Mouse lung automated segmentation tool for quantifying lung tumors after micro-computed tomography

Applying deep learning to segmentation of murine lung tumors in pre-clinical micro-computed tomography

Lung cancer remains a leading cause of cancer-related death, but scientists have made great strides in developing new treatments recently, partly owing to the use of genetically engineered mouse models (GEMMs). GEMM tumors represent a translational model that recapitulates human disease better than implanted models because tumors develop spontaneously in the lungs. However, detection of these tumors relies on in vivo imaging tools, specifically micro-Computed Tomography (micro-CT or µCT), and image analysis can be laborious with high inter-user variability. Here we present a deep learning model trained to perform fully automated segmentation of lung tumors without the interference of other soft tissues. Trained and tested on 100 3D µCT images (18,662 slices) that were manually segmented, the model demonstrated a high correlation to manual segmentations on the testing data (r2=0.99, DSC=0.78) and on an independent dataset (n = 12 3D scans or 2328 2D slices, r2=0.97, DSC=0.73). In a comparison against manual segmentation performed by multiple analysts, the model (r2=0.98, DSC=0.78) performed within inter-reader variability (r2=0.79, DSC=0.69) and close to intra-reader variability (r2=0.99, DSC=0.82), all while completing 5+ hours of manual segmentations in 1 minute. Finally, when applied to a real-world longitudinal study (n = 55 mice), the model successfully detected tumor progression over time and the differences in tumor burden between groups induced with different virus titers, aligning well with more traditional analysis methods. In conclusion, we have developed a deep learning model which can perform fast, accurate, and fully automated segmentation of µCT scans of murine lung tumors.

Measurement of mouse head and neck tumors by automated analysis of CBCT images

Animal experiments are often used to determine effects of drugs and other biological conditions on cancer progression, but poor accuracy and reproducibility of established tumor measurement methods make results unreliable. In orthotopic mouse models of head and neck cancer, tumor volumes approximated from caliper measurements are conventionally used to compare groups, but geometrical challenges make the procedure imprecise. To address this, we developed software to better measure these tumors by automated analysis of cone-beam computed tomography (CBCT) scans. This allows for analyses of tumor shape and growth dynamics that would otherwise be too inaccurate to provide biological insight. Monitoring tumor growth by calipers and imaging in parallel, we find that caliper measurements of small tumors are weakly correlated with actual tumor volume and highly susceptible to experimenter bias. The method presented provides a unique window to sources of error in a foundational aspect of preclinical head and neck cancer research and a valuable tool to mitigate them.

A fully automated micro‑CT deep learning approach for precision preclinical investigation of lung fibrosis progression and response to therapy

Micro-computed tomography (µCT)-based imaging plays a key role in monitoring disease progression and response to candidate drugs in various animal models of human disease, but manual image processing is still highly time-consuming and prone to operator bias. Focusing on an established mouse model of bleomycin (BLM)-induced lung fibrosis we document, here, the ability of a fully automated deep-learning (DL)-based model to improve and speed-up lung segmentation and the precise measurement of morphological and functional biomarkers in both the whole lung and in individual lobes. µCT-DL whose results were overall highly consistent with those of more conventional, especially histological, analyses, allowed to cut down by approximately 45-fold the time required to analyze the entire dataset and to longitudinally follow fibrosis evolution and response to the human-use-approved drug Nintedanib, using both inspiratory and expiratory μCT. Particularly significant advantages of this µCT-DL approach, are: (i) its reduced experimental variability, due to the fact that each animal acts as its own control and the measured, operator bias-free biomarkers can be quantitatively compared across experiments; (ii) its ability to monitor longitudinally the spatial distribution of fibrotic lesions, thus eliminating potential confounding effects associated with the more severe fibrosis observed in the apical region of the left lung and the compensatory effects taking place in the right lung; (iii) the animal sparing afforded by its non-invasive nature and high reliability; and (iv) the fact that it can be integrated into different drug discovery pipelines with a substantial increase in both the speed and robustness of the evaluation of new candidate drugs. The µCT-DL approach thus lends itself as a powerful new tool for the precision preclinical monitoring of BLM-induced lung fibrosis and other disease models as well. Its ease of operation and use of standard imaging instrumentation make it easily transferable to other laboratories and to other experimental settings, including clinical diagnostic applications.

Micro-CT acquisition and image processing to track and characterize pulmonary nodules in mice

X-ray computed tomography is a reliable technique for the detection and longitudinal monitoring of pulmonary nodules. In preclinical stages of diagnostic or therapeutic development, the miniaturized versions of the clinical computed tomography scanners are ideally suited for carrying out translationally-relevant research in conditions that closely mimic those found in the clinic. In this Protocol, we provide image acquisition parameters optimized for low radiation dose, high-resolution and high-throughput computed tomography imaging using three commercially available micro-computed tomography scanners, together with a detailed description of the image analysis tools required to identify a variety of lung tumor types, characterized by specific radiological features. For each animal, image acquisition takes 4-8 min, and data analysis typically requires 10-30 min. Researchers with basic training in animal handling, medical imaging and software analysis should be able to implement this protocol across a wide range of lung cancer models in mice for investigating the molecular mechanisms driving lung cancer development and the assessment of diagnostic and therapeutic agents.

Automated segmentation of lungs and lung tumors in mouse micro-CT scans

Here, we have developed an automated image processing algorithm for segmenting lungs and individual lung tumors in in vivo micro-computed tomography (micro-CT) scans of mouse models of non-small cell lung cancer and lung fibrosis. Over 3000 scans acquired across multiple studies were used to train/validate a 3D U-net lung segmentation model and a Support Vector Machine (SVM) classifier to segment individual lung tumors. The U-net lung segmentation algorithm can be used to estimate changes in soft tissue volume within lungs (primarily tumors and blood vessels), whereas the trained SVM is able to discriminate between tumors and blood vessels and identify individual tumors. The trained segmentation algorithms (1) significantly reduce time required for lung and tumor segmentation, (2) reduce bias and error associated with manual image segmentation, and (3) facilitate identification of individual lung tumors and objective assessment of changes in lung and individual tumor volumes under different experimental conditions.

Automated monitoring of respiratory rate as a novel humane endpoint: A refinement in mouse metastatic lung cancer models

In oncology research, while xenograft tumor models are easily visualized and humane endpoints can be clearly defined, metastatic tumor models are often based on more subjective clinical observations as endpoints. This study aimed at identifying objective non-invasive criteria for predicting imminent distress and mortality in metastatic lung tumor-bearing mice. BALB/c and C57BL/6 mice were inoculated with CT26 or B16F10 cells, respectively. The mice were housed in Vium smart cages to continuously monitor and stream respiratory rate and locomotion for up to 28 days until scheduled euthanasia or humane endpoint criteria were met. Body weight and body temperature were measured during the study. On days 11, 14, 17 and 28, lungs of subsets of animals were microCT imaged in vivo to assess lung metastasis progression and then euthanized for lung microscopic evaluations. Beginning at day 21, most tumor-bearing animals developed increased respiratory rates followed by decreased locomotion 1-2 days later, compared with the baseline values. Increases in respiratory rate did not correlate to surface tumor nodule counts or lung weight. Body weight measurement did not show significant changes from days 14-28 in either tumor-bearing or control animals. We propose that increases in respiratory rate (1.3-1.5 X) can be used to provide an objective benchmark to signal the need for increased clinical observations or euthanasia. Adoption of this novel humane endpoint criterion would allow investigators time to collect tissue samples prior to spontaneous morbidity or death and significantly reduce the distress of mice in the terminal stages of these metastatic lung tumor models.