Automatic Detection Of Immunogold Particles From Electron Microscopy Images

Immunogold particle detection is a time-consuming task where a single image containing almost a thousand particles can take several hours to annotate. In this work we present a framework for the automatic detection of immunogold particles that can leverage significantly the burden of this manual task. Our proposal applies a Laplacian of Gaussian (LoG) filter to provide its detection estimates to a Stacked Denoising Autoencoder (SdA). This learning model endowed with the capability to extract higher order features provides a robust performance to our framework. For the validation of our framework, a new dataset was created. Based on our work, we determined that solely the LoG detector attained more than 74.1% of accuracy and, when combined with a SdA the accuracy is improved by at most 11.4%.