Application of the Flocking Method for Spatial Analysis of Brain

Activity in Optogenetics Datasets

Margarita Zaleshina

and Alexander Zaleshin

Moscow Institute of Physics and Technology, Moscow, Russia

Institute of Higher Nervous Activity and Neurophysiology, Moscow, Russia

Keywords: Brain Imaging, Pattern Recognition, Optogenetics, Mouse Brain.

Abstract: This work introduces a new approach for spatial analysis of assumed dynamics of neuronal activity in mouse

brain images obtained by light-sheet fluorescence microscopy methods (LSM). In calculations we used

flocking algorithms based on neuronal activity distributions from slice to slice with a time delay that occurs

during scanning. We applied GDAL Tools and LF Tools in QGIS for topological processing of multi-page

TIFF files with LSM datasets. As a result, we identified localizations of sites with small movements of group

neuronal activity passing in the same locations (with retaining localization) from slice to slice. An important

advantage of this result is the ability to reveal locations with pronounced neuronal activity in a sequence of

several adjacent slices, as well as to identify set of sites with interslice activity.

1 INTRODUCTION

This paper presents a new approach in spatial analysis

of optogenetic data using a flocking method.

Optogenetics is a widely used method to study

neuronal activity in living organisms at the cellular

level. Genetically encoded indicators enable high

spatiotemporal resolution optical recording of

neuronal dynamics in behaving mice (Patriarchi et al.,

2018). These recordings further makes it possible to

collect and process brain images, revealing important

indicators of activities of sets of neurons in various

behavioural tasks, as well as in the study of

spontaneous activity.

Optogenetic data are usually presented as a multi-

page TIFF file consisting of a set of stitched 2D slices.

Specialized programs such as NeuroPG (Avants et

al., 2015), MicroMator (Fox et al., 2022) have

already been developed to view and analyse data.

Moreover, optogenetic image processing is also used

in other Platforms for Optogenetic Stimulation and

Feedback Control (Kumar and Khammash, 2022).

Optogenetics uses sets of images of registered

neuronal activity in the form of high-resolution 2D

slices. Typical number of slices in one multi-page

TIFF file is from 100 to 1000. Each pixel in these

https://orcid.org/0000-0001-5273-6579

https://orcid.org/0000-0001-9356-9615

images represents an activity of neurons at a specific

point, which is linked to a relative coordinate system.

Taking into account the fact that the time of 2D

recordings is nonzero, the change in activity from

slice to slice can also be used in computational

operations to determine the dynamics of activity.

When detecting neuronal activity, the main

problem faced by researchers is noises. A noise in

images, on the one hand, leads to the detection of a

‘false activity’ (false positives), but on the other hand,

makes it difficult to identify the existing activity

(false negatives), lowering the overall recognition

quality. In areas with redundant information the

influence of noise is higher. As a result, those zones

that are in the middle range of activity are of interest

for analysis, and allow to distinguish and highlight

weak effects.

The main goal of this work is to develop and

apply computational methods and tools that help

reduce the influence of noise on recognition of

neuronal activity and increase the predictability of

dynamic optogenetic (neuronal) activity through

brain slices. To eliminate the problem of noise in

optogenetic images, in this paper we propose a new

approach based on the principles of spatial analysis of

flock trajectories (flocking method).

Zaleshina, M. and Zaleshin, A.

Application of the Flocking Method for Spatial Analysis of Brain Activity in Optogenetics Datasets.

DOI: 10.5220/0012154100003595

In Proceedings of the 15th International Joint Conference on Computational Intelligence (IJCCI 2023), pages 471-478

ISBN: 978-989-758-674-3; ISSN: 2184-3236

471

The flocking method is based on keeping the

distances and co-direction of movements of elements

in the flock and can be used in analysis of dynamic

changes in brain activity. The principles of flocking

are already applied in brain imaging analysis

(Aranda, Rivera and Ramirez-Manzanares, 2014). In

our paper, their scope is expanded to analyse the

neuronal activity of fluorescently activated mouse

brain cells. The main scheme of our work is presented

in Figure 1.

Figure 1: Sources, elementary processed units and tools in

the processing of optogenetic images. A. Typical multi-

TIFF image. Schematically shows that multi-TIFF image

includes slices. Each subsequent slice is recorded with a

shift along the brain and with a time delay relative to the

previous one. B. Image analysis: B1. Primary image in

grayscale mode. B2. Identification of activity points (shown

as orange dots) by activity on a pair of neighboring slices.

B3. Identification of ensemble locations (shown as lilac

dots). B4. Identification of circuit tube between

neighboring slices (shown as blue tube), which connect of

circuit points (shown as green dots). С. Cross-slice

projection between slices 1_2 and N-1_N when identifying

circuit locations, taking into account the buffer zones

(shown with solid blue lines on the upper slices and dashed

lines on the lower slices).

Spatial relationships and neighbourhood in neural

networks in fluorescence microscopy datasets enrich

the possibilities of processing connectivity. The fact

that the activity of neurons is related not to a single

element but to a set of elements makes it possible to

process data by methods of spatial analysis for flocks,

taking into account the joint distribution of ‘neuronal

ensemble’ activity. The presence of a spatiotemporal

sweep between slices during scanning makes it

possible to take into account the direction of

movement of neuronal activity.

The usage of spatial analysis methods made it

possible to reveal data from pixels of multi-TIFF

images and conduct inter-slice spatiotemporal

analysis using flocking method.

2 BACKGROUND

2.1 Brain Imaging Methods and Tools

The purpose of brain imaging analysis is usually to

process data on brain structure, neuronal activity and

their interrelationships. The possibilities and ways of

analyzing the obtained data expand with the

development of medical and research equipment used

to obtain images of the brain. Thus, as image spatial

resolution and the accuracy of localization of

individual elements in the image increase, the ability

to identify the topology of structures and individual

brain areas improves. Also, with appropriate

resolution, the level of detail is improved for

describing processes in a healthy brain.

The possibility of separating activity of different

neuronal populations in the brain tissue was

investigated in experiments using various injections

to detect activity. Thus, in (Klapoetke et al., 2014) it

was shown that two channel rhodopsins can detect

two-color neural activation of spiking and

downstream synaptic transmission in independent

optical excitation of distinct neural populations.

Topology of structures is considered in certain

size ranges typical for these structures. Within small

areas of interest, an influence of topology from other

scales will reduce. Curved surfaces of the brain

directly affect the overall measurement of activity of

ensembles from different segments. Spatial analysis

in the recognition of images of brain tissue images

allows individual smaller elements on a curved

surface to become similar to linear elements. A

formation of dynamically stable ensembles with a

self-sustaining configuration can remain in its

localization for a prolonged time.

Segmentation, Connectivity

The main task for understanding the functioning of

both healthy and damaged brains is segmentation,

selection of areas of interest, and identifying

NCTA 2023 - 15th International Conference on Neural Computation Theory and Applications

472

connectivity between individual parts of the brain.

Linking experimental results to spatial and temporal

reference points is necessary for comparative analysis

of multiple heterogeneous data sets of brain structure

and activity, obtained from different sources, with

different resolutions, and in different coordinate

systems. Evaluation of automatic labelling detection

is investigated by Papp et al. (Papp et al., 2016), who

propose a new workflow for spatial analysis of

labelling in microscopic sections.

Tractography

Modern methods of Brain Imaging analysis apply a

transition from selecting areas of interest to

tractography techniques that allow visualizing

pathways of the brain (white matter tracts) using

tractography algorithms. Comparison of tractography

algorithms for detecting abnormal structural brain

networks presented in (Zhan et al., 2015), influence

of pre-processing and comparison of tract selection

methods in DTI analysis presented in (Ressel et al.,

2018). Sets of images obtained with a fiber-bundle

micro probe immersed at different depths inside a

fixed brain tissue were processed in (Doronina-

Amitonova et al., 2012).

Spatial Analysis of High-Resolution Images

As measurements become more detailed, researchers

have an opportunity to monitor not only summary

results in the form of connection or tracts but also to

identify detailed elements at the cellular level. To do

this, analysis algorithms are enhanced. Thomas L.

Athey et al. (Athey et al., 2023) presented BrainLine,

an open source pipeline that interacts with existing

software to provide registration, axon segmentation,

soma detection, visualization and analysis of results..

2.2 Artificial Neural Networks for

Spatial Processing

Artificial neural networks are frequently used in

segmentation of biomedical images. To solve the

problem of image processing in differentiated zoom

levels of images, mixed sized biomedical image

segmentation based on training U-Net (Benedetti,

Femminella and Reali, 2022) and DeepLabV3

(Furtado, 2021) are used. Time overlap strategy used

in U-Net (Ronneberger, 2017) allows for seamless

segmentation of images of arbitrary size, and the

missing input data is extrapolated by mirroring.

However, U-Net performance can be influenced by

many factors, including the size of training dataset,

the performance metrics used, the quality of the

images and, in particular, specifics of brain functional

areas to be segmented.

Despite the development of convolutional neural

networks (CNNs) is limited because both efforts

required to prepare training dataset, and time spent on

data recognition in trained neural networks are still

too great. As a result, it is more convenient to solve

this type of tasks using spatial analysis methods that

allow performing multi-operations and selecting not

all objects, but only those that are of interest for

further study. These approaches can be employed

either individually or in conjunction with CNNs

during pre-processing or post-processing stages of

data processing. In addition, a well-chosen

segmentation labelling algorithm (Lee et al., 2022)

helps to optimize work with neural networks.

2.3 Optogenetics in Studying of

Neuronal Activity

In 2005, Boyden and Deisseroth published the results

of the first optogenetics experiments. Their work

(Boyden et al., 2005) reported the ability to control

neuronal spiking with a millisecond resolution by

expressing a natural occurring membrane localized

light-gated ion pump. In further research, Deisseroth

explored possibilities of using optogenetics to control

brain cells without surgical intervention (Deisseroth,

2010). With the advancement of optogenetics, its

experimental applications have spread to all areas of

brain activity research. Optogenetic tools are enabling

causal assessment of the roles that different sets of

neurons play within neural circuits, and are

accordingly being used to reveal how different sets of

neurons contribute to emergent computational and

behavioral functions of the brain (Boyden, 2011).

Currently, optogenetics is actively used to study

the neuronal activity of living animals, allowing deep

immersion into the brain without destroying its

structure. Researchers conduct a variety of

optogenetic experiments on mice, including the study

of social and feeding behaviour (Jennings et al.,

2019), False Memory creation in certain parts of the

brain (Ramirez et al., 2013), and, if possible,

activating or suppressing the activity of brain cells

with a light flash, while affecting the general

behaviour of mice (Yang et al., 2021).

Light-sheet microscopy (LSM) was developed to

allow for fine optical sectioning of thick biological

samples without the need for physical sectioning or

clearing, which are both time consuming and

detrimental to imaging. The functioning principle of

LSM is to illuminate the sample while collecting the

fluorescent signal at an angle relative to the

illuminated plane. Optogenetic manipulation coupled

to light-sheet imaging is a powerful tool to monitor

Application of the Flocking Method for Spatial Analysis of Brain Activity in Optogenetics Datasets

473

living samples (Huisken and Stainier, 2009;

Maddalena et al., 2023).

2.4 Usage of Flocking Method in

Detection of Neuronal Activities

In this paper, we have extended the application of

flocking method to the spatial analysis of optogenetic

datasets. Flocking is a common behaviour observed

in nature, defining the collective behaviour of a large

number of interacting individuals with a common

aim. Nearby members of a flock should move in

approximately the same direction and at the same

speed. For studying of collective motion or

population dynamics in short trajectories is of-ten

applied the flocking method, which based on analysis

of joint directions and intersections of trajectories

with a time lag. Flock methods analyse a behaviour

of multi-sets of similar elements in research on

collective behaviour in biology and even in robotics

(Vicsek and Zafeiris, 2012; Ban et al., 2021;

Papadopoulou et al., 2023). The methods used to

calculate a behaviour of animals in a flock can also be

extended to model neural networks (Battersby, 2015).

Collective motion is also investigated for the analysis

of cumulative behaviour of cells (Ascione et al.,

2023).

The possibility of organizing parallel calculations

by using the processing of activity patterns with

spatial reference to individual tiles is shown by Marre

(Marre et al., 2012); as an extension of this work, the

paper (Goldin et al., 2022) shows the possibility of

using the CNN model to calculate context

dependence for predicting the activity of retinal cells

depending on the content of natural images.

In the case of neuronal activity, we are dealing

with a set of simultaneously working elements, where

behaviour of each of the elements depends on both its

neighbours and the environment (Degond, Frouvelle

and Merino-Aceituno, 2017; Levis, Pagonabarraga

and Liebchen, 2019; Escaff and Delpiano, 2020;

Rouzaire and Levis, 2022). Doursat suggested that

‘Neuron flocking’ must happen in phase space and

across a complex network topology’ (Doursat, 2013).

‘Flocking’ behaviour as presented in this work

has components comparable to the ‘delay activity’,

which was observed by Miyashita (Miyashita, 1988)

and theorized by other researchers (Hamid and Braun,

2019). Specifically, the current work postulates that

the activity of a neuron within a set of simultaneously

neurons (neuronal network) depends on its

neighbours within the neuronal network and, hence,

the topology of the network. The formation of mental

representations, being based on the temporal statistics

of the environment, involves the establishment of

stable neural patterns. These patterns of reverberating

activity act as attractors within the neural network,

enabling efficient encoding and retrieval of

information (Hamid and Braun, 2019).

In the paper (Aranda, Rivera and Ramirez-

Manzanares, 2014) Aranda et al. have shown that

algorithms, based on information about spatial

neighbourhood such as tractography methods, as well

as the flocking paradigm, can improve a calculation

of local tracks. Aranda et al. (Aranda, Rivera and

Ramirez-Manzanares, 2014) made an assumption for

calculations what ‘the flock members are particles

walking in white matter for estimating brain structure

and connectivity’. The authors applied calculation

methods in accordance with Reynolds' rules of

flocking behaviour (Reynolds, 1987). This

assumption makes it possible to calculate the

behaviour of individual sets of elements piece by

piece, without using of collective information.

The application of optogenetic scanning made it

possible to identify the main components of neuronal

activity and various types of activity changes in the

same locations over time.

We assume that the topological properties of

distribution of individuals in a moving flock are able

to represent information about the environment in the

same way as it is realized by a network of neurons. In

the methodology used in this paper, we further show

that considering a set of elementary components of

neuronal activity in the form of a flock improves the

extraction of meaningful information.

In calculations to study dynamics of neuronal

activity, we used the time delay that occurs when

moving from slice to slice during scanning using

optogenetic methods. An important advantage of

using this method is the identification of locations

where a pronounced directionality of neuronal

activity trajectories can be observed in a sequence of

several adjacent slides, as well as the identification of

areas of through intersection of activities.

3 METHODS

3.1 Calculations

The proposed flocking method for interslice image

analysis allows to identify activity of neural

ensembles in the mouse brain, which were obtained

using optogenetic technologies.

By applying the flocking principles to the analysis

of activity of a set of neurons, it becomes possible to

NCTA 2023 - 15th International Conference on Neural Computation Theory and Applications

474

reduce the influence of noise and replenish the sites

with missed activity.

The registered optogenetic highlighting that we

considered is caused by neural ensembles.

Illuminated elements are presented in the form of

pixels of varying degrees of brightness, with an area

of 1x1 sq. pixels (10x10 sq. μm). The characteristic

size of a single detected ensemble was up to 10x10

sq. pixels. These ranges are typical for cells,

ensembles, and agglomerations of cells (Bonsi et al.,

2019). All calculations were performed on the basis

of the characteristic features of neural circuits,

including common intersections and overlapping

buffer zones of different track.

Figure 2: Processing to save or remove intersection points

from layers.

The following operations were performed with

each of the multi-page TIFF files:

Split Multi-Page TIFF Files into Distinct Slices in

TIF Format

Image Pre-Processing and Interpolation

(a) Create contour lines of intensity (in the form of

isolines, the applied parameter is 25 pixels) for

each of the distinct slices.

(b) Apply LF Tools Extend lines plugin to a set of

contours in each of the distinct slices and

creating extended lines of contours at their start

and/or end points, 100 μm in length.

Flocks Identifying

from neighboring slices.

(d) Search for intersection points of extended lines

from neighboring slices that are located at a

distance in the range of 0.25-0.5 pixel (Figure

2).

(e) Remove all intersection points from the

previous item that are present in more than one

on the same extended line.

(f) Search for remaining intersection points from

three neighboring slices, which (points) are not

more than 0.25 pixel away from each other.

(g) Search for all intersection points from the

previous item that are more than one on the

same extended line.

This operation reveals either a long marginal

chain (more than 10 pixels in length) in several

neighboring slices or the movement of a large

object (10x10 pixels). The spread of

intensively of these identified objects occurs

over areas of distinct slices.

(h) Remove all intersection points from 3(d) item

that are more than one on the same extended

line.

(i) Search for remaining intersection points from

the previous item.

As a result, small movements (movement

within the identified localization, 10x10 pixels)

of small objects (3x3 pixels) are revealed.

Plotting of Flock Trajectories

(j) Splice of intersection points from 3(g) item

(defining the localization of small movement)

into a sequence corresponding to the sequence

of transitions from slice to slice, if the

intersection points from 3(g) item are not more

than 10 pixels away from each other.

Table 1: Spatial data processing applications.

Plu

in Descri

tion

Extracts contour lines

https://docs.qgis.org/3.2

8/en/docs/user_manual/

processing_algs/gdal/ras

terextraction.html

Generate a vector contour

from the input raster by

joining points with the same

parameters. Extracts contour

lines from any GDAL-

supported elevation raster.

Nearest neighbour

analysis

https://docs.qgis.org/3.1

6/en/docs/user_manual/

processing_algs/qgis/ve

ctoranalysis.html#qgisn

earestneighbouranalysis

Performs nearest neighbour

analysis for a point layer. The

output presents how data are

distributed (clustered,

randomly or distributed).

LF Tools

https://github.com/LEO

XINGU/lftools/wiki/LF-

Tools-for-QGIS

Tools for cartographic

production, surveying, digital

image processing and spatial

analysis (Extended lines)

The parameters used for the calculations were

established by selecting and optimizing the number of

intersection points connecting lines from two

different neighboring slices, taking into account the

Nearest neighbor analysis. An extended line is

Application of the Flocking Method for Spatial Analysis of Brain Activity in Optogenetics Datasets

475

constructed according to the distance between the

cells.

3.2 Applications for Spatial Analysis

In our work we processed optogenetic mouse brain

images using Open Source Geographic Information

System QGIS v.3.

Applications and special plug-ins (see Table 1)

were used for spatial analysis of data both within

single slices and between sets of closely spaced slices.

4 EXPERIMENTS AND RESULTS

4.1 Datasets

Our work considered optogenetic datasets on 23

mice. Datasets were presented as multi-page TIFF

files, which were exported to QGIS (http://qgis.org).

Recognition of multipoint activity and spatial

analysis of the distribution of neuronal activities

according to fluorescence microscopy datasets was

performed based on data packages published in an

open repository (https://ebrains.eu). As source

material, we used fluorescence microscopy datasets:

Set 1 (see Table 2): We used whole-brain datasets

(Silvestri et al., 2019) from transgenic animals with

different interneuron populations (PV, SST and VIP

positive cells) which are labelled with fluorescent

proteins. These datasets were obtained from 11 mice

(male animals, on post-natal day 56). The data was

represented in 48 multi-page TIFF files. Each multi-

page TIFF included 160 - 288 slices with dorsal or

ventral projections of the mouse brain. The data

resolution is 10.4x10.4x10 μm.

Set 2 (see Table 2): We used whole-brain datasets

(Silvestri, Di Giovanna and Mazzamuto, 2020)

obtained using LSM in combination with tissue

clearing. These datasets were obtained from 12 mice

(male animals, on post-natal day 56). The data was

represented in 14 multi-page TIFF files. Each multi-

page TIFF file included 800 slices with dorsal or

ventral projections of the mice brain. The data

resolution is 10x10x10 μm.

By processing using CLARITY-TDE method

(Chung et al., 2013; Costantini et al., 2015) images

have been partially cleaned up.

Allen Mouse Common Coordinate Framework

(Wang et al., 2020) served in our work as a frame of

data reference to spatial coordinates.

Table 2: Source material.

Set 1 Set 2

Number of mice 11 mice 12 mice

Gender and age of

animals

male

animals,

post-natal

day 56

male

animals,

post-natal

day 56

Parvalbumin-positive

interneurons parvalbumin

(

)

4 Animals 5 Animals

Somatostatin-positive

interneurons somatostatin

(

SST

)

3 Animals 3 Animals

VIP-positive

interneurons vasoactive-

intestinal peptide (VIP)

4 Animals 4 Animals

Number of multi-page

TIFF files (several files

er mouse)

48 multi-

page TIFF

files

14 multi-

page TIFF

files

Tissue clearing method CLARITY/

TDE

CLARITY/

TDE

Resolution 10.4x10.4x

10x10x10

Number of slices in one

multi-

age TIFF file

about 288

slices

800 slices

Size of one slice about

1200x1500

ixels

1140x1500

pixels

4.2 Results

Spatial Processing in QGIS for Identifying Tiles

With Activity Locations (see Section 3.1)

Output: set of tiled rasters with activity locations,

which contain the ID-numbers of individual slices

The values obtained as a result of our work after

spatial processing in QGIS:

─ mean number of localization sites in neighboring

slices, averaged over multi-page TIFF files, is 124;

─ the percentage of sites with identified ‘small

movement’ (movement within the identified

localization, 10x10 sq. pixels) to the total number

of identified localizations, averaged over all multi-

page TIFF files, is 73.4%.

As a result of spatial processing in QGIS,

localizations of sites of (10x10 sq. pixels) with

‘flocks’ were identified, based on the intersection

points of extended lines in these localizations

Figure 3

(Figure 3).

Further, only tiles with the identification of

activity locations will be processed. This is much less

than the full data from multi-page TIFF sets, and

therefore the volume of processed materials is

reduced by 10 or more times.

NCTA 2023 - 15th International Conference on Neural Computation Theory and Applications

476

Figure 3: Slice-by-slice activity near the identified

localization (marked by yellow dots). Scale bar: 30 µm.

The identified sites were further used as

segmentation labelling for training U-Net and

DeepLabV3Plus neural networks.

Post-processing of Tiles with Identified Activity

Locations Using Convolutional Neural Networks

for Image Segmentation

Output: set of activity coordinates inside of tiles

which contain the ID-numbers of individual slices

and tiles. The finished results can be uploaded to

JSON files.

After training used our segmentation labelling, U-

Net showed next results in Precision, Recall, and F1-

score (F1-score = 2 * Precision * Recall / (Precision

+ Recall): 81.4%, 76.2%, and 78.7%, respectively;

and DeepLabV3Plus showed next results in

Precision, Recall, and F1-score: 76.4%, 79.4%, and

77.9% respectively.

Further Using of Final Results

Final results can be used in external applications, both

for calculating the trajectories of activity movements,

and for constructing tractograms inside 3D multi-

page TIFF files.

5 CONCLUSION

In our work, we applied the flocking method to

analysis of spontaneous brain activity. We selected

different cell groups and determined areas occupied

by ensembles of cell groups in mouse brain. When

performing computational experiments, we analyzed

the interslice propagation of neuronal activity for sets

of mouse brain images.

In summary, the contributions of this work are as

follows:

─ We performed a spatial analysis of mouse brain

optogenetic images using the flocking method.

─ We have shown that using the flocking method, it

is possible to detect more accurately both areas and

tracks of neuronal activity, identifying the

connectivity of extended areas of activity

─ We were able to identify localizations of sites with

small movements of group activity (stably

localized flickering of activity with small

movements).

In the future, the flocking method can be used not

only in processing of optogenetic images but also in

the analysis of other tracks, including the analysis of

data obtained by diffusion-weighted magnetic

resonance imaging (DW MRI) + High Angular

Resolution Diffusion Imaging (HARDI).

REFERENCES

Aranda, R., Rivera, M. and Ramirez-Manzanares, A. (2014)

‘A flocking based method for brain tractography’,

Medical Image Analysis, 18(3), pp. 515–530. doi:

10.1016/j.media.2014.01.009.

Ascione, F. et al. (2023) ‘Collective rotational motion of

freely expanding T84 epithelial cell colonies’, Journal

of The Royal Society Interface, 20(199):20220719.

Athey, T. et al. (2023) ‘BrainLine: An Open Pipeline for

Connectivity Analysis of Heterogeneous Whole-Brain

Fluorescence Volumes’. doi: 10.1101/2023.02.28.

530429.

Avants, B. W. et al. (2015) ‘NeuroPG: open source

software for optical pattern generation and data

acquisition’, Frontiers in Neuroengineering, 8.

Ban, Z. et al. (2021) ‘Self-Organised Collision-Free

Flocking Mechanism in Heterogeneous Robot

Swarms’, Mobile Networks and Applications, 26(6), pp.

2461–2471. doi: 10.1007/s11036-021-01785-7.

Battersby, S. (2015) ‘News Feature: The cells that flock

together.’, Proceedings of the National Academy of

Sciences of the United States of America. United States,

112(26), pp. 7883–5.

Benedetti, P., Femminella, M. and Reali, G. (2022) ‘Mixed-

Sized Biomedical Image Segmentation Based on U-Net

Architectures’, Applied Sciences, 13(1), p. 329. doi:

10.3390/app13010329.

Bonsi, P. et al. (2019) ‘RGS9-2 rescues dopamine D2

receptor levels and signaling in DYT1 dystonia mouse

models.’, EMBO molecular medicine, 11(1).

Boyden, E. S. et al. (2005) ‘Millisecond-timescale,

genetically targeted optical control of neural activity.’,

Nature neuroscience, 8(9), pp. 1263–8.

Boyden, E. S. (2011) ‘A history of optogenetics: the

development of tools for controlling brain circuits with

light.’, F1000 biology reports. England, 3, p. 11.

Chung, K. et al. (2013) ‘Structural and molecular

interrogation of intact biological systems.’, Nature.

England, 497(7449), pp. 332–7.

Costantini, I. et al. (2015) ‘A versatile clearing agent for

multi-modal brain imaging.’, Scientific reports, 5(1), p.

9808. doi: 10.1038/srep09808.

Degond, P., Frouvelle, A. and Merino-Aceituno, S. (2017)

‘A new flocking model through body attitude

Application of the Flocking Method for Spatial Analysis of Brain Activity in Optogenetics Datasets

477

coordination’, Mathematical Models and Methods in

Applied Sciences, 27(06), pp. 1005–1049.

Deisseroth, K. (2010) ‘Controlling the Brain with Light’,

Scientific American. United States, 303(5), pp. 48–55.

Doronina-Amitonova, L. V et al. (2012) ‘Multicolor in vivo

brain imaging with a microscope-coupled fiber-bundle

microprobe’, Applied Physics Letters, 101(23), p.

233702. doi: 10.1063/1.4767386.

Doursat, R. (2013) ‘Bridging the mind-brain gap by

morphogenetic ‘neuron flocking’: The dynamic self-

organization of neural activity into mental shapes’,

AAAI Fall Symposium - Technical Report, FS-13-02,

pp. 16–21.

Escaff, D. and Delpiano, R. (2020) ‘Flocking transition

within the framework of Kuramoto paradigm for

synchronization: Clustering and the role of the range of

interaction’, Chaos: An Interdisciplinary Journal of

Nonlinear Science, 30(8), p. 083137.

Fox, Z. et al. (2022) ‘Enabling reactive microscopy with

MicroMator’, Nature Communications, 13, p. 2199.

Furtado, P. (2021) ‘Testing Segmentation Popular Loss and

Variations in Three Multiclass Medical Imaging

Problems’, Journal of Imaging, 7(2), p. 16.

Goldin, M. A. et al. (2022) ‘Context-dependent selectivity

to natural images in the retina’, Nature

Communications, 13(1), p. 5556.

Hamid, O. H. and Braun, J. (2019) ‘Reinforcement

Learning and Attractor Neural Network Models of

Associative Learning’, in Sabourin, C. et al. (eds)

Computational Intelligence: 9th International Joint

Conference, IJCCI 2017 Funchal-Madeira, Portugal,

November 1-3, 2017 Revised Selected Papers. Cham:

Springer International Publishing, pp. 327–349.

Huisken, J. and Stainier, D. Y. R. (2009) ‘Selective plane

illumination microscopy techniques in developmental

biology’, Development, 136(12), pp. 1963–1975.

Jennings, J. H. et al. (2019) ‘Interacting neural ensembles

in orbitofrontal cortex for social and feeding

behaviour’, Nature, 565(7741), pp. 645–649.

Klapoetke, N. C. et al. (2014) ‘Independent optical

excitation of distinct neural populations’, Nature

Methods, 11(3), pp. 338–346.

Kumar, S. and Khammash, M. (2022) ‘Platforms for

Optogenetic Stimulation and Feedback Control’,

Frontiers in Bioengineering and Biotechnology, 10, p.

918917. doi: 10.3389/fbioe.2022.918917.

Lee, Jonghyeok et al. (2022) ‘A pixel-level coarse-to-fine

image segmentation labelling algorithm’, Scientific

Reports, 12(1), p. 8672.

Levis, D., Pagonabarraga, I. and Liebchen, B. (2019)

‘Activity induced synchronization: Mutual flocking and

chiral self-sorting’, Physical Review Research.

American Physical Society, 1(2), p. 023026.

Maddalena, L. et al. (2023) ‘Optogenetics and Light-Sheet

Microscopy’, in, pp. 231–261.

Marre, O. et al. (2012) ‘Mapping a Complete Neural

Population in the Retina’, The Journal of Neuroscience.

United States, 32(43), pp. 14859–14873. doi:

10.1523/JNEUROSCI.0723-12.2012.

Miyashita, Y. (1988) ‘Neuronal correlate of visual

associative long-term memory in the primate temporal

cortex’, Nature, 335(6193), pp. 817–820.

Papadopoulou, M. et al. (2023) ‘Dynamics of collective

motion across time and species’, Philosophical

Transactions of the Royal Society B: Biological

Sciences, 378(1874). doi: 10.1098/rstb.2022.0068.

Papp, E. A. et al. (2016) ‘Brain-Wide Mapping of Axonal

Connections: Workflow for Automated Detection and

Spatial Analysis of Labeling in Microscopic Sections’,

Frontiers in Neuroinformatics, 10.

Patriarchi, T. et al. (2018) ‘Ultrafast neuronal imaging of

dopamine dynamics with designed genetically encoded

sensors.’, Science (New York, N.Y.). United States,

360(6396). doi: 10.1126/science.aat4422.

Ramirez, S. et al. (2013) ‘Creating a false memory in the

hippocampus.’, Science (New York, N.Y.), 341(6144),

pp. 387–91. doi: 10.1126/science.1239073.

Ressel, V. et al. (2018) ‘Comparison of DTI analysis

methods for clinical research: influence of pre-

processing and tract selection methods’, European

Radiology Experimental, 2(1), p. 33.

Reynolds, C. W. (1987) ‘Flocks, herds and schools: A

distributed behavioral model’, ACM SIGGRAPH

Computer Graphics, 21(4), pp. 25–34.

Ronneberger, O. (2017) ‘Invited Talk: U-Net

Convolutional Networks for Biomedical Image

Segmentation’, in Maier-Hein geb. Fritzsche, K. H. et

al. (eds) Bildverarbeitung für die Medizin 2017.

Springer Berlin Heidelberg, pp. 3–3.

Rouzaire, Y. and Levis, D. (2022) ‘Dynamics of topological

defects in the noisy Kuramoto model in two

dimensions’, Frontiers in Physics, 10, p. 976515.

Silvestri, L. et al. (2019) ‘Whole brain images of selected

neuronal types’. Human Brain Project

Neuroinformatics Platform. doi: 10.25493/68S1-9R1.

Silvestri, L., Di Giovanna, A. P. and Mazzamuto, G. (2020)

‘Whole-brain images of different neuronal markers’.

Human Brain Project Neuroinformatics Platform. doi:

10.25493/A0XN-XC1.

Vicsek, T. and Zafeiris, A. (2012) ‘Collective motion’,

Physics Reports, 517(3–4), pp. 71–140.

Wang, Q. et al. (2020) ‘The Allen Mouse Brain Common

Coordinate Framework: A 3D Reference Atlas’, Cell,

181(4), pp. 936-953.e20.

Yang, Y. et al.

(2021) ‘Wireless multilateral devices for

optogenetic studies of individual and social behaviors’,

Nature Neuroscience, 24(7), pp. 1035–1045. doi:

10.1038/s41593-021-00849-x.

Zhan, L. et al. (2015) ‘Comparison of nine tractography

algorithms for detecting abnormal structural brain

networks in Alzheimer’s disease.’, Frontiers in aging

neuroscience, 7(APR), p. 48.

NCTA 2023 - 15th International Conference on Neural Computation Theory and Applications

478