2026-09-19T01:18:43Zhttps://dspace.mit.edu/server/oai/requestoai:dspace.mit.edu:1721.1/1061222026-06-06T01:03:06Zcom_1721.1_7582com_1721.1_7581col_1721.1_131023
Nir Shavit.
Jakubiuk, Wiktor
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science.
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science
2016-12-22T16:29:53Z
2016-12-22T16:29:53Z
2015
2016
http://hdl.handle.net/1721.1/106122
965799815
Thesis: M. Eng. in Computer Science and Engineering, Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, February 2016.
"December 2015." Cataloged from PDF version of thesis.
Includes bibliographical references (pages 83-88).
By investigating neural connections, neuroscientists try to understand the brain and reconstruct its connectome. Automated connectome reconstruction from high resolution electron miscroscopy is a challenging problem, as all neurons and synapses in a volume have to be detected. A mm3 of a high-resolution brain tissue takes roughly a petabyte of space that the state-of-the-art pipelines are unable to process to date. A high-performance, fully automated image processing pipeline is proposed. Using a combination of image processing and machine learning algorithms (convolutional neural networks and random forests), the pipeline constructs a 3-dimensional connectome from 2-dimensional cross-sections of a mammal's brain. The proposed system achieves a low error rate (comparable with the state-of-the-art) and is capable of processing volumes of 100's of gigabytes in size. The main contributions of this thesis are multiple algorithmic techniques for 2- dimensional pixel classification of varying accuracy and speed trade-off, as well as a fast object segmentation algorithm. The majority of the system is parallelized for multi-core machines, and with minor additional modification is expected to work in a distributed setting.
by Wiktor Jakubiuk.
M.Eng. in Computer Science and Engineering
88 pages
eng
Massachusetts Institute of Technology
M.I.T. theses are protected by copyright. They may be viewed from this source for any purpose, but reproduction or distribution in any format is prohibited without written permission. See provided URL for inquiries about permission.
http://dspace.mit.edu/handle/1721.1/7582
Electrical Engineering and Computer Science.
High performance data processing pipeline for connectome segmentation
Thesis
application/pdf
false
Publication
unknown
unknown
<Publication xmlns="https://www.openaire.eu/cerif-profile/1.1/" id="4d4b423b-9f90-475f-ba00-42aef483d591">
<Type xmlns="https://www.openaire.eu/cerif-profile/vocab/COAR_Publication_Types">http://purl.org/coar/resource_type/c_1843</Type>
<Language>eng</Language>
<Title>High performance data processing pipeline for connectome segmentation</Title>
<PublishedIn>
<Publication>
</Publication>
</PublishedIn>
<PublicationDate>2016</PublicationDate>
<Authors>
<Author>
<DisplayName>Jakubiuk, Wiktor</DisplayName>
<Affiliation>
<OrgUnit>
</OrgUnit>
</Affiliation>
</Author>
</Authors>
<Editors>
</Editors>
<Publishers>
<Publisher>
<DisplayName>Massachusetts Institute of Technology</DisplayName>
<OrgUnit />
</Publisher>
</Publishers>
<License>http://dspace.mit.edu/handle/1721.1/7582</License>
<Keyword>Electrical Engineering and Computer Science.</Keyword>
<Abstract>By investigating neural connections, neuroscientists try to understand the brain and reconstruct its connectome. Automated connectome reconstruction from high resolution electron miscroscopy is a challenging problem, as all neurons and synapses in a volume have to be detected. A mm3 of a high-resolution brain tissue takes roughly a petabyte of space that the state-of-the-art pipelines are unable to process to date. A high-performance, fully automated image processing pipeline is proposed. Using a combination of image processing and machine learning algorithms (convolutional neural networks and random forests), the pipeline constructs a 3-dimensional connectome from 2-dimensional cross-sections of a mammal's brain. The proposed system achieves a low error rate (comparable with the state-of-the-art) and is capable of processing volumes of 100's of gigabytes in size. The main contributions of this thesis are multiple algorithmic techniques for 2- dimensional pixel classification of varying accuracy and speed trade-off, as well as a fast object segmentation algorithm. The majority of the system is parallelized for multi-core machines, and with minor additional modification is expected to work in a distributed setting.</Abstract>
<Access xmlns="http://purl.org/coar/access_right"
>
</Access>
</Publication>