Automatic intensity quantification of fluorescence targets from microscope images with maximum likelihood estimation

We introduce a method to determine the quantitative intensity values and sub-pixel locations of closely located small targets from noisy fluorescence microscope images. We model the microscope image with a mixture of point spread functions and the image noise with a stochastic process containing Poisson distribution. Maximum likelihood estimation is used to find the optimal parameters for the model. Numerical ML estimation is performed with differential evolution optimization algorithm. To evaluate the methods, noisy simulated images were created with closely located targets. Methods were compared to conventional methods based on low-pass filtering and Gaussian mixture fitting, and the simulations show better accuracy for the new method. A real microscope image is also quantified to show that the model is applicable in practice.