Image segmentation techniques for morphometric measurement of fish blood cells: A comparative study

Automated ellipse fitting methods offer a promising tool for measuring Morphometric Parameters of Fish Blood Cells (MPFBCs) from digital microscope images. However, their performance depends greatly on the quality of the binary mask obtained from the image segmentation. Poor segmentation often results in a binary mask with a broken cell border, leading to incorrect ellipse fitting thereby compromising MPFBC measurement results. The current paper employs nine well-established image segmentation techniques to conduct their comprehensive performance analysis, identifying the top-performing methods in terms of measuring the correct area of fish blood cells. The results of the implemented methods are compared with measurements obtained manually from expert biologists. Results demonstrate that each single segmentation method exhibits some outlier measurements compared with the biologist's confidence interval. Therefore, the identified top methods are combined to avoid outliers, aiming to improve MPFBC measurements. Results indicate that the proposed combination of the top three methods can reduce outliers, achieving the lowest median absolute error of 51 pixels and an Interquartile Range (IQR) of 60 pixels for the cell area measurement. For the nucleus area, among all methods, Global Thresholding-based segmentation achieves the lowest median error of 12 pixels and an IQR of 17 pixels, without needing to combine more segmentation approaches.