A novel deep-learning approach for detecting neurons of the primary visual cortex and orientation selectivity

Calcium imaging microscopic technique uses fluorescent calcium indicators that bind with calcium ions (Ca
²⁺
) to measure their levels in cells or tissues optically. Two-photon calcium imaging allows tracking the activity of thousands of neurons in mammalian’s brain at high resolution by monitoring the changes in Ca
²⁺
levels. Neurons express orientation selectivity in the Primary Visual Cortex (V1) and can be determined by studying their activities. Drawing regions of interest manually requires a great deal of time to extract data for each neuron. Calcium imaging data must be analyzed to understand neurons' functions, but it is computationally expensive to process. This research presents a pipeline that begins with detecting V1 neurons automatically by combining deep learning models. The resulting average F-1 score is 0.89 based on testing data. These results outperform state-of-the-art methods that don't surpass 0.85 in testing F-1 score. Using signal processing techniques, each neuron's preferred orientation was determined by analyzing the data extracted from it. The orientation selectivity of V1 neurons has inspired new biological discoveries related to the origin of vision and the development of modern computer vision models such as Convolutional Neural Networks (CNN).