Olli V. Lounasmaa, Low Temperature Laboratory, Helsinki University of Technology,

02015 HUT, Finland

Heikki Seppä, Measurement Technology, VTT Automation, 02015 VTT, Finland

Keywords: Brain, Cardiomagnetism, Current dipole, Flux transformer, Functional magnetic resonance imaging, Heart, Imaging techniques, Josephson junction, Magnetic field, Magnetic resonance imaging, Magnetic shielding, Magnetocardiography, Magnetoencephalography, Magnetometer, Multichannel instruments, Neuromagnetism, Pick-up coil, SQUID, superconductors

1. General comments

2. Magnetoencephalography and magnetocardiography

2.1. Introduction

2.2. The human brain

2.3. Analysis of MEG and MCG measurements

2.4. Magnetic shielding

3. SQUIDs

3.1. Properties of superconductors

3.2. The rf SQUID

3.3. The dc SQUID

3.4. The pick-up loop and the flux transformer

3.5. Manufacturing technology

3.6. A modern SQUID

3.7. Other SQUID designs

4. Multichannel instrumentation for MEG

4.1. The Neuromag system

4.2. The CTF Systems neuromagnetometer

4.3. The neuromagnetometer of Biomagnetic Technologies

5. Examples of MEG studies

5.1. Somatosensory responses to finger stimuli

5.2. Dynamic brain activation during picture naming

5.3. Face-sensitive areas in the human brain

5.4. Action viewing

5.5. Increase of cortical representation with use

5.6. Clinical applications

6. Cardiomagnetic studies

6.1. The human heart

6.2. Methods and instrumentation in magnetocardiography

6.3. Detection of heart abnormalities

6.4. Estimation of life-threatening arrhythmia

6.5. Discussion on MCG

7. Comparisons of MEG, MCG, EEG, ECG, PET, and fMRI

8. Conclusions

9. Acknowledgments

10. Bibliography

Action potential: The duration of an action potential (AP) in the heart is 300…400 ms, and the APs spread electrically from cell to cell in a synchronized manner.

Arrhythmia: A disturbance in the normal heart rhythm. Tachyarrhythmia is a disturbance where the heart rate rises up to 200…300 beats per minute, while in a bradyarrhythmia the heart rate is slowed.

Biomagnetic Technologies Inc: Pacific Heights Boulevard, San Diego, CA 92121.

CAT: Computer assisted tomography. A revolving X-ray tube and a bank of counters are used to create a picture by analyzing the recorded X-ray intensities by means of a computer.

CTF Systems Inc: Port Coquitlam, Canada V3C 1M9.

Current dipole: A pointlike current element, with a position and direction, is a current dipole. Thousands of parallel currents through neighboring neurons are needed to form a current dipole which is sufficiently strong to be detected by MEG.

DC: Direct current.

ECG: Electrocardiography. Electrodes are placed at suitable points on the subject's chest. See also the description for EEG.

EEG: Electroencephalography. Electrodes placed on the scalp measure the same neuronal currents as in MEG, in theory. But since the electric-potential distribution is distorted by the skull, the scalp and other tissues that have electrical conductivities that differ from those in the brain itself, EEG is better for detecting currents that originate deep inside the brain or are radially oriented.

FET: Field effect transistor, described in any modern textbook on electronics.

Fissures: The grooves in the cortical surface are called fissures. Since the current dipoles in the brain are perpendicular to the surface of the cortex, dipoles in the walls of the fissures are tangential to the surface of the head. Therefore, MEG primarily measures activity in the fissures. This is, however, not so serious because 2/3 of the cortical surface is in the fissures and because the most important regions (for instance the somatosensory, motor, auditory and visual cortices) are mainly in the fissures.

Flux transformer: A superconducting flux transformer is used for transmitting magnetic flux from the region where measurements are made to the SQUID. The flux transformer is based on the fact that the magnetic flux, which penetrates a surface surrounded by a superconductor, is constant.

fMRI: functional magnetic resonance imaging. See MRI.

HUCH: Helsinki University Central Hospital, 00029 HUCH, Finland

HUT: Helsinki University of Technology, 02015 Espoo, Finland.

IBM: International Business Machines Inc.

Intercellular current: The current between cells.

Inverse problem: By measuring the magnetic field outside the head one can, in principle, use a computer to calculate the sources in the brain which have generated the field. Unfortunately, however, there is an infinite number of solutions to this so called inverse problem, as was shown by Hermann von Helmholtz 150 years ago. This is a mathematical fact, the problem is "ill posed". In order to get a definite solution, special conditions must be assumed. Usually it is postulated that the source within the spherical brain is a current dipole or several such dipoles. Then the locations, directions and strengths of the dipoles can be calculated uniquely.

Josephson junction: A weak link between two superconducting electrodes.

LTL: Low Temperature Laboratory at HUT.

MCG: Magnetocardiography. Secondary subject of this article.

MEG: Magnetoencephalography. Main subject of this article.

MRI: Magnetic resonance imaging. The subject is exposed to a high static magnetic field and a small alternating radiofrequency field. Magnetic resonance signals produced by hydrogen nuclei in different tissues are measured, either to create an image of proton density in the anatomical structures (conventional MRI) or to monitor changes related to brain metabolism and blood flow (functional MRI).

MSI: magnetic source imaging; another name for MEG.

Mu-metal: Mu-metal is an alloy of iron and nickel and small amounts of other metals, which has a high magnetic permeability, considerably higher than soft iron. It is used for magnetic shielding of the closed volume where MEG measurements are made. Magnetic field lines are "trapped" in the mu-metal instead of the free space inside.

PTB: Physikalische-Technische Bundesanstalt (10587 Berlin, Germany).

QRS complex: The tri-phasic main waveform in an ECG or MCG recording, caused by depolarization of the ventricular muscle cells.

Source modelling: When experimental data are mathematically analysed, it is necessary to assume the basis on which the formulae were derived. In MEG, it is usually assumed that the source is a current dipole in a spherical volume conductor.

SQUID: Superconducting QUantum Interference Device.

Tachycardia: Abnormally fast heartbeat.

Tangential dipole: It is a mathematical fact that dipoles radial to the nearest surface of the brain do not produce a magnetic field outside. Tangential dipoles, on the other hand, produce a field outside.

VTT: Technical Research Centre of Finland (02015 VTT, Finland).

Activity of the brain and heart involves small electrical currents which flow in the biological tissue. These currents, in turn, produce tiny magnetic fields which can be measured outside the body, using ultra-low-noise Superconducting QUantum Interference Devices (SQUIDs). Two related techniques, magnetoencephalography (MEG) and magnetocardiography (MCG) have emerged to enable non-invasive utilization of the magnetic data. Clinical applications of both methods have evolved and further developments are in progress. MEG is sometimes called magnetic source imaging (MSI).

A great deal has been learned already about the cerebral anatomy and physiology, but the fundamental question of how the brain stores, retrieves, and processes information is still largely unknown. Neural computers are the future tools of designing and decision making. Recent developments in this new field of computing are one important source of inspiration to study the functional principles of the human brain, in addition to many interesting questions of basic and clinical neuroscience. For the first time it is now possible to investigate information processing in the living human brain, without philosophical speculations.

The heart in conceptually a much simpler device than the brain but, nevertheless, cardiology is a very important field of medical research. Heart deseases are the number-one cause of deaths in developed countries.

This paper deals mainly with SQUIDs and their use in MEG studies. We naturally emphasize the MEG- and instrumentation work carried out under Riitta Hari and Olli Lounasmaa in the Low Temperature Laboratory of HUT, and the development of SQUIDs done under Heikki Seppä in VTT. A brief section towards the end describes the experimental arrangement and some results of MCG studies of the heart, emphasizing work carried out under Toivo Katila in the Laboratory of Biomedical Engineering of HUT and in the BioMag Laboratory jointly owned by HUCH and HUT.

Figure 1 about here

Magnetoencephalography and magnetocardiography are fully based on the SQUID which is a superconducting ring interrupted by one or two Josephson junctions. Phase coherence of charge carriers in a superconductor gives rise to flux quantization in the SQUID ring. Owing to the smallness of the flux quantum, about 2.07 fWb, the SQUID can be made into a very sensitive parametric amplifier.

Pioneers in the MEG field are James Zimmerman, the inventor of the practical SQUID in 1970, David Cohen who, with Zimmerman, first detected magnetic brain signals, and Samuel Williamson whose group did much of the earlier work. At present, about 100 laboratories worldwide are engaged in magnetoencephalographic studies of the human brain; some of the interdisciplinary teams consist of a dozen or more researchers, including physiologists, physicists, psychologists, linguists, and information scientists. Comprehensive reviews of the MEG technique and of experimental data are available (see Bibliography). The state-of-the-art in biomagnetism is reflected in the proceedings of the biennial International Conferences on Biomagnetism.

Figure 2 about here

Most regions of the cerebral cortex have been mapped functionally in animals and also in man during brain surgery. The area just in front of the nearly vertical Rolandic (= central) fissure, in the frontal lobe, is the motor cortex, concerned with control and integration of muscular activity. Behind this fissure, in the parietal lobe, the primary somatic sensory cortex is located. This structure is involved with the reception and analysis of information from the skin, joints, and muscles. The left and right auditory cortices, in the temporal lobes, are largely buried within the Sylvian fissures. The primary visual cortex is in the occipital lobe at the back of the head. Wernicke's and Broca's areas, which are important for the analysis and production of speech, are also indicated in Figure 2. Most of the remaining cortical regions are known as association areas where inputs from several sensory modalities are integrated.

Signals in the brain are processed by neurons, which communicate with each other via weak electrical impulses. The neurons are able to generate so-called action potentials, which are voltage pulses of about 100 mV that last for about 1 ms. Action potentials travel along axons, the long branches of individual neurons. When the pulse reaches a synapse where two neurons meet, it causes the release of chemical neurotransmitters that trigger a postsynaptic current in the next neuron. This allows transmission of interneuronal currents to continue through the brain.

Owing to spontaneous brain activity and incoherent "noise", individual signals resulting from 20 to 500 successive sensory stimuli are usually averaged. In this way the evoked response, for example after "beeps" of sound, emerges from the background. To collect enough data for calculating the magnetic field distribution, measurements must be made at many, typically at 50 to 100 different locations. In earlier MEG experiments this was done sequentially, over one or a few sites at a time. This was a slow procedure which also endangered the reliability of the data: The subject could not be expected to remain in the same state of vigilance throughout a long measurement session. Now this "human" problem has been overcome by the use of neuromagnetometers covering simultaneously the whole head (see Chapter 4).

Figure 3 shows, at left, a typical magnetic signal evoked in the auditory cortex by 50-ms tones presented to the subject's left ear. The field gradient, in fT/cm, is plotted as a function of time. By combining such signals from many locations, a computer can calculate the magnetic field distribution shown at right. The active region in the brain, underneath the center of the red arrow, may be determined from this type of data. The field pattern varies with time because the activated region in the cortex changes.

Figure 3 about here

In interpreting MEG data, one is dealing with the electromagnetic inverse problem, i.e., with the calculation of the source currents responsible for the externally measured magnetic field. The inverse problem has no unique solution. One must, therefore, use a suitable source model to interpret the data. When the neural currents are confined to a small region in the cortex one may assume that, looking from the outside, the source is a current dipole. The magnitude, direction, and position of the active brain area can then be deduced unambiguously with a least-squares fit of the experimental data as shown on the right side of Figure 3. The dipole lies in the middle of the two field extrema, at a 90° angle to the line joining them.

Since biomagnetic signals are small, rejection of all external disturbances is of extreme importance. Other body parts also generate signals. For example, the heart's magnetic field on the chest is two to three orders of magnitude larger than the brain signals just outside the head. The vulnerability of the system to external magnetic noise can be reduced greatly by a gradiometric design of the superconducting flux transformer which gathers the magnetic signal and transmits it to the SQUID. Two configurations are in common use (see Section 3.4): a flat figure-of-eight geometry and two axial but oppositely wound coils in series. Both loop systems are insensitive to spatially uniform background fields but respond to changes in an inhomogeneous field as generated by the nearby brain.

In addition, MEG experiments are usually carried out inside a magnetically shielded room, which in the LTL at HUT employs for noise reduction two layers of aluminum, that attenuate very effectively the high frequency band of the external magnetic disturbances, and two layers of mu-metal for rejection of noise at low frequencies. In addition, the magnetic field entering the outer mu-metal shield is detected by demagnetization coils. A controller and Helmholtz-type feedback coils outside the room are employed to compensate for low frequency magnetic fields. There is also a special feedback circuit to cancel the 50 Hz magnetic field.

3. SQUIDs

Since the late 1960's, SQUIDs have been familiar devices in most low temperature laboratories. Multi-SQUID applications, however, have materialized only after 1980, owing to the slow progress in fabrication technology and small demand. Today, reliable and stable thin-film Josephson circuits can be manufactured in large quantities. The cost of one complete sensor unit, including the dc SQUID and the pick-up loop, is about $ 150, but the price will be reduced to $ 50 if production is markedly increased.

There are several ways to accomplish a flux detector based on Josephson junctions, but only the rf and dc SQUIDs will be discussed here.

When a radio frequency flux is superimposed, via a tank circuit, on the low frequency flux threading the superconducting ring, dissipation in the system becomes proportional to the signal flux. This is because the phase of the sinusoidal pump signal, at the moment when the Josephson junction is switched to its resistive state, depends on the low frequency flux. By monitoring the amplitude of the rf signal across the tank circuit, the applied flux can be measured. Since the change in the voltage amplitude of the tank circuit is reasonably high, several hundres of microvolts per flux quantum, this so called hysteretic rf SQUID became the first practical SQUID magnetometer. Owing to its complicated readout electronics the rf SQUID is no longer used in commercial neuromagnetometers.

The interference pattern in a ring with two junctions was first demonstrated in 1965. As an example, Figure 4a shows the current-voltage characteristics of a dc SQUID for three different applied fluxes. The current in a voltage biased SQUID, as a function of the external magnetic field, is illustrated in Figure 4b.

Figure 4 about here

Owing to quantization of the flux in a superconducting ring, a closed loop can be used to pick-up the magnetic field from a large area, close to signal sources, and to transfer it into the SQUID ring. A multiloop signal coil wound around the SQUID compresses the magnetic field into the SQUID ring. The supercurrent induced by the applied field is inversely proportional to the inductance of the flux transformer and thus the signal coil inductance, i.e, the number of turns, should be as small as possible. On the contrary, high mutual inductance between the signal coil and the SQUID ring maximizes the flux threading the SQUID ring and thus a compromise is necessary. A simple calculation shows that inductance matching between the pick-up loop and the signal coil maximizes the transfer efficiency to convert magnetic field to flux in the SQUID ring.

There is variety of ways to accomplish a pick-up loop. High inductance is realized by using a multiloop configuration but connecting the loops in parallel leads to low inductance. Since the SQUID inductance is very small, the signal coil is made of several turns in series, in order to secure a high signal coil inductance.

The pick-up coil can be a single loop sensing homogenous magnetic field or it may comprise of two or more loops in series but in opposite directions sensing thus only gradients of the field. Gradiometers are easier to operate in a noisy environment, because the rapid decay of the gradient field as a function of distance makes the sensor most sensitive to nearby source currents, provided that the gradiometers are balanced sufficiently well. Immunity to external noise sources can further be improved by using a pick-up unit comprising of three separate coils. This arrangement leads to a second order gradiometer which is insensitive not only to the homogenous field but also to its gradient. Higher order gradiometers are employed in biomagnetic studies which are performed in unshielded environments. Magnetometers, on the other hand, are good for extracting information about deep sources but they unavoidably respond to distant noise sources, too.

Several possibilities of arranging magnetometers and first order gradiometers are shown in Figure 5. The tangential and planar gradiometers can be integrated on a silicon chip but the axial configuration requires a flexible board or it can be made of a superconducting wire.

Figure 5 about here

To avoid the problems of wire-wound flux transformers, which are expensive to manufacture and have only a modest balance, planar integrated sensors offer an elegant solution. The main advantages of flat thin-film gradiometers are their compact structure and excellent dimensional precision, providing a good intrinsic balance. Figure 6 depicts a SQUID-system designed and fabricated by the Technical Research Centre of Finland (VTT). The sensor is specially optimized for the frequency range between 0.1 and 1000 Hz, which is important for biomagnetic measurements. The white flux noise corresponds to an intrinsic energy resolution of 5h, where h is Planck's constant, the theoretical limit. The VTT SQUID, which is one of the best, will be described in some detail below.

Figure 6 about here

The sensor optimization for neuromagnetism is rather complex because many parts make up the complete system: the SQUID with the signal coil, the pickup coil, the inductance matching circuitry, the high frequency damping circuitry, the output transforming circuits, and the readout amplifier. The sensor should have sufficiently low magnetic field noise at the frequency band of interest and it must be sensitive to the field components that carry information about the source currents in the subject's brain or heart. The system should also have enough tolerance against environmental disturbances, and be robust and easy to manufacture. The chosen strategy on one of the parts unavoidably restricts choices available in designing the other parts. Furthermore, most of the design parameters influence each other. Due to these facts, it is best to optimize the sensor as a whole rather than divide it into subsystems optimized separately.

The pick-up loops collect magnetic field over a large area and feed it to the small SQUID loop. The larger the area, the smaller is the effective magnetic field noise. The 28-mm chip size, even when including gradiometers, allows noise levels to be comparable to the parasitic noise sources in the vicinity of the sensor, such as the thermal radiation shields of the dewar. Owing to these noise sources, enlarging the coil size further would not increase sensitivity but rather limit the highest field that can be tolerated owing to the non-linearity of the SQUID. The loop size is small enough to be implemented economically on a silicon chip. The vacuum gap of the dewar and the patient's skullbone separate the sensors and the source currents in the brain by about 35 mm. Therefore, the sensors can be assembled close enough to each other to permit sufficient spatial sampling of the magnetic field.

The SQUID loop with its three Josephson junctions resides on the SQUID chip, and on top of it is a 16-turn signal coil which forms the first stage of the inductance-matching transformer. The inductance of the SQUID loop is chosen to be 8 pH, as small as can be manufactured without degrading its coupling to the signal coil too much. This low inductance needs to be matched to the 10000 times larger pick-up inductances. A single stage transformer would need a very large number of turns, which would create a lot of parasitic capacitance across the loop. Thus a two-stage transformer is used. The second transformer stage occupies most of the SQUID chip area.

The energy resolution of the SQUID in the white noise region is proportional to the product of the loop inductance L and the junction capacitance C. The magnetic field noise referred to the pick-up coil, on the other hand, is independent of L because the flux transfer coefficient decreases at the same rate as the energy resolution increases with diminishing loop inductance. However, a small L is desirable because it reduces the 1/f noise caused by critical current fluctuations in the junctions. The second benefit of a small L is an increase in the dynamic range, which is particularly advantageous when magnetometer-type pick-up coils are operated in a noisy environment.

The resistive terminations of the transmission lines created by the signal coil and by the pick-up loop circuitry produce noise which must be taken into account in sensor optimization. The improvement in the magnetic field noise, when the pick-up loop area is increased, is partially offset by this noise: longer transmission lines are then necessary and a lower line impedance is needed in order to avoid stray inductances. The required lower termination resistance gives rise to higher noise currents, impairing the field sensitivity.

Simplicity in the SQUID readout circuitry is a clear benefit when a large number of sensor channels must be employed in a single instrument. In a simple direct-readout scheme there is, however, the risk that the noise of the room-temperature amplifier dominates the system noise, since the SQUID output signal is rather weak and its output impedance is too small for a low noise room temperature amplifier. A small SQUID loop inductance helps since it increases the output signal.

If a circuit composed of an additive feedback coil and a cooled FET, acting as a tunable resistor, is set across the voltage biased SQUID, the so called "noise cancellation electronics" is obtained (see Figure 7). The circuit converts the noise voltage originating from the preamplifier into a flux and, furthermore, the SQUID converts it into a current. Finally, the voltage biased SQUID output current is converted to a voltage at the output of the preamplifier by a feedback resistor. The noise voltage at the input of the preamplifier conributes at the output directly and indirectly via the SQUID. The signals have an opposite sign, therefore, and by tuning the channel resistance of the FET, the contribution of the input noise voltage at the ouput of the SQUID can be made negligibly small.

Figure 7 about here

Tanks to the noise cancellation electronics, the readout system is very simple and thus suitable for mutichannel devices. Unfortunately, lack of bias modulation prevents elimination of the 1/f noise, which originates from critical current fluctuations. This is why very large Josephson junctions placed close to the SQUID washer were used.

The VTT SQUID sensor described above has proven to be quite succesful. With several thousand sensors in operation around the world right now, it may be the most used SQUID device ever made.

Figure 8 about here

The main advantage of the PTB magnetometer is its simple construction; only one chip is needed to accomplish the complete magnetometer. Owing to the high SQUID inductance, the flux-to-current conversion efficiency is lower than that of a low inductance SQUID. Consequently, monitoring the weak output signal with room temperature electronics becomes more difficult. In addition, the flux noise of the high inductance SQUID is higher, which reduces the available dynamic range.

To amplify a weak output signal without adding noise, the feedback technique was adopted. An additive feedback coil converting the output voltage into a flux via an external resistor was integrated into the SQUID chip. Owing to direct external feedback the second slope of the periodic flux-to-voltage response becomes steeper increasing the efficiency. Since an integrated non-tunable resistor was used in the PTB design to convert the output voltage to a current, the positive feedback was controlled by the current bias.

The PTB device thus replaces the pick-loop, the flux trasforming circuits, and the SQUID with a simple multi-loop SQUID and increases the gain of the SQUID via positive feedback with integrated circuitry. The PTB system forms an elegant and compact magnetometer.

Figure 9 about here

Figure 10 about here

Figure 11 about here

5. Examples of MEG studies

We will next describe briefly seven recent MEG experiments. The technique has been used mainly to study the way the brain processes signals, such as those arising from our senses of hearing, touch, smell and pain, as well as those associated with voluntary movements of the body. MEG has also been employed to examine the neural basis of more complex brain functions, such as attention, sensory memory and language. Magnetoencephalography is at its best for measuring signals from the cerebral cortex but, by assuming the source locations in advance, clear responses from the cerebellum, thalamus and hippocampus have been recorded by Claudia Tesche of the LTL at HUT.

Figure 12 about here

By naming objects humans try to bring order into their environment. This important process necessarily involves several regions of the brain, including those specific to speech analysis. In the LTL of HUT, picture naming was investigated in six subjects. The experimental procedure was as follows: Line drawings of common items were presented, once every 5 s, on a computer screen, seen through a hole in the wall of the magnetically shielded room. The subjects were instructed to pronounce, as soon as possible, the names of the objects shown.

Figure 13 illustrates how the dominant active area in the brain changed with time in one subject. The early visual response (Source 1) at the back of the head was followed by more lateral posterior signals in the left and right hemispheres (Sources 2 and 3), suggesting a contribution from Wernicke's area (see Figure 2), which is important for understanding a language, and its right-sided counterpart. Within the same time window (gray area in the magnetic field versus time plots), a left frontal site (4) close to the cortical face representation area showed activation, as in preparation for mouth movements.

Figure 13 about here

About 500 ms after the picture had been shown, signals emerged from a site anterior to the face representation region in the motor cortex, reflecting activation of Broca's area (5), which is known to be involved in vocalized word production, and its counterpart in the right hemisphere (6). These signals were immediately followed by activity at the top of the brain (7), probably generated in the supplementary motor cortex which programs complex tasks such as articulation. Information processing by the brain during picture naming is beautifully demonstrated by the experimental sequence in Figure 13. Only the MEG technique has sufficient resolution to reveal the order of events.

Figure 14 about here

In other words, when we see someone performing a particular action, it triggers neuronal activity in the same areas of our own brain that we would use to perform that action ourselves. These "mirror neurons" are believed to form the basis of non-verbal communication between social animals, including humans, who spend much of their waking time in trying to predict the intentions and feelings of other members in their social group. MEG studies have suggested (see Figure 15) that Broca's region, which is a speech producing area in the left hemisphere of the human brain, plays a central role in this system.

Figure 15 about here

When playing music, the second to the fifth digits in the left hand of violin players are continuously fingering the strings. This task involves considerable manual dexterity and increased sensory stimulation. The right hand, which manipulates the bow, participates in a task involving much less individual finger movement. Therefore, in string players, inputs to the right and left somatosensory cortices differ considerably.

Thomas Elbert and his colleagues, while at the University of Münster in 1995, used MEG to study stringed-instrument players who had started their musical training at an early age. The group found that the evoked fields produced in response to tactile stimuli, applied to the fingers on the left hand, were much stronger in musicians than in non-musicians. They also observed that the strength of the response depended on the age at which the subjects had started playing the violin or cello. The result suggests that the cortical territory, occupied by the representation of the left hand digits, had increased in string players as compared with that of controls.

Clinical uses of the MEG method are emerging rapidly. Up to now, the most important applications are in the study of patients suffering from epilepsy and brain tumors.

People who suffer from epilepsy can usually be helped with drugs, but if medication does not work, neurosurgery is an option. However, before the operation takes place it is vital for the surgeon to know more about the epileptic "discharges" that manifest themselves in such patients. These discharges are sudden "electrical storms" that occur in the brain. In many patients, epileptic activity arises in a small local area, from which it rapidly spreads throughout the cortex. The discharges are much stronger, up to 1 pT, than evoked responses. Therefore, they can be detected directly without signal averaging.

MEG has been used to study these electrical storms, and it can help the surgeon to ascertain whether the spikes arise from restricted areas of the brain and, if so, how many areas are involved and how closely in time they fire. MEG can also give information about how near these areas are to such parts of the brain that should be left intact during an operation, for example the regions that control movements and speech. It is also important to determine how stable the distribution of the electrical discharges is with time.

The patient of Figure 16 obviously has two epileptogenic areas in the right hemisphere of his brain, one in the parieto-occipital lobe and another in the temporal lobe. In both locations the generators of epileptic discharges are distributed over fairly large areas. In this patient, the part of the parieto-occipital lobe that showed atrophy in MRI was operated on, which resulted in over 80% reduction in epileptic seizures.

Figure 16 about here

Combination of brain's structural data from MRI and functional information from MEG is today routinely used in some hospitals for preoperative examination of patients before surgical intervention. Figure 17 shows, as an example, MEG localization of the somatomotor cortex in a patient with a brain tumor. Somatosensory stimulations of foot, hand and lip and contraction of the forearm muscles was used to determine the course of the central sulcus. The noninvasive functional localization and the display of cortical blood vessels on the MRI reconstruction help the neurosurgeon in selecting suitable patients, in planning the operation and in navigation during actual surgery. The LTL of HUT and the HUCH have collaborated to combine anatomical and functional information in preoperative evaluation of over sixty patients with very encouraging results.

Figure 17 about here

The rhythmic sequence of the human heart (see Figure 18) is controlled by the sino-atrial (S-A) node, which initiates the electric pulse that subsequently triggers the ionic currents causing muscle contractions. The activation propagates through the atria to the atrio-ventricular (A-V) node, which slows down the entry of the pulse to the ventricular muscles so that the atria are emptied of blood before the filled ventricles contract. The sinus node acts as the heart's pacemaker; its firing rate is regulated by input from the autonomous nervous system.

Figure 18 about here

When an electrical impulse travels across the heart it causes a time-dependent action potential. Currents responsible for muscle contractions in the heart are partly transmitted to the surrounding tissue and some of the electrical activity reaches the body surface, where it can be measured as a potential difference between suitably placed electrodes. The output, voltage versus time, is called an electrocardiogram (ECG).

The Laboratory of Biomedical Engineering at the Helsinki University of Technology has carried out a large number of MCG studies in close collaboration with cardiologists at the Helsinki University Central Hospital. Their joint venture, the BioMag Laboratory, has now been operating for five years near the clinics. For heart studies, the hospital unit is equipped with a state-of-the-art 67-SQUID cardiomagnetometer, specifically designed for patients (see Figure 19). The system allows studies of very weak signals, normally hidden under experimental noise. The 30-cm diameter coverage also offers a unique possibility to explore the spatiotemporal dynamic changes from one heartbeat to the next. The position of the dewar containing the SQUID-sensors, in respect of the patient's torso, is determined by measuring the magnetic field produced by small marker coils attached to the chest.

Figure 19 about here

Even though magnetocardiography is not yet a routine clinical tool, non-invasive MCG investigations have been reported on a number of clinically important cases. These include localization of arrhythmia-causing regions of the heart, assessment of the risk of life-threatening arrhythmias, and characterization and localization of myocardial ischemia. Recently many groups have investigated fetal MCG's which can indicate heart abnormalities in unborn children.

Figure 20 depicts a recording made in the BioMag Laboratory using the 67-channel cardiomagnetometer. After its installation in April 1995, the system has been used in over 800 patient studies. Typically, data are collected for 5 min, and the whole measurement procedure requires less than 30 min. The main focus in these investigations has been on various cardiac arrhythmias. The example shown exhibits a spontaneous abnormal but large beat following the regular cardiac cycle. Such additional contractions usually originate in the ventricles and can sometimes trigger life-threatening arrhythmias of over 250 heart beats per minute.

Figure 20 about here

The most promising application of magnetocardiography so far has been the localization of sites where cardiac arrhythmias are generated. Multichannel measurements allow feasible and quick patient studies. MR imaging provides the necessary link to relate the MCG localization results accurately to the anatomy of the heart. The clinical locating accuracy of the MCG method has been tested by numerous authors for localizing the site of the earliest abnormal ventricular activation in patients with the Wolff-Parkinson-White syndrome or other focal ventricular arrhythmias. When drug therapy does not cure the problem, the area causing abnormal activity must be invasively eliminated. In addition, MCG localization has been reported in patients suffering from atrial tachycardias. An example is presented in Figure 21; the result was in excellent agreement with the invasive localization obtained during catheter ablation.

Figure 21 about here

Disparity and variability of ventricular repolarization is a good measure of sudden cardiac death. Figure 22 is an example of detection of abnormal repolarization by magnetocardiographic mapping in a 14 year-old young woman, suffering from an inherited cardiac desease called the long-QT syndrome. The normal and abnormal T-wave patterns are clearly different.

Figure 22 about here

Recently an MCG group in Berlin has made an interesting approach to estimate a patient's future risk of serious arrhythmia. Their method is based on spatial distribution of the high-frequency components in magnetocardiographic signals during the QRS complex. The data were recorded in a magnetically shielded room using 49 magnetogradiometer channels in a low noise SQUID system. A score S was defined as a measure of fragmentation in the bandpass-filtered QRS complex.

The S-score was examined to determine its sensitivity and specificity for discriminating between 34 healthy volunteers, 42 post-myocardial infarction patients, and 43 patients with coronary heart disease and with a history of malignant sustained ventricular tachycardia (VT) or ventricular fibrillation (VF). The multichannel information was visualized by two-dimensional mapping of the S-scores in single channels (see Figure 23). By averaging the S-scores for the seven central channels and for all 49 channels, and by calculating the standard deviation for all channels, a higher sensitivity and specificity for identifying patients with VT or VF was reached than by an analysis of single channels. Combination of these parameters furnishes a sensitivity of 90% and a specificity of 70% for identifying patients prone to VT/VF.

Figure 23 about here

In addition, several groups have reported signal-processing methods in the time and frequency domains, resulting in satisfactory results for detecting patients with arrhythmia risk in symptom-free persons.

Magnetocardiography has many advantages to become a routine clinical method. The technique is totally non-invasive, and it is not necessary to attach electrodes or other sensors in contact with the patient. With a multichannel cardiomagnetometer, a clinical study can be carried out in less than five minutes. The spatio-temporal resolution of MCG is much higher than that of conventional ECG.

There are, nowadays, many methods to study the intact human brain and heart. Anatomical structures can be investigated precisely by means of computer assisted X-ray tomography, CAT, and by magnetic resonance imaging, MRI. Both these techniques provide high quality but static pictures of living organs. Information about the working brain or heart can be recorded by means of positron emission tomography, PET, and by functional magnetic resonance imaging, fMRI. For all these techniques the inverse problem has a unique solution, which makes the spatial localization quite reliable. These methods allow studies of the brain or heart without opening the skull or chest, but the techniques are weakly invasive because the subject is exposed to X-rays, to small time-varying and large static magnetic fields, or to radioactive marker substances. The temporal resolution of PET and fMRI, which depend on metabolic changes in cells, is 500 ms at best. This is insufficient to determine the order in which the different sites in the brain or heart become activated.

Electroencephalography, EEG, i.e., the measurement of electric potentials on the scalp, the so-called "brain waves" in layman's language, is a widely applied clinical technique. EEG is closely related to MEG. In both methods, the recorded signals are generated by the same synchronized currents that flow in the neural network. The time resolution of these two techniques is better than a millisecond, which exceeds that of PET and fMRI by several orders of magnitude. With MEG and EEG it is thus possible to follow rapid changes of cortical activity that reflect signal processing in the brain. A further very important advantage of these two techniques is their complete non-invasiveness. The subject is not exposed to radioactivity or to high magnetic fields. One only measures what comes out of the brain, spontaneously or as a result of natural stimuli. Electrocardiography, ECG, and magnetocardiography, MCG, have similar advantages as EEG and MEG.

In heart studies, likewise, magnetocardiography has to compete with electrocardiography; both methods have again their advantages and drawbacks. In MCG body contacts are not needed while in ECG the skin-electrode impedance may often disturb the data. In addition, the SQUID sensor locations in MCG can be defined with millimeter precision in respect to the heart, while it is much more difficult to determine the precise electrode locations in ECG. The main drawback of MCG is the high price of an installation, more than a factor of ten over a comparable ECG unit.

During the first decades of the new millennium, much progress will be made in all subfields of neural and cardiac studies. Magnetoencephalography and magnetocardiography should play important roles in this development. How the brain processes information is a problem which probably will be solved with the help of mathematicians and information scientists, but a large amount of data by neuroscientists will be needed to guide the various theorists. Clinical uses of MEG will increase. It is also likely that MCG research will give hints for the best cure in many heart diseases.

The first MEG measurements using a SQUID were published in 1970, at about the same time as the first PET experiments were carried out. The development of MEG was slow because too much time was spent with 1-SQUID devices and with studies of the unsolvable inverse problem. Whole scalp covering PET instruments became available in 1975; their construction was carried out briskly by nuclear physicists, who were used to multimillion dollar research grants. MEG was in the hands of solid state scientists who were not familiar with big money and, at first, did not seriously think about whole head covering devices, even though IBM had in its coffers plenty of high quality SQUIDs in the early 1980's. Another difficulty later arose because many scientists believed, MEG pioneer David Cohen among them, that the much cheaper EEG technique was just as good.

Research using MEG, not just developing the method, is finally coming up in force, many years behind PET and even fMRI. Since MEG is eminently suited for certain tasks, like functional mapping of cortical areas, diagnostic uses of the method are now emerging. Cognitive neuroscientists, likewise, have understood that MEG is a unique technique for investigating temporal aspects of signal processing and of other complex phenomena occurring in the human brain. MEG and EEG offer complementary information, but the two techniques have not been employed much in conjunction so far. Similarly, fMRI and PET data should be combined with MEG and EEG.

The use of MEG will expand, if more applications emerge in medical diagnostics, follow-up, and elsewhere; if not, the possibilities of MEG will not be utilized adequately, which would be a pity, both for basic brain research and for modern patient care. However, the price of MEG instruments should become significantly lower first. There is still plenty of interdisciplinary work to be done by physicists and mathematicians, together with basic and cognitive neuroscientists and clinicians, to develop MEG still further.

MEG and EEG will remain, at least for the time being, the only fully non-invasive techniques for monitoring brain activity on a millisecond scale. The very rapid development of magnetic resonance imaging has provided an ideal medium for linking the functional MEG information to anatomical MRI data.

Multichannel instruments for MEG and MCG are the largest users of SQUIDs by far. Late in 1970's IBM had decided to develop a supercomputer based on Josephson junctions. However, in 1982 the project was abandoned and the Company let the LTL in Helsinki use some of their SQUIDs in non-commercial neuromagnetometers. Today many companies manufacture dc SQUIDs, but the volume of production is relatively small.

We thank Riitta Hari and Jukka Nenonen for valuable comments on the manuscript and Mark Mehtonen for skillful digital preparation of illustrations. We also like to thank Mikko Kiviranta for his contributions to Chapter 3.

Andrä W. and Nowak H., Editors (1998). Magnetism in Medicine. 511 pp. Berlin: Wiley-VCH. [Discusses all uses of magnetism in medical practice and research, including SQUIDs, MEG, MCG, MRI and fMRI]

Hämäläinen M., Hari R., Ilmoniemi R.J., Knuutila J. and Lounasmaa O.V. (1993). Magnetoencephalography — Theory, Instrumentation and Applications to Noninvasive Studies of the Working Human Brain. Reviews of Modern Physics 65, 413-497. [A comprehensive treatment of MEG]

Hämäläinen M. and Nenonen J. (1999). Magnetic Source Imaging. Encyclopedia of Electrical Engineering 12. Webster, J., Editor. 133-148 pp. New York: Wiley & Sons [A recent overview of source imaging techniques in MEG and MCG].

Hari R. and Salmelin R. (1997). Human Cortical Oscillations: A Neuromagnetic View through the Skull. Trends in Neuroscience 20, 44-49. [ Review of MEG measurements on rhythmic oscillations generated in the human brain]

Lounasmaa O.V. (1996). Medical Applications of SQUIDs in Neuro- and Cardiomagnetism. Physica Scripta T66, 70-79; Lounasmaa O.V., Hämäläinen M., Hari R. and Salmelin, R. (1996). Information Processing in the Human Brain: Magnetoencephalographic Approach. Proceedings of the National Academy of Sciences of the USA 93, 8809-8815; Hari R. and Lounasmaa O.V. (2000). Neuromagnetism: Tracking the Dynamics of the Brain. Physics World 13, 33-38. [The present paper uses liberally material from these publications]

Nenonen J., Montonen, J. and Mäkijärvi M. (2000). Principles of Magnetocardiographic Mapping. Cardiac Mapping, Shenasa M., Borgreffe M. and Breithardt G., Editors. Mount Kisco, NY: Futura Publishing Co. [Short review of MCG mapping to be published in 2000]

Nicholls J.G., Martin A.R. and Wallace B.G. (1992). From Neuron to Brain. 805 pp. Sunderland: Sinauer Associates. [An excellent text on the human brain]

Ryhänen T., Seppä H., Ilmoniemi R. and Knuutila. J. (1989). SQUID Magnetometers for Low Frequency Applications. Journal of Low Temperature Physics 76, 287-386. [Comprehensive review of SQUIDs describing their fabrication, electronics and operation, as well as the use of SQUIDs in several applications]

Williamson S.J., Romani G.-L., Kaufman L. and Modena I., Editors (1983). Biomagnetism — An Interdisciplinary Approach. 706 pp. New York: Plenum Press. [This book treats in detail all basic aspects of biomagnetism but is now partly outdated]

Yoshimoto, T., Kotani, M., Kuriki, S., Karibe, H. and Nakasato, N. (1999). Recent Advances in Biomagnetism — Proceedings of the 11th International Conference in Biomagnetism. 1161 pp. Sendai: Tohoku University Press. [Short papers on recent advances in MEG and MCG]

Figure 1. The first whole-head MEG instrument, installed to the Low Temperature Laboratory at HUT. During actual measurements the double doors to the shielded room must be tightly closed. The concave, helmet shaped bottom of the helium dewar, with the flux transformer coils near its tip, is brought as close as possible to the subject's head. The 122 SQUIDs in this prototype instrument were manufactured by the IBM Company of Yorktown Heights, NY.

Figure 2. Illustration of the cerebral cortex from the left side.

Figure 3. Left: Typical magnetic response from the brain as a function of time; the N- and P-numbers refer to latencies in milliseconds after the stimulus was given. Right: Topographic map of the magnetic field gradient from which the active cortical area (center of the red arrow) can be determined. Gray indicates field out of the head and white into.

Figure 7. Block diagram of the direct readout scheme based on the noise cancellation method. The voltage noise of the amplifier is detected by the SQUID via an external feedback coil and a FET acting as a tunable resistor. The flux-to-voltage characteristics for different values of the FET gate voltage illustrate cancellation of the voltage noise in the amplifier. The noise of the preamplifier is artificially increased to emphasize the noise cancellation process. NC is the noise cancellation parameter.

Figure 9. Photograph of the 306-SQUID Vectorview™ instrument. Figure provided by Neuromag Ltd of Helsinki.

Figure 13. Information processing in the brain during picture naming. The locations of the primary visual (V), auditory (A), and somatosensory (S) cortices are marked by small squares. The cortical surfaces were generated from MRI scans of the subject. Magnetic signals recorded as a function of time from each site, during three different experimental paradigms, are shown as well. Figure modified from Salmelin et al., Nature 368, 463 (1994).

Figure 14. Face-specific detectors in the human brain. Columns illustrate source strengths in the fusiform cortex (white dots on both sides in the inserted MRI picture) at 165 ms after the stimulus (shown below each column) was presented. Description of stimuli: f = face, c = color, g = grayscale, b = black-and-white, s = scrambled, r = randomized, a = animal, wb = whole body, o = object. Figure modified by Tommi Raij from Halgren et al., Cerebral Cortex 10, 69 (2000).

Figure 15. Result of an experiment in which the subject was asked to look at another person pinching a manipulandum. Above: An example of the hand movements that the subject viewed. Below left: Areas of the brain's left hemisphere activated during viewing, according to MEG measurements. As expected, the visual cortex at the back of the head (violet dot) was responding first because the subject saw the moving hand. The next activation occurred in the left Broca's region (black dot), followed by the left motor cortex (red dot). Below right: The last area to be activated was the primary motor cortex (green dot) in the right hemisphere. Figure modified from Nishitani and Hari, Proc. Natl. Acad. Sci. USA 97, 913 (2000).

Figure 16. Generator areas of epileptic spikes recorded from a patient who had suffered partial epilepsy for many years. An analysis of the MEG spikes showed two epileptogenic regions in the right hemisphere of his brain. Activation of sources in the two regions revealed no systematic time differences. Figure modified by Nina Forss from Lamusuo et al., Epilepsia 40, 921 (1999).

Figure 17. MR image of the brain's left side of a patient with a large tumor (oligoastrocytoma) in the vicinity of the central sulcus. The somatosensory cortex was traced by MEG responses generated by stimulating the right tibial nerve at the ankle (red dot), the right median nerve at the wrist (purple dot) and the right lip (yellow dot). The primary motor cortex (green dot) was identified during extension of the right wrist. The violet dot indicates the site of the auditory cortex. Large veins on the brain surface were made visible by injecting a small amount of gadolinium into the patient's bloodstream during MR imaging. Figure modified by Nina Forss from Mäkelä et al., Human Brain Mapping (to be published in 2000).

Figure 19. Photograph of the 67-SQUID cardiomagnetometer in the BioMag Laboratory of the Helsinki University Central Hospital. The cylindrical dewar bottom has a diameter of 30 cm and a curvature of 79 cm. Figure provided by Neuromag Ltd of Helsinki.

Figure 20. Magnetic heart signal recorded by the 67-SQUID cardiomagnetoimeter of Figure 19 on a patient suffering from arrhythmia. The normal heartbeat is followed by a larger abnormal beat spontaneously initiated in the ventricles. Figure provided by Juha Montonen of the Laboratory of Biomedical Engineering at HUT.

Figure 21. An example of MCG localization of continuous atrial tachycardia in a 24-year old patient. Left: Spatial MCG distribution during the abnormal small-amplitude P-wave (marked by the vertical line). Right: Two independent MCG localizations of the P-wave, superimposed on a transaxial MR image of the heart. The results were in excellent agreement with invasive localization obtained during catheter ablation. Figure provided by Markku Mäkijärvi of the Division of Cardiology at HUCH; see reference by Nenonen et al. in Bibliography

Figure 22. Detection of abnormal repolarization by MCG mapping. (a) Time points displayed on the T-wave are at 340 ms, 400 ms, and 460 ms after the onset the main QRS-peak at left. (b) Normal magnetic field distribution during ventricular repolarization. (c) MCG recording on a symptomatic 14-year old girl with a long-QT syndrome. The magnetic field pattern during the T-wave is clearly abnormal and displays typical multipolar structure. Figure provided by Jukka Nenonen of the Laboratory of Biomedical Engineering at HUT.

Figure 23. Examples of spatial distributions of the QRS fragmentation score S in a healthy volunteer (a), in a patient with old myocardial infarction but without arrhythmias (b), and in a patient with coronary heart disease and vith VT/VF (c). The dots indicate the locations of the 49 SQUID sensors above the anterior chest of the subjec. Color code shows the fragmentation score in various locations. Figure modified from Müller et al., Phys. Med. Biol. 44, 105, 1999.

Total number of words: 13370

# 1

Hierarchial number of contribution: 6.8.5.2.

Lounasmaa

Olli Viktor

Low Temperature Laboratory, Helsinki University of Technology, Box 2200, 02015 HUT, Finland

Telephone: +358-9-4512952 (office); +358-9-481541 (home)

Telefax: +358-9-4512969

E-mail: olli.lounasmaa@hut.fi

Y

Author Agreement No. EOLSSAA 001891

# 2

Hierarchial number of contribution: 6.8.5.2.

Seppä

Heikki Juhani

Measurement Technology, VTT Automation, Box 1304, FIN-02044 VTT, Finland

Telephone: +358-9-4566419 (office); +358-9-555414 (home)

Telefax: +358-9-4567029

E-mail: heikki.seppa@vtt.fi

N

Author agreement must be made with Heikki Seppä