Machine learning enhanced neuroimaging

Recent advances in brain imaging and molecular analysis techniques offer unprecedented opportunities for a deeper understanding of neurodegenerative disorders such as ALS, Alzheimer's, and Parkinson's. High-throughput physiological approaches, including two-photon microscopy, are powerful tools for revealing subtle functional changes. Complementary genomics and connectomics approaches, including MAPseq [1] and spatial transcriptomics methods like BARseq2 [2], can reveal changes in brain wiring and gene expression at the single cell level with high throughput. Spatial transcriptomics relies on high-resolution optical microscopy techniques, which are often low throughput. Hence, techniques that provide higher throughput, faster acquisition with a higher resolution are desirable. Higher resolution could help in improving the efficiency of feature detection, resulting in higher accuracy in deciphering transcriptomic reads. The aim is, therefore, to enhance the spatial resolution of spatial transcriptomics/connectomics (BARseq2) using deep learning. We demonstrate a deep-learning-enabled resolution enhancement technique and computational aberration correction for optical microscopy.

References:

1. Kebschull, Justus M., et al. "High-throughput mapping of single-neuron projections by sequencing of barcoded RNA." Neuron 91.5 (2016): 975-987.

2. Sun, Yu-Chi, et al. "Integrating barcoded neuroanatomy with spatial transcriptional profiling enables identification of gene correlates of projections." Nature Neuroscience 24.6 (2021): 873-885.