Using SILMAS to improve machine learning-assisted quantification of pathology
(2025) In Optics Letters 50(10). p.3465-3468- Abstract
In this study, we investigate the benefits of using structured illumination light-sheet microscopy with axial sweeping (SILMAS) in the context of neural network-assisted quantification of pathology in volumetric data. A bottleneck for training a neural network is the availability of manually labeled training data that adequately covers the variance in unseen samples, allowing the network to correctly extrapolate. Volumetric data obtained using SILMAS imaging have enhanced uniformity of depth resolution and contrast, which could lead to faster convergence during training and require less training data. Here, we compare the performance of a neural network trained on SILMAS data with that of one trained on data from non-structured axially... (More)
In this study, we investigate the benefits of using structured illumination light-sheet microscopy with axial sweeping (SILMAS) in the context of neural network-assisted quantification of pathology in volumetric data. A bottleneck for training a neural network is the availability of manually labeled training data that adequately covers the variance in unseen samples, allowing the network to correctly extrapolate. Volumetric data obtained using SILMAS imaging have enhanced uniformity of depth resolution and contrast, which could lead to faster convergence during training and require less training data. Here, we compare the performance of a neural network trained on SILMAS data with that of one trained on data from non-structured axially swept light-sheet microscopy. The networks are trained on data from a cleared transgenic mouse brain tissue with the objective of quantifying pathological aggregations of alpha-synuclein (aSyn), which plays an integral role in the development of Parkinson’s disease. Overall, our findings demonstrate that the optical filtering enabled by SILMAS reduces the amount of manual labeling required when training neural networks for quantification.
(Less)
- author
- Andersson, David
LU
; Frantz, David
LU
; Kirik, Deniz
LU
; Berrocal, Edouard LU and Kristensson, Elias LU
- organization
- publishing date
- 2025-05-15
- type
- Contribution to journal
- publication status
- published
- subject
- in
- Optics Letters
- volume
- 50
- issue
- 10
- pages
- 4 pages
- publisher
- Optical Society of America
- external identifiers
-
- scopus:105005222119
- ISSN
- 0146-9592
- DOI
- 10.1364/OL.551597
- language
- English
- LU publication?
- yes
- additional info
- Publisher Copyright: Journal © 2025 Optica Publishing Group.
- id
- eb05b33e-9f62-4936-9ce2-5237c6855dbb
- date added to LUP
- 2025-08-05 13:47:42
- date last changed
- 2025-08-05 13:48:11
@article{eb05b33e-9f62-4936-9ce2-5237c6855dbb, abstract = {{<p>In this study, we investigate the benefits of using structured illumination light-sheet microscopy with axial sweeping (SILMAS) in the context of neural network-assisted quantification of pathology in volumetric data. A bottleneck for training a neural network is the availability of manually labeled training data that adequately covers the variance in unseen samples, allowing the network to correctly extrapolate. Volumetric data obtained using SILMAS imaging have enhanced uniformity of depth resolution and contrast, which could lead to faster convergence during training and require less training data. Here, we compare the performance of a neural network trained on SILMAS data with that of one trained on data from non-structured axially swept light-sheet microscopy. The networks are trained on data from a cleared transgenic mouse brain tissue with the objective of quantifying pathological aggregations of alpha-synuclein (aSyn), which plays an integral role in the development of Parkinson’s disease. Overall, our findings demonstrate that the optical filtering enabled by SILMAS reduces the amount of manual labeling required when training neural networks for quantification.</p>}}, author = {{Andersson, David and Frantz, David and Kirik, Deniz and Berrocal, Edouard and Kristensson, Elias}}, issn = {{0146-9592}}, language = {{eng}}, month = {{05}}, number = {{10}}, pages = {{3465--3468}}, publisher = {{Optical Society of America}}, series = {{Optics Letters}}, title = {{Using SILMAS to improve machine learning-assisted quantification of pathology}}, url = {{http://dx.doi.org/10.1364/OL.551597}}, doi = {{10.1364/OL.551597}}, volume = {{50}}, year = {{2025}}, }