Differentiation of arterioles from venules in mouse histology images using machine learning

Abstract

Analysis and morphological comparison of the arteriolar and venular components of a microvascular network are essential to our understanding of multiple diseases affecting every organ system. We have developed and evaluated the first fully automatic software system for differentiation of arterioles from venules on high-resolution digital histology images of the mouse hind limb immunostained with smooth muscle [Formula: see text]-actin. Classifiers trained on statistical and morphological features by supervised machine learning provided useful classification accuracy for differentiation of arterioles from venules, achieving an area under the receiver operating characteristic curve of 0.89. Feature selection was consistent across cross validation iterations, and a small set of two features was required to achieve the reported performance, suggesting the generalizability of the system. This system eliminates the need for laborious manual classification of the hundreds of microvessels occurring in a typical sample and paves the way for high-throughput analysis of the arteriolar and venular networks in the mouse.

Keywords: arteriole venule classification; feature analysis; machine learning; vasculature quantification; whole slide analysis.

Fig. 1

Arteriole with the DAB stain (a) and the binary mask (b) output from the automated segmentation. Scale bar is $50\mu \mathrm{m}$.

Fig. 2

Arterioles and venules are shown in the normal and regenerated tissues. The microvessels have distinguishable differences in the normal vasculature, where the smooth muscle layer is thicker and more pronounced on the arteriolar side and have a visibly darker stain. This difference is less apparent in the regenerated vasculature.

Fig. 3

Block diagram to demonstrate the method followed in this experiment for a LOMO cross validation.

Fig. 4

AUC (a) and error rate (b) as a function of number of features used for classification. The error bars represent the standard error of the mean. All features were chosen using a forward feature selection, and classifiers were trained with a LOMO cross validation.

Fig. 5

Three features chosen from forward feature selection over the 10 cross validations.

Fig. 6

ROC curves for each of the classifiers trained on two features. LOGLC AUC: 0.89, SVM AUC: 0.89, RFC AUC: 0.84.

Fig. 7

Histogram of the confidences from the LOGLC (a, b), SVM (c, d) and RFC (e, f) classification using two features with their respective confidence thresholds, where “none” indicates no confidence threshold was applied and all vessels were categorized.

Fig. 8

AUC (a) and error rate (b) as a function of number of features used for classification. All features were chosen using a forward feature selection, and classifiers were trained with an independent training set.

Fig. 9

ROC curves for each of the classifiers from classifier trained on two features. LOGLC AUC: 0.91, SVM AUC: 0.92, RFC AUC: 0.89.

Fig. 10

Histogram of the confidences from the LOGLC (a,b), SVM (c,d) and RFC (e,f) classifications with their respective confidence thresholds.

Fig. 11

Incorrectly (a)–(e) and correctly (f)–(j) classified arterioles by the LOGLC. Scale bar is $20\mu \mathrm{m}$.