Biophysical simulation enables segmentation and nervous system atlas mapping for image first spatial omics.

Spatial omics (SO) produces high-definition mapping of subcellular molecules within tissue samples. Mapping transcripts to anatomical regions requires segmentation, but this remains challenging in tissue cross-sections with tubular structures like axons in peripheral nerve or spinal cord. Neural networks could address misidentification but are hindered by the need for extensive human annotations. We present SiDoLa-NS (Simulate, Don't Label-Nervous System), an image-driven (top-down) approach to SO analysis in the nervous system. We utilize biophysical properties of tissue architectures to design synthetic images of tissue samples, eliminating reliance on manual annotation and enabling scalable training data generation. With synthetic samples, we trained supervised instance segmentation convolutional neural networks (CNNs) for nucleus segmentation, achieving precision and F1-scores>0.95. We further identify macroscopic tissue structures in mouse brain (mAP50=0.869), spinal cord (mAP50=0.96), and pig sciatic nerve (mAP50=0.995). This framework sets the stage for transferable models across species and tissue architectures-accelerating SO applications in neuroscience and beyond.