Performance evaluation of an image estimation method based on principal component analysis (PCA) developed for quantitative depth-variant fluorescence microscopy imaging

In 3D wide-field computational microscopy, the image formation process is depth variant due to the refractive index mismatch between the imaging layers. In a previous study, an image estimation method based on a principle component analysis (PCA) model for the representation of the depth varying point spread function (DV-PSF) was presented and demonstrated with noiseless simulations. In this study, the performance of the PCA-based DV expectation maximization algorithm (PCA-DVEM) was further evaluated with noisy simulations. Different levels of Poisson noise were used in simulated forward images of a synthetic object computed using theoretically-determined DV-PSFs approximated by the PCA approach. The noise influence on the reconstructed images obtained with PCA-DVEM was evaluated. We found that without regularization, the algorithm performs well when the signal-to-noise ratio (SNR) is 14 dB or higher. The relationship of the number of PCA components, B, to the image reconstruction performance was also investigated on both noiseless and noisy simulated data. In both cases, we found that the number of PCA components has limited effect on the image reconstruction performance for B > 1. To reduce computational complexity while maintaining image estimation performance, B = 2 is suggested for processing experimental data.