AI-based mouse behavior analysis in pathology: A focus on movement disorders

Quantifying behavior in animal models is essential for understanding neurological disorders, yet traditional scoring methods often fail to capture the complexity and heterogeneity of motor dysfunction. Recent advances in artificial intelligence (AI) have enabled high-resolution, scalable analysis of rodent behavior through video-based tracking, behavioral decomposition, and inertial sensor-based motion tracking. In this review, we provide a comprehensive synthesis of AI-based approaches for behavioral analysis, with a specific focus on pathological motor phenotypes, particularly in rodent models of Parkinson's disease and L-DOPA-induced dyskinesia. We compare supervised, unsupervised, and hybrid pipelines, examining how tools such as MoSeq, B-SOiD, VAME, SimBA, A-SOiD, and inertial measurement units (IMUs)-based frameworks extract latent motor motifs and classify complex behaviors. We highlight the growing importance of multimodal strategies integrating video, inertial sensing, and neural recordings to link behavioral features to underlying neural activity. Beyond technical advances, we discuss key conceptual challenges, including interpretability, cross-laboratory generalization, and the alignment of AI-derived behavioral units with clinically meaningful motor symptoms. Together, these advances point to a paradigm shift in preclinical phenotyping: from descriptive scoring to biologically informed, AI-powered behavioral analysis.