The document presents a detailed examination of neurite tracing using a marked point process (MPP) model, supervised by Prof. Ooi Wei Tsang and Prof. Daniel Racoceanu. It discusses the challenges and results of automated digital reconstruction and quantification of neuronal morphology, highlighting the unsupervised extraction of 3D neuronal networks while addressing issues like tile stitching artifacts and false connections. Key methodologies include energy optimization techniques and the integration of geometric constraints for improved object detection in neuronal imagery.

1. Marked Point Process for Neurite
Tracing
Sreetama Basu
NUS-IPAL
Supervised by
Prof. OOI Wei Tsang
And
Prof. Daniel Racoceanu

2. Presentation topics
• Neurite tracing
• Marked Point Process(MPP)
• MPP model for neurite tracing
• Results

3. Digital reconstruction
Segmentation followed by quantification
•Structure determines functions;
linked to higher order cognitive functions,
neuro-degenerative diseases
(Alzheimer’s, Parkinson’s etc).
•Brain Atlas of common lab animals;
•Huge volume of (sparse) data, Manual reconstruction takes days and months
• C.elegans ~ 15 years; EM connectomic data ~800 terabytes
•Digital reconstruction : Easier to archive, exchange, analyze
Figure adapted from : Ascoli, G. A., J. L. Krichmar, et al. (2001). "Generation, description and storage of dendritic morphology
data." Philos Trans R Soc Lond B Biol Sci 356(1412): 1131-1145.

4. Challenges of automated reconstruction
Tile stitching artifacts Fuzzy structures
False connections, Halos around branches
overlaps, crossing mistaken as parallel
processes
Intra operator
Branch gaps And inter operator
And Variability in manual
discontinuities reconstruction
by experts
Brown, K. M., G. Barrionuevo, et al. (2011). "The DIADEM data sets: representative light microscopy images of neuronal
morphology to advance automation of digital reconstructions.“ Neuroinformatics 9(2-3): 143-157

5. Presentation topics
• Neurite tracing
• Marked Point Process
• MPP model for neurite tracing
• Results

6. Marked Point Process
Extraction of tree crowns
Extraction of arbitrarily-shaped objects using stochastic multiple birth-and-death dynamics and active contours, Maria S. Kulikova, Ian H. Jermyn,
Xavier Descombes, Elena Zhizhina, and JosianeZerubia, IS&T/SPIE Electronic Imaging 2010 meeting, San Jose, California, USA
Extraction of
road networks
A Point Process for Fully Automatic Road Network Detection in Satellite and Aerial Images Pierre, C. Descombes, X. Zhizhina, E. Problems of
Information Transmission 10 3 247-256 2010

7. Marked Point Process
Intermediate Image data
0th iteration
iteration MPP Objects
Energy optimization through
A B A B Birth and death dynamics
C C Sampling embedded in
D E E
D Simulated Annealing
F F
A B A B
C C
D E D E
F F
Iteration stops when all objects Intermediate
are detected, corresponds to minimum energy iteration

8. Marked Point Process
• Set of objects= realization of a MPP
K: point process whose realization is a
random number of points given by a
random variable
• Marked point process objects defined in S= K x M
M: mark space
of the object
• Example: M= [Rmin,Rmax]
• And (x,y, R) defines a disc object
A K
B
centre radius C
D E
F

9. Marked point process
• image is viewed as a Gibbs energy model, where the maximum
object density corresponds to minimum energy
1
P ( X x=
= ) exp( − β E ( x))
K
Energy
Inverse temperature
• Optimization: Multiple Birth and Death dynamics embedded in
Simulated Annealing
Advantages:
Unsupervised extraction not involving any specific configuration
initialization (unknown number of objects)
Allows incorporation of prior – geometric and interaction constraints
among objects
Takes into account data at macroscopic scale , at object level instead of
pixels (HR images)

10. Presentation topics
• Neurite tracing
• Marked Point Process
• MPP model for neurite tracing
• Results

11. MPP model for neuronal network
extraction
• Aim: unsupervised 3D neuronal network extraction
• Model: object process (objects: sphere ~ U ([rmin , rmax ])
specified by an Energy:
U = Ud + Ui + Up
Prior term
Data term
•Favors connectivity
• internal homogeneity
•Contrast with background Interaction term
•Prevents overlap of objects
•Prevents crowding together of objects
• Optimization: Multiple Birth and Death dynamics
embedded in Simulated annealing

12. MPP model for neuronal network
extraction
Red circle =
cut of
spherical
object on a
Data Energy response slice of the
image stack
Internal homogeneity Both constraints
No contrast are satisfied
constraint is violated
with background

13. MPP model for neuronal network
extraction
Pair-wise interaction constraints
Considering radiometric properties + interaction constraints among objects

14. MPP model for neuronal network
extraction
Prior Energy: based on number of neighbors,
Models multi-object interaction; favors
connectivity and elongation of network

15. Presentation topics
• Neurite tracing
• Marked Point Process
• MPP model for neurite tracing
• Results

16. Data
Olfactory Projection data stack: input to MPP 2D projection (max intensity) of the
Olfactory Projection data stack

17. Preliminary Results
Olfactory Projection data stack with MPP objects Volume rendering of image stacks

18. Preliminary Results
Approximation of the neuronal network using Construction of Minimum Spanning Tree from
MPP with spheres as objects object centres on 2D projection (max intensity)
of the Olfactory Projection data stack

19. Preliminary Results
Gold Standard Manual
Reconstruction of MPP +MST reconstruction
the same Axon

20. Thank You