Automatic Diabetes Detection from Histological Images of Rats Phrenic Nerve Using Two-Dimensional Sample Entropy

Abstract

In microscopy, morphological characteristic of the axon are the most common features assessed in histological images of nerves. Although morphometric indexes are widely used to describe histological data, the calculation of those indexes is a highly time-consuming task that demands great manual effort from the specialist. Recently, two-dimensional sample entropy (SampEn2D) was proposed to quantify the degree of irregularity present in an image, based on the spatial patterns of pixels. Here, we propose the use of SampEn2D as a suitable metric for classifying diabetic status of rats from histological images of the phrenic nerve. Microscopy images of three different Wistar rats groups (untreated diabetic (N = 24), insulin-treated diabetic (N = 9), and non-diabetic control (N = 11)) were analysed. The results show that for the optimal SampEn2D parameters (m = 1, r = 0.1), control rats have significantly (p < 0.01) lower entropy (3.76 ± 0.26) as compared to both insulin-treated (5.09 ± 0.20) and untreated (5.30 ± 0.13) diabetic animals. Performance of SampEn2D for image classification between untreated and control groups was assessed by ROC analysis and area under the ROC curves (AUROC = 0.96). SampEn2D reaches a sensitivity of 87% and specificity of 82% when a threshold of SampEn2D = 4.73 is taken to separate the two assessed groups. In conclusion, SampEn2D method arises as a useful tool in the screening of diabetic rats. The method may be useful to pre-select animals for further morphometric evaluation, reducing the burden of manual processing.