End-To-End Automated Mean Linear Intercept Measurement System

Abstract

The Mean Linear Intercept (MLI) measurement is a quantitative metric for assessing air space size in histopatho- logical images of lung tissue. Currently, MLI measurement in- volves human raters conducting manual image assessment that is time-consuming, labour-intensive, and subject to inter- and intra-rater variability. Our system utilizes a deep learning ap- proach for semantic segmentation to achieve a fully automated MLI measurement system with a graphical user interface. The system was trained on mouse lung images and tested on a rat lung image to investigate the generalizability to other animal models. The system computed an MLI score of 62.20 within 90 minutes using 8255 field-of-view (FOV) images extracted from the whole slide image. The human rater found the MLI score to be 62.49 within 41 minutes using 500 randomly selected FOVs. This result suggests that the system maintains its accuracy with rat lung images. Although the system took twice as long as the human rater, it processed >16× more FOVs, which leads to lower standard error in the MLI score.