ID-9-O-2596 Automatic computation of emphysema maps on histological lung samples of treated mice

Introduction

Emphysema manifests as a component of chronic obstructive pulmonary disease (COPD), which is expected to be the third most common cause of death in 2020. Animal models such as elastase-induced emphysema mice have been used to study COPD. In this context, a method for a reliable quantification of emphysema lesions is required to evaluate the stage of the disease and the effect of treatments. The second moment of the equivalent diameter variable (D₂) has shown a robust behaviour in emphysema quantification since it does not depend on the shape of the airspace and is sensible to a heterogeneous distribution of them¹. However, the value of D₂ by itself does not provide information about the degree of emphysema severity.

Data

All experimental protocols involving animal manipulation were approved by the University of Navarra Experimentation Ethics Committee. Lung lobe sections (H&E stained) from control and emphysema-treated mice were obtained using an automated Axioplan 2ie Zeiss microscope (Carl Zeiss, Jena, Germany). Each slide was initially acquired with a Plan-Neofluar objective (numerical aperture NA = 0.035, magnification 1.25x, pixel resolution 3.546 μm/pixel). The images of fourteen lung sections were available for the present study: 12 training images (6 control and 6 treated) and 2 test images (1 control and 1 treated). A total of 399 patches (751x751 pixels) were extracted from the training sections: 190 from normal (N) and 209 from emphysematous (E) tissue areas of control and treated mice, respectively.

Methods

Given a lung lobe image, semi-automatic segmentation based on morphological operators was applied to identify parenchyma pixels. Subsequently, a 751x751 window was centred on each pixel and the corresponding value of D₂ was computed. A Bayesian approach was used to map a D₂ value onto a probability index indicating emphysema severity. The posterior probability of being emphysematous was obtained as p(E|D₂) = p(D₂|E)p(E)/p(D₂), where p(D₂) = p(D₂|E)p(E) + p(D₂|N)p(N). The density functions p(D₂|E) and p(D₂|N) were obtained from samples in the training set using the Parzen’s method.

Results

Figures 1 and 2 show the emphysema map of the two lung sections in the test set (higher intensity corresponds to higher probability of emphysema). The percentage of parenchyma labelled as emphysematous, i.e., p(E|D₂) > 0.5, was 17.9% and 83.3% for the control and the treated mice, respectively.

Conclusion

The proposed method could serve as an objective tool for the evaluation of emphysema severity in the context of COPD study.

[1] Parameswaran H, Majumdar A, Ito S et al., “Quantitative characterization of airspace enlargement in emphysema,” J. Appl. Physiol., 100, 186-193 (2006).