Brain virtual histology and volume measurement of a lizard species (Podarcis bocagei) using X-ray micro-tomography and deep-learning segmentation.

There is an increasing emphasis on understanding individual variation in brain structure-such as overall brain size and the size of specific regions-and linking this variation to behaviour, cognition, and the driving social and environmental factors. However, logistical challenges arise when studying the brain, especially in research involving wild animals, such as dealing with small sample sizes and time-consuming methods. In this study, we used wild lizards, Podarcis bocagei, as our model. We developed an efficient and accurate method that combines X-ray micro-tomography and deep-learning segmentation to estimate the volume of six main brain areas: the olfactory bulbs, telencephalon, diencephalon, midbrain, cerebellum, and brain stem. Through quantitative comparisons, we show that a sufficiently trained deep-learning neural network can be developed with as few as five samples. Using this trained model, we obtained volume data for the six brain regions from 29 brain samples of Podarcis bocagei. This approach drastically reduced the time needed for manual segmentation from several months to just a few weeks. We present a comprehensive protocol detailing our methods, which includes sample preparation, X-ray tomography, and 3D volumetric segmentation. This work collectively provides valuable resources that can assist researchers not only in animal behaviour and physiology, but also in biomedical research and computer sciences.