realSEUDO for real-time calcium imaging analysis
Authors: Iuliia Dmitrieva, Sergey Babkin, Adam S. Charles
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate comparable performance to offline algorithms (e.g., CNMF), and improved performance over the current on-line approach (On ACID) at speeds of 120 Hz on average. We benchmarked our speedups against the original SEUDO on 45000 frames across 50 cells from [12]. Simulated data experiments: We first applied all three algorithms to a simulated video created with Neural Anatomy and Optical Microscopy simulation (NAOMi) [30]. Applications to in-vivo mouse CA1 recordings We applied real SEUDO to an in vivo calcium imaging recordings from mouse hippocampal area CA1 previously described in Gauthier et al. 2022 [12]. |
| Researcher Affiliation | Collaboration | Iuliia Dmitrieva Applied Math and Statistics Johns Hopkins University Baltimore, MD 21218, USA Sergey Babkin Microsoft Research Redmond, WA, USA Adam S. Charles Biomedical Data Science Center for Imaging Science Mathematical Institute for Data Science Kavli Neuroscience Discovery Institute Johns Hopkins University Baltimore, MD 21218, USA |
| Pseudocode | Yes | Algorithm 1 real SEUDO Algorithm. Algorithm 2 Modified FISTA algorithm. |
| Open Source Code | No | While the data is all freely available, the implementation still requires some fine-tuning to be easy to use for the general public. We plan to release before the conference itself, and, provide full algorithmic and parameter selection details in the paper. |
| Open Datasets | Yes | Simulated data experiments: We first applied all three algorithms to a simulated video created with Neural Anatomy and Optical Microscopy simulation (NAOMi) [30]. Applications to in-vivo mouse CA1 recordings We applied real SEUDO to an in vivo calcium imaging recordings from mouse hippocampal area CA1 previously described in Gauthier et al. 2022 [12]. Additional in-vivo tests: As final test we applied all three algorithms (real SEUDO, On ACID, CNMF) to a 2000-frame mesoscope video example collected by the Yuste lab at Columbia University and provided with the On ACID github package as a demo. |
| Dataset Splits | No | The paper uses simulated and in-vivo datasets for evaluation. While a 'training set' is mentioned in the supplementary material for a separate neural network optimization experiment, the main experiments on calcium imaging data do not specify dataset splits (e.g., train/validation/test) for model training or validation. |
| Hardware Specification | Yes | Algorithmic performance was measured on an x86-64 computer with 48 CPU cores (Intel Xeon 6248R), 2 hyperthreads per core, 78 GB of memory, and without the use of a GPU. |
| Software Dependencies | No | The paper mentions the use of MATLAB, TFOCS, FISTA, C++, POSIX threads, and TPOPP library, but it does not provide specific version numbers for these software components, which are required for reproducible descriptions. |
| Experiment Setup | Yes | For a kernel with diameter of 30 pixels this improves the performance by a factor of over 100 without substantial degradation of false transients removal or the recognition of the interfering components shape W c. We benchmarked the patch-based parallel processing of real SEUDO 80x80 pixel patches. Specifically we simulated the neural activity over 20000 frames at 30Hz with fame size of 500x500 pixels. They consisted of 36 videos, each sized 90x90 pixels with 41750 frames sampled at 30 Hz. For a reasonable value of λ = 0.15, real SEUDO had a true positive rate of 75% and a false positive rate of 24%. |