Johannes Friedrich, Ph.D.
Speaker: Johannes Friedrich, Ph.D.
Associate Research Scientist, Neuroscience
Topic: Online methods for real-time analysis of calcium imaging data
Calcium imaging methods enable researchers to measure the activity of genetically-targeted large-scale neuronal populations. Whereas earlier methods required the specimen to be stable, e.g. anesthetized or head-fixed, new brain imaging techniques using microendoscopic lenses and miniaturized microscopes have enabled deep brain imaging in freely moving mice.
Previously, a constrained matrix factorization approach (CNMF) has been suggested to extract the activity of the imaged neuronal sources. It has been extended further to handle the very large background fluctuations in microendoscopic data (CNMF-E). However, both approaches rely on offline batch processing of the entire video data and are demanding both in terms of computing and memory requirements, in particular CNMF-E. Moreover, in some scenarios we want to perform experiments in real-time and closed-loop — analyzing data on-the-fly to guide the next experimental steps or to control feedback –, and this calls for new methods for accurate real-time processing.
Here we address both issues by introducing an online framework for the analysis of streaming calcium imaging data, including i) motion artifact correction, ii) neuronal source extraction, and iii) activity denoising and deconvolution. Extending previous work on online dictionary learning and calcium imaging data analysis, we first present online adaptations of the CNMF as well as the CNMF-E algorithm, which dramatically reduces memory and computation requirements. Secondly, we propose an algorithm that uses a convolution-based background model for microendoscopic data that enables even faster (real time) processing on GPU hardware.
We apply our algorithms on a variety of experimental datasets that employ 2-photon, lightsheet, and microendoscopic imaging techniques, and show that they yield similar high-quality results as the popular offline approaches, but outperform them with regard to computing time and memory requirements.
Our algorithms enable faster and scalable analysis, and open the door to new closed-loop experiments.