At the end of the first part, we stated that spatial coding is a certain spatial relation between the small volumes of the brain, and we asked whether one can find shared spatial coding in the brain areas for the given stimulus.

For example, if faces are presented to a subject in the fMRI scanner, one could expect a similar spatial relationship between the occipital cortex where visual information arrives and face-specific brain areas (e.g., the fusiform cortex, the orbitofrontal cortex).
To explore this issue, one needs to take a 3D pattern of the occipital cortex activity during face perception and compare it with other areas in the brain cortex.

To ensure that the observed activity is face-specific, one must compare face perception and a non-specific visual perception like images of totally degraded faces. This contrast image would be the mean difference between many presentations of faces and degraded faces.

However, there is a catch. If one takes a cube at the peak of the face-specific occipital activity, there is no sense in just moving it upwards-backward or in the anterior-posterior directions and comparing it with other parts of the cortex. The brain cortex is curved. It means that we should not only move the initial cube while comparing it with other areas of the cortex but also to rotate it. Testing various rotations can permit us to find the most appropriate orientation of the initial cube, which gives the highest correlation with activity at the given site.

The initial cube of activity is called a seed pattern. We applied the following steps of analysis using a seed pattern:

• The seed pattern (10x10x10 voxels) is centered on each voxel in the brain.
• Rotations around the X, Y, and Z axes with a step of 10 degrees, testing all possible combinations of rotations around the axes.
• The maximum correlation among all pattern rotations is selected and saved.

As a result of these manipulations, we obtained the 3D image, in which each voxel contained the maximum correlation among the rotations of the seed pattern centered at this voxel.

Such an image was obtained for each subject and statistical analysis of the group of subjects revealed statistically significant correlations (with correlation coefficients 0.5-0.8) in the contralateral occipital cortex and in face-specific regions:

Similarly, in the images of auditory word perception (contrasted with a non-specific auditory baseline), when we took the seed in the left auditory cortex, we found statistically significant correlations at the group level in the vicinity of this pattern and in the contralateral auditory cortex.

Given that we performed thousands of correlations per voxel with the rotated seed, we verified whether significant correlations could be obtained in a group of subjects just by chance. In the groups of subjects formed by randomized brain activity, no correlations were statistically significant. It means that there is a stimulus-specific overlap between the subjects in the areas correlated with the seed pattern. Thus, spatial coding is shared between the involved brain areas and is stimulus-specific.

Our method of rotational cross-correlations can be applied to functional brain images and structural ones. For example, we applied it to the PET image of amyloid plaques in a patient with Alzheimer’s disease. Placing the seed in the occipital cortex, one can see that at the correlation threshold of 0.5, similar spatial distributions of amyloid are observed in the temporal and paracingulate cortex. The diagnostic value of this approach presents a clinical perspective.

Of course, applications of the method to functional neuroimaging are also an interesting perspective for clinical research. For example, one can explore patterns of auditory activity in patients with asymmetric hearing loss. One can predict that the contralateral distribution of auditory activity in hearing loss (anacusis) is different from normal hearing listeners (NHL). Some hints for this can be found in the recent study of our group using classical neuroimaging analysis:

Sadoun A, Chauhan T, Mameri S, Zhang YF, Barone P, Deguine O, Strelnikov K. (2020) Stimulus-specific information is represented as local activity patterns across the brain. NeuroImage.

Cognitive neuroimaging is a branch of neuroscience, which studies brain activations in response to cognitive acts. There is a fundamental question in cognitive neuroimaging – what is brain activation? Why we see blobs of different intensities composed of points of different intensities?

I got interested in this question after my Ph.D. thesis and having found no satisfactory answer I published my own definition of brain activation. Brain activations can be defined like this:

Information-driven reorganization of energy flows in and among populations of neuroglial units, leading to a total increase of energy utilization in these populations.

Strelnikov K. (2010) Neuroimaging and neuroenergetics: brain activations as informationdriven reorganization of energy flows. Brain and Cognition.

This definition requires some clarifications:

Neuroglial units are collaborating neural and glial cells. Energy flows are transformations of energy that propagate along the cellular structures (axons, dendrites, synapses, etc.). In these flows, metabolic energy is transformed into the energy of electromagnetic fields. At rest, the organization of this system reflects predictive coding; the system is reorganized by stimulation.

Predictive coding can be well seen in our study of initially deaf cochlear implanted patients. At rest, without any stimulation, they had differences from controls in visual, auditory, temporal, and speech-related areas:

Strelnikov K, Rouger J, Demonet JF, Lagleyre S, Fraysse B, Barone, P. (2010) Does brain activity at rest reflect adaptive strategies? Evidence from speech processing after cochlear implantation. Cerebral Cortex

In all these areas, there was an increase in activity at rest six months after the implantation:

Thus, these areas at rest support their increased predictive coding for audiovisual language.

Schematically, the relationship between informational input and predictive coding can be clarified like this:

When there is a large deviation of the received information from predictions, we need much energy to treat it (A). When our predictions are the same as the received information, almost no supplementary energy is needed with respect to rest (C). Energy usage needs to be economized (optimized) in the brain; this is why we have specialized cellular populations in the brain that predict various incoming information types.

The proposed definition of brain activations speaks of the unity between the two cell types in the brain – neurons and glia. These neuroglial populations exist in 3D. Can we, for computational purposes, represent these populations as cubes? Of course, this is a silly idea. Neuroglial populations have different sizes and shapes. However, curiously, we do not have a choice in neuroimaging. Any neuroimaging technique represents the brain as a set of small cubes called voxels (like pixels in pictures):

In functional neuroimaging, we have the representation of the brain as a 3D activity field where each cube has a signal value, which reflects the average level of metabolic energy in this cube:

What happens when a signal arrives from one neuroglial population to another? Electrophysiological studies show a burst of activity at the site of the signal arrival corresponding to the steep slope of energy level at this place.

Kepecs A, Wang XJ, Lisman J (2002) Bursting neurons signal input slope. J. of Neuroscience
La Camera G, Giugliano M, Senn W, Fusi S (2008) The response of cortical neurons to in vivo-like input current: theory and experiment. Biological Cybernetics

For the curve in one dimension, the slope is mathematically defined as the derivative of the curve; in several dimensions, the slope is mathematically defined by the gradient vector, which points to the maximal increase of the signal.

If we are lucky to divide the brain into voxels well, we can see that in one voxel (cube), there is no burst, and in the adjacent voxel, there is a burst of activity leading to the high gradient between these voxels.

For example, in this scheme for a voxel with a signal level 3, the highest spatial difference will be in the voxel’s direction with signal level 8.

This illustration for the one slice of brain activity demonstrates that gradient vectors can form divergences (spreading out from a source) and convergences:

We asked whether it is possible by calculating gradient vectors in the 3D brain activity (corresponding to the highest differences between the adjacent voxels) to deduce from the statistical difference between stimulation and baseline the stable activity flows specifically for the given type of cognitive load. For example, one could expect sources in the occipital cortex and propagation of activity in the frontal cortex for face perception.

Having analyzed two groups of subjects, one with visual stimuli (faces) and the other with auditory stimulation (words), indeed, we found statistically significant sources in the occipital cortex for faces and in the temporal auditory areas for auditory words.

Strelnikov K, Barone P (2012) Stable Modality-specific activity flows as reflected by the neuroenergetic approach to the fMRI weighted maps. PLoS ONE

We also found significant directions of stable activity flows, for example, in the forward-backward direction for faces (Y-axis for the brain):

In another study, we found a high spatial overlap between the gradient directions in fMRI activity and the distributed spatial sources of EEG and MEG activities for the same stimuli suggesting that these gradients do not reflect only one neuroimaging technique.

Strelnikov, Barone (2014), Overlapping brain activity as reflected by the spatial differentiation of Functional Magnetic Resonance Imaging, Electroencephalography and Magnetoencephalography data. J. of neuroscience and neuroengineering.

Thus, our results indicate stimulus-specific spatial organization within activated regions in the brain. It reflects the spatial interaction of neuroglial populations and encodes the presented stimulus in a certain spatial coding within spatial patterns of voxels.

From the fundamental perspective, spatial coding may be placed at the intersection of the three physical properties: information, energy, and space.

One can ask if spatial coding is a certain spatial relation between the small volumes of the brain, can one find shared spatial coding in the brain areas for the given stimulus? For example, during face perception, one could expect similar relations between voxels in the occipital areas, basal temporal and frontal areas known to be active during face processing. This question was addressed by our recent study¹, which we will consider in the next post.

¹Sadoun A, Chauhan T, Mameri S, Zhang YF, Barone P, Deguine O, Strelnikov K. (2020) Stimulus-specific information is represented as local activity patterns across the brain. NeuroImage.

Much of our behavior is shaped by feedback from the environment. We repeat behaviors that previously led to rewards and avoid those with negative outcomes. At the same time, we can learn in many situations without such feedback. Our ability to perceive sensory stimuli, for example, improves with training even in the absence of external feedback.

Feedback signals derived from the participants’ confidence reports activated the same brain areas typically engaged for external feedback or reward. Moreover, just as these regions were previously found to signal the difference between actual and expected rewards, so did they signal the difference between actual confidence levels and those expected on the basis of previous confidence levels. This parallel suggests that confidence may take over the role of external feedback in cases where no such feedback is available. Finally, the extent to which an individual exhibited these signals predicted overall learning success.

http://dx.doi.org/10.7554/eLife.13388.002

The widely used approach to quantify our beliefs and their relation to observations is Bayesian statistics. For example, I can belive that the average height of men in China is 160 cm, most of the men being between 150 and 170 cm. Then I go to China and start measuring men there. I look at my results and see that they are distributed not exactly as in my prior belief. How should I update my belief? A clear non-mathematical introduction on Bayesian statistics explains the idea of the method:
https://towardsdatascience.com/a-zero-math-introduction-to-markov-chain-monte-carlo-methods-dcba889e0c50

Popular science is a well-established way of spreading the scientific culture and simplifying the meaning, results and consequences of scientific works. Often appearing to be a fruitful portal, popular science allows the general public to get an inkling of what is going on in the scientific world. In addition, it is also a good method of “translating” science for non-scientific experts working in the field of the humanities – for example, philosophers.  Indeed, writing popular science articles can be seen as equivalent to translating texts between languages of different ethnic groups. However, as with every translation effort, attention has to be paid to the fact that translation is an imperfect process. It gives only a general picture. Moreover, for a successful translation from one language to another, we need not only an expert’s understanding of both languages but also someone with expertise in translation itself.

Likewise, it is possible for non-scientific people or students of the humanities to find through popular science a way to build a vision of scientific knowledge. Unfortunately, on many occasions, I have happened to read stories that have been badly popularized, where the author misunderstood the scientific study and, in the translation, has misled the general public and also non-expert scholars. Moreover, a lack of global vision and of deep knowledge can inevitably lead to erroneous interpretations and the consequences could be disastrous.

To avoid the pseudo-knowledge of non-experts, scientific knowledge should be popularized by experts in the field who have at the same time mastered a good method of explaining it to the general public. Indeed, popular science is a powerful and useful tool for scientists to achieve these important goals (*):

• To translate complex scientific works for people far from the scientific field.
• To disentangle some misunderstood concepts or ideas used by some pseudoscientific books or magazines.
• To explain to the general public that some ideas are not really scientific ones and provide ways and tools in order for them to comprehend the unadulterated reality of science.

Thus, having a large population getting a clear picture of scientific knowledge will empower citizens able to make reasoned decisions on issues related to, for example, science and the medical field and will prevent society from reverting to the dark ages of history.

Amirouche Sadoun

*Laura Bonetta. 2007. Scientists Enter the Blogosphere. Cell 129, pp. 443-445.

Modern scientific breakthroughs have raised many philosophical questions covering several domains, such as mathematics, genetics, quantum physics, artificial intelligence, psychosurgery and cognitive neuroscience. For example, the complexity of the brain, as revealed by the new techniques of imaging and other technology, demonstrates a level of organization that even scientists are only starting to understand and opens an immensely rich field of thought that philosophers are struggling with. Contemporary philosophers, such as Gaston Bachelard, regretted that a lot of philosophers, like Emile Meyerson, did not take much serious consideration of the diversity of scientific knowledge, a necessary step to conceive any philosophy of science. More specifically, Meyerson did not foresee that changes in scientific paradigms would lead to changes in the conception of epistemology itself (*).

Indeed, the considerable development of science, the complexity and richness of its terminology and its myriad concepts in our modern period make it more and more difficult to those who do not have (at least not yet) a scientific background to grasp what happens in several interconnected fields of science. How could a student of philosophy provide the analysis of scientific knowledge (epistemological analysis) without being able to understand at least one of the scientific areas, without being primarily a scientist? How is it possible in this condition to be able to construct and develop a critical analysis of the scientific method, its inferences and logical forms, if one does not have scientific training? It is certainly almost impossible to approach this goal with just a rudimentary scientific knowledge. Would that mean that modern science is destroying the philosophers’ field of thinking and reasoning? Certainly not. In my opinion, the access to the philosophy of science is always possible if one has a minimum of scientific background and an understanding of the basics of scientific methodology and recent advances. Students of philosophy already have a highly charged curriculum, so a good alternative would be for the scientists themselves to try to approach philosophy in search of a deeper or more general understanding of scientific problems and controversies.

Amirouche Sadoun

* Frédéric Fruteau de Laclos. 2008. « Le bergsonisme, point aveugle de la critique bachelardienne du continuisme d’Émile Meyerson ». In Bachelard et Bergson (2008), pp. 109-122. Ed. Presses Universitaires de France.

We think that either space and time perception or, for example, face perception are inevitably engrained in the physical laws of our body. These laws seem subjective when perceived from the interoceptive point of view. However, they reflect the same physical laws as in the rest of the universe. Our internal state has the same objectivity as any other event in nature and it needs to be studied to discover physical laws behind it.

Even if we imagine a fantastic object, this imagination is just another phenomenon of nature resulting from objective physical laws in the brain. E.g. if we study water flows in the ocean, a certain flow may seem fantastic or even absurd but it is always based on a set of physical laws. It may seem fantastic only because we do not know yet physical laws behind it. This concerns any observation in nature including our thoughts and internal representations.

Kuzma Strelnikov and Amirouche Sadoun