Prediction of fast subcellular calcium signals based on label-free second harmonic generation microscopy measurements

Abstract

Calcium (Ca²⁺) sparks were predicted based on the label-free second harmonic generation (SHG) signal of myosin filaments with pix2pix, a conditional adversarial network. Predicted Ca²⁺ sparks and ground truth Ca²⁺ sparks show similar properties. Predicted Ca²⁺ spark frequencies and a predominant location at the periphery of skeletal muscle fibers were in good agreement with ground truth data. High frequencies of predicted Ca²⁺ sparks could be observed at y-shaped structures of myosin filaments. Important image pre-processing was applied in order to enhance the signal of Ca²⁺ sparks used for training. Only open-source software was applied. I provide the image pre-processing algorithm and several algorithms to evaluate the quality of predicted Ca²⁺ sparks. This technique can be extended to other label-free microscopy techniques and other fluorescent indicators developed during the last years. It remains a difficult task to predict the exact location of Ca²⁺ sparks.