Kensuke Sekihara and Srikatan S. Nagarajan Series Editor: Joachim H. Nagel Series in Biomedical Engineering Editor-in-Chief Prof. Dr. Joachim H. Nagel Institute of Biomedical Engineering University of...

REVIEW OF LITERATURE
The neural mechanisms associated with human behavior and the minute details of the human
ain can be studied by functional
ain imaging. Its emergence has helped to disclose the neural co
elates of various behaviors. For instance: understanding the language, learning the new skills and remembering the essential facts. Furthermore, it’s important for patients with
ain tumors and epileptic patients for the mapping of normal and abnormal
ain functions.1 (Sekihara K.,& Nagarajan S)
Literature studies2-4 revealed that the functional
ain imaging was conducted using the positron emission tomography (PET), a nuclear-medicine-based imaging method which helps to determine the
ain’s metabolic activity. Over the last few decades, due to rapid advances in the field of science and technology, functional magnetic resonance imaging (fMRI) came into existence by replacing the PET. It was proven to be beneficial for understanding the studies based on mapping of the human
ain. (Bertero, M., & Piana, M. (2006), Jatoi MA, Kamel N, Malik AS, Faye I (2014), Horacek J et al,2007).
There has been a growing body of evidence that the usage of fMRI and PET is limited to only to the metabolic or the neurovascular activities. However, there weren’t of any use in measuring the neuronal activities. Based on a few studies and the above-mentioned factors, it can be stated that real-time imaging becomes hard as it limits the time resolution.1
For the measurement of neuronal activities within the time frame (millisecond scale), magnetoencephalography (MEG) and electroencephalography (EEG) were introduced. The electric potentials of the scalp and the weak magnetic fields present outside the scalp are generated by the electrophysiological activity.1,2,4
The minute details of the magnetic fields produced by the human
ain are measured by the MEG whereas, the EEG measures electric potentials present on the scalp because of its
ain activity. Over the past few years, the Electrocorticography (ECoG) came into the field of technology possessing some of the potential benefits. ECoG is a neuroimaging technique that records directly on the cortical surface and the electric potential is produced by the neural cu
ents. For instance: when surgery is indicated in pharmaco-resistant epileptic patients, the ECoG is performed. This is only confined to the human subjects during the clinical applications in order to reach the surface of the cortex.3 (Jatoi MA, Kamel N, Malik AS, Faye I ,2014). It utilizes the measurements of the voltage potential at various locations on the scalp (in the order of micro volts (μV)) and then applies signal processing techniques to estimate the cu
ent sources inside the
ain.
The mechanism of EcoG can be best explained by the following factors1:
ECoG signals:
These are recorded by stainless steel electrodes which are fixed in the silastic strips. It helps in limiting the infection as the strips do not cover the entire cortical surface.
Location:
These are placed right above areas which are mapped for the possible surgical intervention.
Design:
The strips are flexible to make sure that cortical regions do not get exposed by the craniotomy.
Advantages:
Provides signal to noise ratio much higher than the EEG.
Has higher temporal and the spatial resolution.
Limitations:
One of the limitations is the higher invasiveness and secondly is the limited coverage.
Due to the above-mentioned qualities, the ECoG signals are being successfully used in the clinical field. Studies have reported that ECoG is possible in nonhuman primates only if the possibility of a large portion of cortex exists (Yaman, F., Yakhno, V. G., & Potthast, R. , 2013)5. For instance: During the research experiments, when a subject undergoes ECoG, the cortical areas are not directly located under the electrodes.
This is due to the grid placement which is led by clinical requirements with no link to neuroscientific purposes. Similarly, in clinical studies, the ECoG grid is misplaced due to the wrong pre-surgical evaluation. Therefore, to study such mentioned areas it is essential to reconstruct
ain activity from the measured ECoG. This can only be possible by solving the electromagnetic inverse problem.
Inverse modeling in imaging
Human
ain’s neural activities generate coherent synaptic and intracellular signals or cu
ents in the cortical part of the
ain. These cu
ents are known to be the critical generators of electroencephalography and magnetoencephalography signals. The cu
ent generated can be noninvasively mapped using electromagnetic
ain imaging. The electromagnetic
ain imaging uses specific mathematical algorithms which constitute two important components which include inverse modeling and forward modeling. In forward modeling, the outputs from the sensors are derived from a known distribution source of neural generators.2 (Bertero, M.,& Piana, M., 2006). Forward modeling works on the principle of linear association between measurements and sources. It works along the sensor lead field, where the sensor-a
ay-sensitivity-profile is represented. Maxwell’s equations are applied in the forward modeling where the sensor-outputs are calculated.
Bertero & Piana (2006)2 stated that inverse modeling is a procedure or an algorithm which helps in reconstructing the source outputs depending on the field of sensor leads and measurements. Inverse modeling algorithms are useful in solving the bio electromagnetic inverse issues. These include the measurement of spatiotemporal distributions of a neural source through bio electromagnetic calculations received outside the human body.5
Types of inverse algorithms
Algorithms of inverse modeling can be categorized into two
oad varieties. They include imaging methods and parameter-estimation. Algorithms of imaging methods do not use or require any information about the sources of signal and thus can prevent the non-linear search in a parameter space which is high dimensional. The imaging methods can be further categorized into two main algorithm classes which include spatial filters and tomographic reconstruction (TR) methods.5 (Yaman, F., Yakhno, V. G., & Potthast, R. , 2013) Voxel discretization is involved in tomographic reconstruction methods which presume that each voxel has a fixed source. Tomographic reconstruction methods measure the source amplitudes at each voxel via the least-squares appropriate to the data measured. These methods usually need additional confinements on the distribution of source as voxel number is generally greater the number of sensors. Algorithms such as minimum norm methods are used in tomographic reconstruction. Also, a larger number of TR methods are reformulated as non-adaptive spatial filters. Non-adaptive spatial filters are those that rely on the measurement geometry.5(Yaman, F., Yakhno, V. G., & Potthast, R. , 2013)
Spatial filters constitute another category of image methods. The spatial filter is a linear operator fit to the data measured. This filter is used to measure the activity strength at a given spatial location/filter-point location. These spatial filters can numerically regulate the sensitivity of the sensor a
ay to make a virtual sensor. The sensitivity pattern of such virtual sensor is usually located around or at its pointing location. The spatial filters are often called beam formers in the signal processing field.  Spatial filters depend on both measurements of the covariance matrix and geometry.
Parameter estimation assumes that measurement of observed sensor data requires only a lesser number of sources. For example, in a case, it is assumed that the spatial source distribution had only minimal point sources. Here, the strengths, locations, and orientations from the point sources make a set of parameters which are unknown. These parameters are estimated using a non-linear least-squares appropriate to the measured data. Such point of sources work on a model called ECD (equivalent cu
ent dipole). If it is assumed to have a single source, the model is known as SDS (single-dipole search).5 (Yaman, F., Yakhno, V. G., & Potthast, R. 2013)
In parameter estimation, the source parameter...