Recent advances in intact brain imaging, such as CLARITY and MAP (Magnified Analysis of the Proteome), enable volumetric imaging of brain tissue at subcellular resolution. Currently there is no effective tool for automated axon tracing at single fiber resolution in densely populated imagery. To address this need, we developed a machine learning-based high performance computing pipeline: 1) a convolutional neural network detects axon fiber voxels, 2) morphological operations extract fiber centerlines, 3) tracking logic connects fiber segments across low-intensity gaps and unresolved fiber crossings. The pipeline was implemented on a CPU cluster and tested on a 250 gigabyte volume of SMI-312 densely labeled axon fibers, imaged from parts of the hippocampus and cortex of a MAP-processed mouse brain. The pipeline automatically traced 221,298 fibers across gray and white matter in 10 hours. Of the traced fibers, 104 exceeded 1 mm, with the longest being 2.16 mm. An accuracy of 84% was reported based on manual evaluation of the 200 longest fibers. While there is room to improve accuracy, this pipeline offers a significant speed-up and increased efficiency over tracing one neuron at a time or manual fiber tracing. As the pipeline is scaled up to evaluate larger regions of the brain, connectivity patterns derived from these automatically traced long-range axons can potentially provide insights into brain function.
|Title of host publication||2018 IEEE 15th International Symposium on Biomedical Imaging, ISBI 2018|
|Publisher||IEEE Computer Society|
|Number of pages||5|
|Publication status||Published - 2018 May 23|
|Event||15th IEEE International Symposium on Biomedical Imaging, ISBI 2018 - Washington, United States|
Duration: 2018 Apr 4 → 2018 Apr 7
|Name||Proceedings - International Symposium on Biomedical Imaging|
|Other||15th IEEE International Symposium on Biomedical Imaging, ISBI 2018|
|Period||18/4/4 → 18/4/7|
Bibliographical noteFunding Information:
DISTRIBUTION STATEMENT A. Approved for public release: distribution unlimited. This material is based upon work supported by the Assistant Secretary of Defense for Research and Engineering under Air Force Contract No. FA8721-05-C-0002 and/or FA8702-15-D-0001. Any opinions, findings, conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the Assistant Secretary of Defense for Research and Engineering.
All Science Journal Classification (ASJC) codes
- Biomedical Engineering
- Radiology Nuclear Medicine and imaging