∗Biomedical Imaging Group Rotterdam, Departments of Medical Informatics and Radiology

In live-cell fluorescence microscopy imaging, quantitative analysis of biological image data generally involves the detection of many subresolution objects, appearing as diffraction-limited spots. Due to acquisition limitations, the signal-to-noise ratio (SNR) can be ex-tremely low, making automated spot detection a very challenging task. In this paper, we quantitatively evaluate the performance of the most frequently used supervised and unsupervised detection meth-ods for this purpose. Experiments on synthetic images of three dif-ferent types, for which ground truth was available, as well as on real image data sets acquired for two different biological studies, for which we obtained expert manual annotations for comparison, re-vealed that for very low SNRs (≈2), the supervised (machine learn-ing) methods perform best overall, closely followed by the detectors based on the so-called h-dome transform from mathematical mor-phology and the multiscale variance-stabilizing transform, which do not require a learning stage. At high SNRs (>5), the difference in performance of all considered detectors becomes negligible. Index Terms—Object detection, noise reduction, image filter-ing, machine learning, fluorescence microscopy. 1