Evaluating techniques for neuron identification in complex cultures: a deep learning approach

Microscopy image analysis of neurons cultures represents a formidable challenge due to their complex structure because of the dynamic nature of neurite tissue development, the neuron movement, the morphological changes, and the pres- ence of many elements in the cultures apart from neurons such as glial cells, dead cells, vesicles, etc. A rigorous eval- uation of deep learning techniques to address this intricate problem is undertaken in this study. Several methodolo- gies, including Instance Segmentation and Object Detection models, are scrutinized within a comprehensive experimen- tal framework. The efficacy of the Instance Segmentation model is underscored by the findings, demonstrating superior quantitative results. Precise neuron quantification is facili- tated by this model through the detection of bounding boxes in images, thereby enabling the automation of tasks such as morphological and size analysis of neuronal cells and track- ing individual neurons across sequential images from the same culture. This paper serves as a robust foundation for future investigations aimed at expanding the understanding of neurons cultures.