Advances in intelligent animal pose tracking for neuro-behavioral integration

Animal behavior serves as an external manifestation of neural function and a key window into brain mechanisms and disease states. Traditional behavioral studies relying on manual observation and empirical classification suffer from subjective bias and limited spatiotemporal resolution, hindering their ability to quantitatively link behavior with neural activity. With rapid advances in computer vision and deep learning, pose tracking has evolved from Two-Dimensional (2D) estimation to Three-Dimensional (3D) reconstruction and further toward multimodal alignment. This review summarizes the evolutionary trajectory of machine learning-based animal pose tracking methods, ranging from early unsupervised and supervised paradigms to deep neural frameworks achieving high-precision 2D and 3D tracking, and further to the integration of frontier technologies—including Foundation Models, Generative Models, and Novel Neural Networks—alongside multimodal alignment with neural activity. Deep learning enables high-fidelity behavioral phenotyping and pose-neural mapping, driving advancements in multi-animal tracking and neuro-medical research. Despite remarkable progress, challenges persist in cross-species generalization, inconsistent annotation standards, and incomplete modeling of behavior–neural causality. Future research will emphasize self-supervised and generative learning to reduce annotation dependency, and multimodal temporal–spatial integration to align neural states with behavioral dynamics. The convergence of computational ethology and neuro-medicine is transforming neuroscience from passive observation to mechanistic understanding, advancing research and intervention in neurological and psychiatric disorders.