SynSeg: A synthetic data-driven approach for robust subcellular structure segmentation

Accurate subcellular segmentation is crucial for understanding cellular processes, but traditional methods struggle with noise and complex structures. Convolutional neural networks improve accuracy but require large, time-consuming, and biased manually annotated datasets. We developed SynSeg, a pipeline generating synthetic training data for a U-Net model to segment subcellular structures, eliminating manual annotation. SynSeg leverages synthetic datasets with varied intensity, morphology, and signal distribution, delivering context-aware segmentations, even in challenging conditions. We demonstrate SynSeg's superior performance in segmenting vesicles and cytoskeletal filaments from cells and live Caenorhabditis elegans, outperforming traditional methods like Otsu's, ILEE, and FilamentSensor 2.0 and a recent deep learning method. Additionally, SynSeg effectively quantified disease-associated microtubule morphology in live cells, uncovering structural defects caused by mutant Tau proteins linked to neurodegeneration. Furthermore, SynSeg enables high-throughput, automated analysis, revealing that BSCL2 disease mutations increase lipid droplet size and showing its broad generalizability for quantitative cell biology. These results highlight the potential of synthetic data to advance biological segmentation.