Segmentation of patchy areas in biomedical images based on local edge density estimation

We suggest an effective approach for the semi-automated segmentation of biomedical images according to their patchiness based on local edge density estimation. Our approach does not require any preliminary learning or tuning, although a couple of free parameters directly controllable by the end user adjust the analysis resolution and sensitivity, respectively. We show explicitly that the local edge density exhibits excellent correlations with the cell monolayer density obtained by manual domain-expert based assessment, characterized by correlation coefficients rho > 0.97. Our results indicate that the proposed algorithm is capable of an efficient segmentation and quantification of patchy areas in various biomedical microscopic images. In particular, the proposed algorithm achieves 95 to 99% median accuracy in the segmentation of image areas covered by the cell monolayer in an in vitro scratch assay. Moreover, the proposed algorithm effectively distinguishes between the native and regenerated tissue fragments in microscopic images of histological sections, indicated by nearly three-fold discrepancy between the local edge densities in the corresponding image areas. We believe that the local edge density estimate could be further applicable as a surrogate image channel characterizing its patchiness either as a substitute or as a complementary source to the conventional cell-or tissue-specific fluorescent staining, in some cases either avoiding or limiting the use of complex experimental protocols. We implemented a simple open-source software tool with for on-the-fly visualization allowing for a straightforward feedback by a domain expert without any specific expertise in image analysis techniques. Our tool is freely available online at https://gitlab.com/digiratory/biomedimaging/bcanalyzer.

Automated summary

The proposed method effectively segments biomedical images based on local edge density and achieves high accuracy without the need for preliminary learning or tuning. The algorithm's segmentation and quantification capabilities are validated in various biomedical microscopic images, showcasing its potential for efficient image analysis in the field.