An Automated Tool to Assess Air Space Size in Histopathology Images of Lung Tissue
Diego A Politis, Sina Salsabili, Adrian D. C. Chan · 2022 IEEE International Instrumentation and Measurement Technology Conference (I2MTC) · 2022
The mean linear intercept (MLI) score is a useful and common approach for quantifying lung structure in histopathological images. In this paper, we describe a computer tool to determine the MLI score in a fully automated manner. A multiresolution semantic segmentation approach is used to segment the various structures within whole slide images (WSIs) of the lungs. Next, multiple field-of-view (FOV) images from the original WSI and masks are extracted. The extracted FOVs are screened, rejecting images that contain bronchi or blood vessels within the region of interest and accepting those that remain. The accepted FOVs are then used to calculate the MLI score using an intersection counting approach. The automated tool was tested using 20 WSIs from mice that were exposed to one of four conditions that affected their lung structure. The root-mean-squared deviation between the MLI score of our proposed method and a human rater was 5.73 (standard deviation 5.65), and there was a very strong correlation (r=0.9931). The proposed automated tool provides an efficient, accurate, and accessible method that could replace current manual and semi-automated techniques.