D-LMBmapX: Generalized Deep-Learning Pipeline for Automated Whole-brain Neural Circuitry Profiling Across Any Developmental Stages

Understanding whole-brain circuitry development in mice has been hindered by the lack of tools for accurate and comprehensive analysis. No existing 3D brain atlas provides daily postnatal resolution, as constructing such atlases is highly labour-intensive. Axonal morphologies change dynamically, making reliable segmentation challenging, and many 2D datasets lack sufficient Z-resolution for cross-modality 3D analysis. Here, we present D-LMBmapX, a deep-learning pipeline for automated whole-brain circuitry profiling across postnatal development. D-LMBmapX constructs high-resolution 3D mouse brain atlases spanning seven postnatal stages and employs an adaptive registration strategy for whole-brain alignment at any postnatal day. It also integrates a foundation model for axon and soma segmentation, enabling the quantitative circuitry assessment across development. We implemented a diffusion model-based style transfer for cross-modality and cross-dimensional registration, validated by aligning genetically defined neuronal types from 2D ISH datasets to our 3D atlas. Using D-LMBmapX, we profiled whole-brain dopaminergic projections throughout postnatal maturation.