Improved passive autofocus algorithm for fluorescence microscopy by non-parametric blurriness

Abstract

In a fluorescence microscope, passive autofocus processing provides the best-focused cellular image in optical visualization during image collection. The steep focus curve is essential to high throughput microscopy for the application of cell images by reducing the number of stack image acquisitions to be focus-measured. The proposed scheme combines the non-parametric blur ratio computed by the two-step Otsu’s threshold method to region of interest (ROI)-based autofocus measures based on the conventional sharpness metrics. The blur ratio expedites the gentle focus curve caused by arbitrary plane selection of an out-focused image to be segmented among stack images. It makes up for irregular focus computation by the improper threshold value of Otsu’s method by bright big cells and epithelial debris. Also, the proposed blurriness curve of stack images identifies coarse and fine search areas in variable step-sized autofocus search algorithms, which decreases the number of image collection for calculating the focus metrics. In computer simulations with nuclei cell images containing U2OS cells stained with Hoechst from the Broad Bioimage Benchmark Collection (BBBC), the proposed autofocus algorithm yields superior sharpness performance of focus curve over the schemes applying the entire image and using focus ROI by Otsu’s method. Four types of cellular fluorescence images are evaluated in terms of normalized focus curves and quantitative criteria of steepness, sharpness ratio, and sensitivity. Additionally, the performance excellence of the proposed autofocus algorithm is assessed by testing the rule-based search algorithm with a varying search size and adapting the blurriness curve in terms of the quantitation of autofocus speed (number of stack image acquisitions) and accuracy of focus position (not to lose best focus position).