Automated 3D mouse lung segmentation from CT images for extracting quantitative tumor progression biomarkers

Citations Over Time

Abstract

Genetically engineered mouse models of lung cancer are essential for preclinical evaluation of disease progression and treatments as well as in drug development. Micro-computed Tomography (microCT) is an imaging modality that is widely used in visualizing the anatomy of subjects in vivo and extracting quantitative and translatable biomarkers. This work demonstrates the use of uCT imaging and image segmentation techniques in large population phenotyping studies of transgenic mouse models of lung cancer. We studied 8 genotypes of transgenic mice with 99 subjects imaged at 4 time points. We developed (1) a high throughput image acquisition technique that acquires 60 subjects in 3 hours at an isotropic resolution of about 100 um, and (2) an automated segmentation algorithm to compute tumor and vasculature volume (TVV), a previously validated biomarker for lung cancer progression. TVV is computed as the difference between the whole lung and the functional lung (air space within lung) volumes. Previous work on automated lung segmentation focused on healthy lung or on segmentation of pulmonary nodules. We automatically compute TVV by determining a lung region of interest (ROI) by using the rib cage, the functional lung volume by thresholding within the lung ROI, and the whole lung volume by iteratively performing morphological hole-fill, bridge, and image close operations on the functional lung. We compare the automated results with that of manual analysis. Automated functional lung volume results were highly correlated to manual results (R2≥0.95) at all the time points. Whole lung volume was well-correlated to manual measurements (R2≥0.8 up to the 2nd time point), but required some manual correction at later time points when the tumors almost filled the lung. Overall this approach provided about 66% time saving compared to manual analysis. Our innovative workflow with high throughput acquisition and automated segmentation enabled efficient phenotyping studies to aid drug development.

Related Papers

• A Survey of Thresholding Methods for Image Segmentation(2002)
• → Counting objects in an image by marker controlled watershed segmentation and thresholding(2013)23 cited
• A survey of thresholding algorithms on yarn images(2010)
• → The Study on the Image Thresholding Segmentation Algorithm(2015)2 cited
• → A Thresholding Algorithm Based on Probabilistic Rough Set for Image Segmentation(2008)1 cited