Deep Learning-Based Automated Detection of Ca ² ⁺ Sparks Using Synthetic Training Data

Calcium (Ca 2 + ) sparks are fundamental signaling events essential for cell excitability. Ca 2 + sparks are usually recorded by confocal imaging. Automated Ca2+ spark image analysis approach is highly demanded. While deep learning offers a powerful approach for automated image analysis, its success typically depends on the availability of large, annotated datasets. Due to the labor consuming nature of Ca2+ imaging experiments, it is difficult to obtain large training datasets for Ca2+ spark detection. To address this limitation, in this study, we developed a synthetic data-driven approach to train deep learning models for automated detection and analysis of Ca2+ sparks. Without relying on large-scale manual annotations, this strategy supports scalable, reproducible training while preserving biological relevance. The trained models demonstrate strong performance on both synthetic and experimental validation datasets. This work introduces a practical and generalizable framework for high-throughput analysis of Ca 2 + signaling. More broadly, it illustrates the potential of synthetic data to expand deep learning applications across biomedical image and signal analysis domains. This abstract was presented at the American Physiology Summit 2026 and is only available in HTML format. There is no downloadable file or PDF version. The Physiology editorial board was not involved in the peer review process.