Contributions
Abstract: EP1675
Type: Poster Sessions
Abstract Category: Therapy - Tools for detecting therapeutic response
Introduction: Reducing myelin injury and enhancing myelin repair are central therapeutic objectives in reversing the progression of multiple sclerosis. Despite the unique identifiable structure of myelin, elaborated by oligodendrocytes (OL), quantifying the injury and repair responses of OLs on a large-scale still poses a significant challenge to researchers. The lack of high-throughput systems capable of extracting cell-specific morphological information has impeded the screening and development of therapeutics aimed at promoting myelin protection and repair.
Objective: To establish automated quantitative approaches that eliminate the time for image acquisition and analysis while successfully matching human analytic quality in assessing OL ensheathment.
Methods: We developed our analytic systems using post-natal rat oligodendrocyte progenitors cultured in nanofiber multi-well plates. The sensitivity of the system to detect biologically relevant effects was validated by treating the cells with brain-derived neurotrophic factor and platelet-derived growth factor to enhance and reduce the number of ensheathing OLs respectively.
Results: We first generated a classic algorithmic approach that used the basic morphological characteristics of OL ensheathments to assess the myelinating potential of OLs. Next, we improved on the classic algorithmic approach by combining automated imaging with deep learning techniques. Our deep learning approach employed a convolutional neural network that followed a “UNet” architecture to learn cell-specific information. During training, the UNet was presented with images appended to single nuclei masks so that the network could learn to associate ensheathments to specific cell nuclei rather than identify ensheathments globally. Through validation on cell-specific and whole-image levels, we demonstrate that both approaches match the accuracy of human segmentations on several parameters, including length distributions and number of sheaths formed per cell. The combination of automated imaging and analysis offers a 5- and 20-fold increase in whole-well analytic speed for the UNet and classic algorithmic approaches respectively
Conclusion: By enhancing analytic speed without sacrificing analytic quality we have developed a high-throughput system capable of quantifying single-cell ensheathments. This new technology permits the detection of nuanced differences associated with myelin injury and repair.
Disclosure: Yu Kang Xu: Nothing to disclose
Daryan Chitsaz: Nothing to disclose
Qiao Ling Cui: Nothing to disclose
Robert A Brown: Nothing to disclose
Matthew Anacleto Dabarno: Nothing to disclose
Jack P Antel: Nothing to disclose
Timothy E Kennedy: Nothing to disclose
Abstract: EP1675
Type: Poster Sessions
Abstract Category: Therapy - Tools for detecting therapeutic response
Introduction: Reducing myelin injury and enhancing myelin repair are central therapeutic objectives in reversing the progression of multiple sclerosis. Despite the unique identifiable structure of myelin, elaborated by oligodendrocytes (OL), quantifying the injury and repair responses of OLs on a large-scale still poses a significant challenge to researchers. The lack of high-throughput systems capable of extracting cell-specific morphological information has impeded the screening and development of therapeutics aimed at promoting myelin protection and repair.
Objective: To establish automated quantitative approaches that eliminate the time for image acquisition and analysis while successfully matching human analytic quality in assessing OL ensheathment.
Methods: We developed our analytic systems using post-natal rat oligodendrocyte progenitors cultured in nanofiber multi-well plates. The sensitivity of the system to detect biologically relevant effects was validated by treating the cells with brain-derived neurotrophic factor and platelet-derived growth factor to enhance and reduce the number of ensheathing OLs respectively.
Results: We first generated a classic algorithmic approach that used the basic morphological characteristics of OL ensheathments to assess the myelinating potential of OLs. Next, we improved on the classic algorithmic approach by combining automated imaging with deep learning techniques. Our deep learning approach employed a convolutional neural network that followed a “UNet” architecture to learn cell-specific information. During training, the UNet was presented with images appended to single nuclei masks so that the network could learn to associate ensheathments to specific cell nuclei rather than identify ensheathments globally. Through validation on cell-specific and whole-image levels, we demonstrate that both approaches match the accuracy of human segmentations on several parameters, including length distributions and number of sheaths formed per cell. The combination of automated imaging and analysis offers a 5- and 20-fold increase in whole-well analytic speed for the UNet and classic algorithmic approaches respectively
Conclusion: By enhancing analytic speed without sacrificing analytic quality we have developed a high-throughput system capable of quantifying single-cell ensheathments. This new technology permits the detection of nuanced differences associated with myelin injury and repair.
Disclosure: Yu Kang Xu: Nothing to disclose
Daryan Chitsaz: Nothing to disclose
Qiao Ling Cui: Nothing to disclose
Robert A Brown: Nothing to disclose
Matthew Anacleto Dabarno: Nothing to disclose
Jack P Antel: Nothing to disclose
Timothy E Kennedy: Nothing to disclose