing the movement of the
robot, acquiring the signals generated by the cells during the training.
The network acquires such information by means of a matrix of floating point values (10000 x 8
for each directional pattern).
After a series of off-line trials the ITSOM was optimized with the following parameters:
500 input neurons
12 competitive layer neurons
Learning rate = 0.003
Forgetting rate = 0.001
τ=0
After finishing the training phase, the ITSOM starts generating the z-scores to compare them
with the z-scores that will be generated during the testing phase.
During the testing phase we send to the neurons several stimulations corresponding to one of
the four patterns, in a random order.
The ITSOM generates new z-scores and compares them to those stored after the training phase.
We carried out the experiment both with non differentiated cells and with mature cells.
The behaviour of the non differentiated cells showed to be random, thus the following
considerations refer to the response of the differentiated cells.
From different off-line tests on portions of the signals we observed that in the first portion of
400 ms the information content is sufficient to a correct classification. The rest of the signal
contains a more debased information and it is advisable to eliminate it.
4. Results
In the described experiment (differentiated cells) we tested the hybrid system with 25
random patterns.
In a previous experiment with a non-tuned ITSOM we collected the following results (Fig. 8):
Correct answers: 8 out of 25
Wrong answers: 13 out of 25
Non classified patterns: 4
Once we tuned the ITSOM with the above mentioned parameters, we obtained the following
results (Fig. 9):
Correct answers: 15 out of 25
Wrong answers: 7 out of 25
Non classified patterns: 3
1
2
3
4 5
6
7
8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
Fig. 8. Histogram of the classification performed by the non-tuned ITSOM: the long bars
describe the correctly classified patterns, the short bars the non classified patterns, the
absence of the bar indicates wrongly classified patterns
The low number of samples is due to the poor lifetime of the human neural stem cells in non
optimal atmosphere and
put through electrical stimulations. In fact before the 25 random
stimulations, the cells were subjected to 40 previous stimulations (10 for each command) necessary
for the training.
1
2
3
4
5
6
7
8
9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
Fig. 9. Histogram of the classification performed by the ITSOM after tuning
We can immediately observe that the classification of the last samples was less effective. The
85.7 % of the wrong answers happened in the last 9 samples, after more than 60 electrical
stimulations that could have damaged the cells.
In Tab. 1 and 2 other results of our experiment are displayed.
Observing the tables we can draw the following considerations.
The classification percentage of the “F” and “B” patterns reach high values (80% and 83.33%).
The B pattern is recognized with a 100% percentage. All the values are very far from the random
value (25%), if we calculate it already on the classifiable z-scores: but it must be stresses that the
network operates a choice among countless non classifiable z-scores. This is due both to the
ITSOM’s effectiveness and to a correct information content of the biological signals.
In order to estimate the quality of this classification we elaborated two-classes confusion
matrixes.
For each confusion matrix we can define four important parameters: False Positive (FP), False
Negative (FN), True Positive (TP), True Negative (TN).
Once we define sensitivity and specificity of a test by means of the following formulas, where
TP are true positive classifications, TN true negative, FN false negative and FP false positive
classifications:
Sensitivity = (TP / ( TP + FN))*100
Specificity = (TN / (TN + FP))*100
we obtain for the four patterns (Tab. 3):
In summary (Tab. 4), the evaluation of the proposed model presents an accuracy of 80.11% and
a precision of 90.50%.
These results allow us to consider the effectiveness of our hybrid classifier quite satisfactory.
5. Discussion and conclusions
Aim of our research is on one side to improve the knowledge of the neurophysiological
learning and memory functionalities; on the other side to evaluate the feasibility of biological
computation, or of non-invasive neurological prostheses, able to improve or substitute damaged
nervous functionalities .
We started from the assumption that the neural signal could have an information content richer
than that formed by the spike trains, therefore it could be important to analyze it in its wholeness,
considering not only the frequencies of spikes, but the wide-ranging frequencies and the amplitudes
of the whole signal.
To reach our purpose we developed a hybrid (biological-electronic) system composed by a
network of human neurons connected to an Artificial Neural Network and to a minirobot *.
A training sequence of simulated perceptions in form of electrical stimulations was delivered to
the biological network.
The output of the biological neurons constitute the input of the ANN, that classifies the
electrical signals coming from the neurons, self-organizing on the basis of the temporal series of
the whole neural electrical signals.
It is known that an artificial neural network acts as a black box able to extract the information
content of the input irrespective of the level of comprehension of the input structure.
We believe that this could be the reason why, avoiding to consider just the spike trains of the
signals but letting the ANN process the signal in its wholeness, we reached interesting results.
Despite the lack of a detailed physiological interpretation of the neural signals , the ANN
worked as a black box and managed to decode the information hidden in the neural response using
the whole signal samples.
In fact similar digital patterns gave rise to output signals containing similar chaotic attractors
corresponding to the same ITSOM code, whereas different patterns lead to attractors corresponding
to different codes.
In the described experiment the hybrid system showed to have learnt and memorized the
patterns in a satisfactory way.
In fact the hybrid system, tested with 25 random patterns has obtained a correct classification of
the four patterns with percentages respectively of 80%, 83,33%, 42,86%, 42,86%. The evaluation
of the proposed model presents an accuracy of 80.11% and a precision of 90.50%.
It is worth remarking that a trial with undifferentiated cells gave completely random results.
Analogous random results were obtained in a testing procedure carried out before the training
phase.
We can infer that a correct classification depends heavily on the degree of vitality and maturity
of the neural cells, and consequently on their ability to self-organize and memorize patterns. We
have also seen that a well-tuned Artificial Neural Network reaches better performances than a nontuned one, but the ANN alone cannot classify the neural signals unless they originate from trained
biological cells.
The wrong classifications of some patterns can be due to several factors:

an attenuated vitality of the cells

a suboptimal ITSOM tuning

an intrinsic limit of its classification algorithm

suboptimal electrical stimulation parameters.
The stimulation parameters could be improved evaluating different waveforms and
polarization/depolarization times, amplitudes, frequencies and length of the stimulations.
It should also be considered the possibility to avoid the use for stem cells and prefer neural
cells with a better resistance outside the incubator.
A better tuning of the Artificial Neural Network is underway. A new algorithm that substitutes
the z-score procedure is also under study.
The new algorithm searches for a set of maximally of winning neurons inside a restricted set of
epochs. In fact the series of winning neurons shows a global preference of the ITSOM towards
groups of neurons specific to each pattern.
During an off-line experiment the new procedure has been able to reach better performances in
the classification of the proposed patterns.
The satisfactory results obtained up to now are encouraging us to improve the complexity of our
hybrid system.
A new hardware controller is being developed that will allow to handle much more complex
patterns and a sharp increase of the number of electrode connections. Once the controller will be
completely tested, we will develop a system able to receive real perceptions from suitable sensors
and to react autonomously to environmental stimulations, exhibiting closed-loop performances. We
will start adopting a controlled environment and low-resolution sensors.
We hope that new progresses could approach us step by step to a better understanding of the
neural code and of the learning process, that in the future could be applied to technological devices
such as brain implants and bionic systems.
Acknowledgements
We are strongly indebted to Prof. G. Degli Antoni (Department of Information Technologies,
University of Milan) for his valuable suggestions and encouragement , to Dr. D. Carne, Dr. A.
Redolfi and Dr. R. Rossoni (Department of Biomolecular Sciences and Biotechnologies, University
of Milan) for their substantial contribution.
_____________________________________________________
* In honor of our town, Crema, we called our hybrid creature “Cremino”. We consider “Cremino” the first hybrid
creature endowed with a small human “brain”. An excerpt of its movements during the described experiment can be
seen in (Pizzi (2006))
References
Akin, T., Najafi, K., Smoke, R.H., Bradley, R.M., 1994. A micromachined silicon electrode for
nerve regeneration applications. IEEE Trans. Biomed. Eng. 41, 305–313.
Bakkum, D.J., Shkolnik, A.C., Ben-Ary ,G., Gamblen, P., DeMarse, T.B., Potter, S.M.(2004).
Removing some ‘A’ from AI: Embodied Cultured Networks. In: Iida, F., Pfeifer, R., Steels, L. ,
Kuniyoshi, Y.(Eds.), Embodied Artificial Intelligence. Springer New York 3139, pp. 130-145.
Bels, B., Fromherz, P., 2002. Transistor array with an organotypic brain slice: field potential
records and synaptic currents. Eur. J. Neurosci. 15, 999–1005.
Bonifazi, P., Fromherz, P., 2002. Silicon chip for electronic communication between nerve cells
by non-invasive interfacing and analog–digital processing. Adv. Mater. 17.
Borkholder, D.A., Bao, J., Maluf, N.I., Perl, E.R., Kovacs, G.T. (1997). Microelectrode arrays
for stimulation of neural slice preparations, J. Neuroscience Methods 7 , 61- 66.
Borst, A. and Theunissen, F.E.(1999). Information theory and neural coding: Natre
Neuroscience. 2, 11, 947-957.
Bove, M., Martinoia, S., Grattarola, M., Ricci, D., 1996. The neuron- transistor junction: linking
equivalent electric circuit models to microscopic descriptions. Thin Solid Films 285, 772–775.
Canepari, M., Bove, M., Mueda, E., Cappello, M., Kawana, A., 1997. Experimental analysis of
neural dynamics in cultured cortical networks and transitions between different patterns of activity.
Biol. Cybernet. 77, 153–162.
Carmena, J.M., Lebedev, M.A., Crist, R.E., O’Doherty, J.E., Santucci,
Chapin, J.K. , Moxon, K.A. , Markowitz, R.S. and Nicolelis, M.A.L. (1999). Real time control
of a robot arm using simultaneously recorded neurons in the motor cortex. Nature Neuroscience 2,
664-670.
D.M., Dimitrov, D.F., Patil, P.G., Henriquez, C.S., 2003. Learning to control a brain–machine
interface for reaching and grasping by primates. M.A.L. PLoS 1, 193–208.
De Marse, T.B., Wagenaar, D.A., Potter, S.M., 2002. The Neurally-controlled Artificial Animal:
A Neural Computer Interface Between Cultured Neural Networks and a Robotic Body. SFN 2002,
Orlando, Florida.
DeMarse, T.B., Wagenaar, D.A., Blau, A.W., Potter, S.M. (2001). The neurally controlled
animat: biological brains acting with simulated bodies. Autonomous Robots 11:305-310.
Egert, U., Schlosshauer, B., Fennrich, S., Nisch, W., Fejtl, M., Knott, T. , Müller, T.,
Hammerle, H. (2002). A novel organotypic long-term culture of the rat hippocampus on substrateintegrated microelectrode arrays. Brain Resource Protoc 2, 229-242.
Ermentrout, B., 1992. Complex dynamics in WTA neural networks
Excitable properties in astrocytes derived from human embryonic CNS stem cells. Eur. J.
Neurosci. 12, 3549–3559.
Fromherz, P., 2002. Electrical interfacing of nerve cells and semiconductor chips. Chem. Phys.
Chem. 3, 276–284.
Fromherz, P., Muller, C.O., Weis, R., 1993. Neuron-transistor: electrical transfer function
measured by the Patch-Clamp technique.
Fromherz, P., Offenhäusser, A., Vetter, T., Weis, J., 1991. A neuron-silicon junction: a Retziuscell of the leech on an insulated-gate field-effect transistor. Science 252, 1290–1293.
Fromherz, P., Schaden, H., 1994. Defined neuronal arborisations by guided outgrowth of leech
neurons in culture. Eur. J. Neurosci. 6.
Galli R, Gritti A, Bonfanti L, Vescovi A.L., 2003. Neural stem cells: an overview. Circulation
Research 92(6), 598-608.
Garcia, P.S., Calabrese, R.L., DeWeerth, S.P., Ditto,W., 2003. Simple arithmetic with firing rate
encoding in leech neurons: simulation and experiment. Proceedings of the XXVI Australian
Computer Science Conference, Adelaide, 16, 55–60.
Gritti, A., Frolichsthal-Schoeller, P., Galli, R., Parati, E.A., Cova, L., Pagano, S.F., Bjornson,
C.R., Vescovi, A., 1999. Epidermal and fibroblast growth factors behave as mitogenic regulators for
a single multipotent stem cell-like population from the subventricular region of the adult mouse
forebrain. J. Neurosci. 19 (9), 3287–3297.
Gritti, A., Galli, R., Vescovi, A.L., 2001. In: Federoff (Ed.), Culture of Stem Cell of Central
Nervous System. Humana Press III, pp. 173–197.
Gritti, A., Rosati, B., Lecchi, M., Vescovi, A.L., Wanke, E., 2000.
Hiroki T., Jun T., Akira M., Konomi K., Nobuo H., 2000. Neurons generated from adult rat
hippocampal stem cells form functional glutamatergic and GABAergic synapses in vitro .
Experimental neurology 165, 1, 66-76.
Hopfield, J.J., 1984. Neural networks and physical systems with emergent collective
computational abilities. Proc. Natl. Acad. Sci. USA, 81.
Jenkner, M., Fromherz, P., 1997. Bistability of membrane conductance in cell adhesion
observed in a neuron transistor. Phys. Rev. Lett. 79, 4705–4708.
Jenkner, M., Muller, B., Fromherz, P., 2001. Interfacing a silicon chip to pairs of snail neurons
connected by electrical synapses. Biol. Cybernet. 84, 239–249.
Jimbo, Y., Robinson, H.P.C., 2000. Propagation of spontaneous synchronized activity in
cortical slice cultures recorded by planar electrode arrays. Bioelectrochemistry 5, 107–115.
Kohonen, T., 1990. Self-Organisation and Association Memory. Springer-Verlag.
Kononov, E., 1996. http://www.myjavaserver.com/~nonlinear/vra/download.html.
Kositsky, M., Karniel, A., Alford, S., Fleming, K.M., Mussa Ivaldi, F.A. (2003). Dynamical
dimension of a hybrid neurorobotic system. Transactions on Neural Systems and Rehabilitation
Engineering 11, 155-159.
Lindner, J.F., Ditto, W., 1996. Exploring the nonlinear dynamics of a physiologically viable
model neuron. AIP Conf. Proc. 1, 375–385.
Maher, M.P., Pine, J., Wright, J., Tai, Y.C., 1999. The neurochip: a new ultielectrode device for
stimulating and recording from cultured neurons. Neurosci. Methods 87, 45–56.
Martinoia, S., Sanguineti, V., Cozzi, L., Berdondini, L., van Pelt, J., Tomas, J., Le Masson, G.,
Davide, F. (2004). Towards an embodied in vitro electrophysiology: the Neurobit project.
Neurocomputing 58-60, 1065-1072.
McKay, R.D.G., 1997. Stem cells in the central nervous system. Science 276, 66–71.
Paninski, L., Fellows, M., Hatsopoulos, N. & Donoghue, J. (2004). Spatiotemporal tuning
properties for hand position and velocity in motor cortical neurons. J. Neurophysiology 91, 515532.
Phys. Rev. Lett. 71, 4079–4082.
Pillow, J.W., Paninski, L., Uzzell, V.J., Simoncelli, E.P., Chichilnisky, E.J. (2005). Prediction
and decoding of retinal ganglion cell responses with a probabilistic spiking model. J Neuroscience
25, 11003-13.
Pizzi, R. (2006). http://www.dti.unimi.it/~pizzi/research.html
Pizzi, R., Fantasia, A., Gelain, F., Rossetti D. & Vescovi, A. (2004). Behavior of living human
neural networks on
microelectrode array support. Proceedings Nanotechnology Conference and
Trade Show. Boston.
Pizzi, R., 1997. Theory of Dynamical Neural Systems with application to Telecommunications,
PhD Dissertation, University of Pavia.
Pizzi, R., de Curtis, M., Dickson, C. (2002). Evidence of chaotic attractors in cortical fast
oscillations tested by an Artificial Neural Network. In J. Kacprzyk (Ed.), Advances in Soft
Computing. Physica Verlag.
Pizzi, R., Rossetti, D., Cino, G., Gelain, F. & Vescovi, A. (2007). Learning in Human Neural
Networks on Microelectrode Arrays. Biosystems Journal 88, 1-2, 1-15, Elsevier.
Potter, S.M.(2001). Distributed processing in cultured neuronal networks. In: Progress in Brain
Research. Nicolelis M.A.L. (Ed.). Elsevier Science B.V.
Potter, S.M., 2001. Distributed processing in cultured neuronal networks. In: Nicolelis (Ed.),
Progress in Brain Research, M.A.L. Elsevier Science B.V.
Potter, S.M., Wagenaar, D.A., DeMarse, T.B. (2005). Closing the loop: Stimulation Feeback
Systems for embodied MEA cultures, in: Taketani, M., Baundry, M. (Eds.), Advances in Network
Electrophysiology Using Multi-Electrode Arrays. Springer New York.
Reger, B., Fleming, K.M., Sanguineti, V., Simon Alford, S., Mussa-Ivaldi, F.A., 2000.
Connecting brains to robots: an artificial body for studying the computational properties of neural
tissues. Artif. Life 6, 307–324.
Rieke F., Warland D., de Ruyter van Steveninck R., Bialek W. (1997). Spikes: Exploring the
Neural Code. MIT Press, 1997.
Ritter, H., Schulten, K., 1986. On the stationary state of Kohonen’s self-organizing sensory
mapping. Biol. Cybernet. 54, 99–106.
Ritter, H., Schulten, K., 1988. Convergence properties of Kohonen’s topology conserving maps:
fluctuations, stability, and dimension selection. Biol. Cybernet. 60, 59–71.
Ruaro, M.E., Bonifazi, P. & Torre, V. (2005). Toward the Neurocomputer: Image Processing
and Pattern Recognition with Neuronal Cultures. IEEE Transactions on Biomedical Engineering, 3.
Saarinen, J., Kohonen, T., 1985. Self-organized formation of colour maps in a model cortex.
Perception, 14, 6, 711 – 719.
Schatzthauer, R., Fromherz, P., 1998. Neuron-silicon junction with voltage gated ionic currents.
Eur. J. Neurosci. 10, 1956–1962.
Schiff, S.J., Jerger, K., Duong, D.H., Chang, T., Spano, M.L., Ditto, W., 1994. Controlling
chaos in the brain. Nature 8, 25.
School, M., Spro¨ssler, C., Denyer, M., Krause, M., Nakajima, K., Maelicke, A., Knoll,W.,
Offenha¨usser, 2000. Ordered networks of rat hippocampal neurons attached to silicon oxide
surfaces. Neurosci. Methods 104, 65–75.
Song, H., Stevens, C.F., Gage, F.H., 2002. Neural stem cells from adult hippocampus develop
essential properties of functional CNS neurons. Nature Neurosci. 5, 5.
Takens, F., 1981. Detecting strange attractors in turbulence. In: Rand, D.A., Young, L.S. (Eds.),
Dynamical Systems and Turbulence—Lecture Notes in Mathematics, 898. Springer-Verlag.
Tank, D.W., Hopfield, J.J., 1989. Neural Architecture and Biophysics for Sequence Recognition
in Neural Models of Plasticity. Academic Press.
Taylor, D.M., Tillery, S.I.H., Schwarz, A.B. (2002). Direct cortical control of 3D
neuroprosthetic devices. Science 296, 5574, 1829-1832.
Vescovi, A.L., Parati, E.A., Gritti, A., Poulin, P., Ferrario, M.,Wanke, E., Fr¨olichsthalSchoeller, P., Cova, L., Arcellana-Panlilio, M., Colombo, A., Galli, R., 1999. Isolation and cloning
of multipotential stem cells from the embryonic human CNS and establishment of transplantable
human neural stem cell lines by epigenetic stimulation. Exp. Neurol. 156, 71–83.
Wagenaar, D.A. and Potter SM (2004) . A versatile all-channel stimulator for electrode arrays,
with realtime control. Journal of Neural Engineering 1, 39-45
Wagenaar, D.A., DeMarse, T.B. and S. M. Potter, S.M.(2001). A toolset for realtime analysis of
network dynamics in dense cultures of cortical neurons . 7th JSNC, University of California at San
Diego, La Jolla.
Weis, R., M¨uller, B., Fromherz, P., 1996. Neuron adhesion on silicon chip probed by an array
of field-effect transistors. Phys. Rev. Lett. 76, 327–330.
Wilson, R.J., Breckenridge, L., Blackshaw, S.E., Connolly, P., Dow, J.A.T., Curtis, A.S.G.,
Wilkinson, C.D.W., 1994. Simultaneous multisite recordings and stimulation of single isolated
leech neurons using planar extracellular electrode arrays. Neurosci. Methods 53, 101–110.
with slow inhibition. Neural Networks 5, 415–431.
Zbilut, J.P., Thomasson, N.,Webber Jr., C.L., 2002. Recurrence quantification analysis as a tool
for nonlinear exploration of nonstationary cardiac signals. Med. Eng. Phys. 24, 53–60.
Zbilut, J.P., Webber, C.L., 1992. Embeddings and delays as derived from quantification of
recurrent plots. Phys. Lett., 171.
Zeck, G., Fromherz, P., 2001. Noninvasive neuroelectronic interfacing with synaptically
connected snail neurons on a semiconductor chip. Proc. Natl. Acad. Sci. 98, 10457–10462.
Tab. 1. Robot Performances
Directions
Input
Total
Pattern F
Pattern B
Pattern L
Pattern R
Correct classification
4
5
3
3
15
Wrong classification
1
0
3
3
7
No classification
0
1
1
1
3
6
7
7
25
Yielded patterns
5
% Classified
100%
83.33%
85.71%
85.71%
88%
% Correctly classified
80%
83.33%
42.86%
42.86%
60%
Tab. 2. Classification Percentage
100%
80%
60%
No classification
40%
Wrong classification
20%
Correct classification
0%
Pattern F Pattern B Pattern L Pattern R
Directions
Totale
Tab. 3. Confusion Matrixes of the model
Confusion matrix of a pattern « P »
P
Non-P
P
TP
FN
Non-P
FP
TN
Confusion matrix pattern B
Confusion matrix pattern F
F
Non-F
F
4
0
Non-F
1
17
Confusion matrix pattern L
L
Non-L
L
3
1
Non-L
3
15
B
Non-B
B
5
6
Non-B
0
11
Confusion matrix pattern R
R
Non-R
R
3
0
Non-R
3
16
Tab. 4. Total Sensitivity and Specificity
Pattern
Pattern
Pattern
Pattern
F
B
L
R
Total
Sensitivity
100%
45.45%
75%
100%
80.11%
Specificity
94.44%
100%
83.33%
84.21%
90.50%