Cell Detection in Microscopy Images with Deep Convolutional Neural Network and Compressed Sensing.

The ability to automatically detect certain types of cells or cellularsubunits in microscopy images is of significant interest to a wide range ofbiomedical research and clinical practices. Cell detection methods have evolvedfrom employing hand-crafted features to deep learning-based techniques. Theessential idea of these methods is that their cell classifiers or detectors aretrained in the pixel space, where the locations of target cells are labeled. Inthis paper, we seek a different route and propose a convolutional neuralnetwork (CNN)-based cell detection method that uses encoding of the outputpixel space. For the cell detection problem, the output space is the sparselylabeled pixel locations indicating cell centers. We employ random projectionsto encode the output space to a compressed vector of fixed dimension. Then, CNNregresses this compressed vector from the input pixels. Furthermore, it ispossible to stably recover sparse cell locations on the output pixel space fromthe predicted compressed vector using $L_1$-norm optimization. In the past,output space encoding using compressed sensing (CS) has been used inconjunction with linear and non-linear predictors. To the best of ourknowledge, this is the first successful use of CNN with CS-based output spaceencoding. We made substantial experiments on several benchmark datasets, wherethe proposed CNN + CS framework (referred to as CNNCS) achieved the highest orat least top-3 performance in terms of F1-score, compared with otherstate-of-the-art methods.