2021
DOI: 10.1038/s41467-021-26255-2
|View full text |Cite
|
Sign up to set email alerts
|

MesoNet allows automated scaling and segmentation of mouse mesoscale cortical maps using machine learning

Abstract: Understanding the basis of brain function requires knowledge of cortical operations over wide spatial scales and the quantitative analysis of brain activity in well-defined brain regions. Matching an anatomical atlas to brain functional data requires substantial labor and expertise. Here, we developed an automated machine learning-based registration and segmentation approach for quantitative analysis of mouse mesoscale cortical images. A deep learning model identifies nine cortical landmarks using only a singl… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
1
1
1

Citation Types

0
23
0

Year Published

2022
2022
2024
2024

Publication Types

Select...
5
3
2

Relationship

1
9

Authors

Journals

citations
Cited by 36 publications
(23 citation statements)
references
References 72 publications
0
23
0
Order By: Relevance
“…First, a flattened cortical map was generated from the reference atlas, and the map was manually trimmed to fit our imaging data by removing a fraction of the most posterior and lateral regions in the neocortex. Such an operation is commonly used in previous studies 22 , 69 , 74 . We then merged small anatomy regions in the map and got the final reference regional map with nine brain regions, including the primary motor cortex (MOs), the secondary motor cortex (Mop), the somatosensory cortex, barrel field (SSb), the somatosensory cortex, upper limb (SSu), the somatosensory cortex, lower limb (SSl), the primary visual cortex (VISp), the association visual cortex (VISs), the dorsal retrosplenial cortex (RSPd), and the lateral retrosplenial cortex (RSPl).…”
Section: Methodsmentioning
confidence: 99%
“…First, a flattened cortical map was generated from the reference atlas, and the map was manually trimmed to fit our imaging data by removing a fraction of the most posterior and lateral regions in the neocortex. Such an operation is commonly used in previous studies 22 , 69 , 74 . We then merged small anatomy regions in the map and got the final reference regional map with nine brain regions, including the primary motor cortex (MOs), the secondary motor cortex (Mop), the somatosensory cortex, barrel field (SSb), the somatosensory cortex, upper limb (SSu), the somatosensory cortex, lower limb (SSl), the primary visual cortex (VISp), the association visual cortex (VISs), the dorsal retrosplenial cortex (RSPd), and the lateral retrosplenial cortex (RSPl).…”
Section: Methodsmentioning
confidence: 99%
“…We recommend generating this mask using the reference frame. Although the automated generation of a brain mask from mesoscale imaging data has been described [48], we find user-drawn masks to be more robust to image artifacts. Inputs are summarized in Table 1 .…”
Section: Resultsmentioning
confidence: 98%
“…All GCaMP responses were movement and hemodynamic artifact corrected by subtracting changes in green reflectance signals from observed green epi-fluorescence (Vanni et al, 2017) and expressed as percentages relative to baseline responses (F-F 0 /F 0 )*100 where F 0 is the baseline from the start of the trial to water reward delivery. For region-based analysis, the brain-to-atlas approach in MesoNet (Xiao et al, 2021) was used to register cortical images to a common atlas using predicted cortical landmarks to determine regions of interest (ROIs). A 5 x 5 pixel region centered in each ROI was used for examination of peak amplitude and baseline standard deviation.…”
Section: Methodsmentioning
confidence: 99%