Quick & accurate neuron population quantification: An interactive, deep-learning accelerated method for neuron population quantification in mice brains

Quantification of neuron populations, stratified across brain regions, provides a foundation for understanding the function of brain circuits. For this reason, accurate quantification is vital. To achieve this, all tissue slices need to be aligned with a reference atlas, have neurons of interest segmented, and subsequently have per-region aggregates computed. These steps are repetitive, tedious, time-consuming, and require expert domain knowledge to perform manually. Automated solutions are therefore desired. However, fully automatic neuron population quantification remains a challenge, due to: High variance of acquisition techniques, non-aligned cutting angles, deformations of tissue, only partially imaged slices, imaging process artifacts, and large amounts of data involved what requires efficient processing. In this work, we address these challenges, aiming towards automated quantification to speed up experiment analysis. But our deep learning-based automations do not yet reach the necessary accuracy. We therefore develop a visual, interactive application, which provides automated suggestions, but let domain experts adjust and fine-tune the image analysis and neuron quantification.We believe this approach, packaged into a tailor-made application accommodating the complete quantification process, will serve to accelerate accurate quantification of neuron populations.Support: Novo Nordisk Foundation Laureate grant & Lundbeck Foundation R310201921