| Diagnosis and monitoring of kidney transplant allografts is supported by microscopic analysis of renal biopsy samples. Visual analysis by pathologists allows for inconsistencies, bias, and inaccuracies; image analysis via digital processing can address these concerns, reduce effort, and potentially provide a second opinion. In this thesis, digital image analysis methods for automatic segmentation of structures in images of renal biopsy samples are presented. Methods for accurate segmentation of the effective biopsy area include opening by reconstruction, morphological closing, and erosion. The results were compared to contours drawn by an experienced pathologist; the mean distance to the closest point was 5.46 ± 3.92 µm (6 ± 4.31 pixels) and the true-positive fraction was 98.25 ± 1.77%. Methods for automatic segmentation of cell nuclei are also presented, including, automatic thresholding, adaptive thresholding, and morphological granulometry. The results were verified against pathologist annotations; true-positive ratios were in the range of 0.80 to 0.93. |
