Investigating the Impact of the Bit Depth of Fluorescence-Stained Images on the Performance of Deep Learning-Based Nuclei Instance Segmentation

Abstract

Nuclei instance segmentation can be considered as a key point in the computer-mediated analysis of histological fluorescence-stained (FS) images. Many computer-assisted approaches have been proposed for this task, and among them, supervised deep learning (DL) methods deliver the best performances. An important criterion that can affect the DL-based nuclei instance segmentation performance of FS images is the utilised image bit depth, but to our knowledge, no study has been conducted so far to investigate this impact. In this work, we released a fully annotated FS histological image dataset of nuclei at different image magnifications and from five different mouse organs. Moreover, by different pre-processing techniques and using one of the state-of-the-art DL-based methods, we investigated the impact of image bit depth (i.e., eight bits vs. sixteen bits) on the nuclei instance segmentation performance. The results obtained from our dataset and another publicly available dataset showed very competitive nuclei instance segmentation performances for the models trained with 8 bit and 16 bit images. This suggested that processing 8 bit images is sufficient for nuclei instance segmentation of FS images in most cases. The dataset including the raw image patches, as well as the corresponding segmentation masks is publicly available in the published GitHub repository.

Keywords: bit depth; computational pathology; deep learning; fluorescence staining; medical image analysis; nuclei segmentation.

Figure 1

Examples of kidney images at different image magnifications with their corresponding labelled and binary segmentation masks.

Figure 2

Example of raw, outlier-removed and normalised 16 bit images of a bone sample at 40× magnification from left to right in the first row, respectively. We also plot a profile intensity line (depicted with dashed red lines in the images) in the second row. The raw 16 bit image was selected from the BitDepth dataset.

Figure 3

Example of raw, outlier-removed and normalised 8 bit images of a bone sample at 40× magnification from left to right in the first row, respectively. We also plot a profile intensity line (depicted with dashed red lines in the images) in the second row. The raw 8 bit image was selected from the BitDepth dataset and corresponds to the 16 bit image presented in Figure 2.

Figure 4

Differences between the AJI and PQ scores (%) for 8 bit and 16 bit images based on the number of nuclei in the images for different image magnifications in the BitDepth dataset. For each magnification, Dataset 4 was used to measure the performance differences between 8 bit and 16 bit images.

Figure 5

Nuclei instance segmentation performance boxplots for Dice, AJI and PQ scores derived from the BitDepth dataset (Y-axes are limited for a better visualisation, and hence, some outliers are not shown).

Figure 6

Nuclei instance segmentation performance boxplots for Dice, AJI and PQ scores derived from the Caicedo et al. dataset [10] (Y-axes are limited for better visualisation, and hence, some outliers are not shown).

Figure 7

Three images from the Caicedo et al. dataset [10] with zero Dice, AJI and PQ scores. There were no nuclei in the ground truth segmentation masks for these three cases.

Figure 8

The average nuclei instance segmentation performance based on Dice, AJI and PQ scores for different image bit depths for the BitDepth and Caicedo et al. datasets.