Quantitative vessel tortuosity: A potential CT imaging biomarker for distinguishing lung granulomas from adenocarcinomas

Mehdi Alilou, Mahdi Orooji, Niha Beig, Prateek Prasanna, Prabhakar Rajiah, Christopher Donatelli, Vamsidhar Velcheti, Sagar Rakshit, Michael Yang, Frank J. Jacono, Robert Gilkeson, Philip A. Linden, Anant MadabhushiScientific 2018 · DOI

Abstract

Adenocarcinomas and active granulomas often present similarly on computed tomography (CT) and positron emission tomography (PET) scans, frequently leading to unnecessary invasive biopsies for benign cases. This study introduces quantitative vessel tortuosity (QVT), a novel CT imaging biomarker, to differentiate these conditions. Using a cohort of 290 patients across two institutions, a machine learning classifier based on QVT features achieved an area under the receiver operating characteristic curve (AUC) of 0.85 in independent validation. This significantly outperformed human experts (AUC ~0.61), traditional radiomic features (AUC 0.73), and deep learning models (AUC 0.76), suggesting QVT's potential as a non-invasive diagnostic tool.

Introduction

Distinguishing between malignant lung adenocarcinomas and benign inflammatory granulomas is a significant clinical challenge. Both can appear "spiculated" on standard CT scans and show high metabolic activity on PET scans. Due to these similarities, approximately 30% of patients with benign nodules undergo invasive surgical procedures to rule out cancer, incurring high costs and risks. This research explores the surrounding vasculature's architecture—specifically its "twistedness"—as a discriminative "radiomic" signature.

Methods

The study developed a comprehensive pipeline to quantify the tortuosity of vessels associated with lung nodules.

Overview of the QVT analysis pipeline including CT imaging, vasculature segmentation, feature extraction, and nodule characterization.

The process involves four primary stages:

  1. Segmentation: Radiologists identify a 3D volume of interest, and a 3D region-growing algorithm isolates the nodule and its associated vasculature.
  2. Skeletonization: Centerlines of the segmented vessels are extracted using a "fast marching" algorithm.
  3. Feature Extraction: 35 QVT features are calculated based on tortuosity (vessel length vs. straight-line distance), curvature (bending degree), and branching complexity.
  4. Machine Learning: A Support Vector Machine (SVM) classifier is trained using the most stable and informative features selected via minimum Redundancy Maximum Relevance (mRMR).

Datasets

The primary cohort included 290 patients, split into a Training Set (N=145) and an Independent Validation Set (N=145). Additionally, the public RIDER dataset (N=31) was used to assess feature stability across test-retest scans.

Results

The QVT-based classifier demonstrated superior diagnostic performance compared to traditional methods and human interpretation.

Visualization of the SVM classifier training using top QVT features on the primary cohort.

Key findings include:

  • High Accuracy: The QVT model achieved an AUC of 0.85 on the independent validation set.
  • Human vs. Machine: Human experts (a radiologist and a pulmonologist) achieved AUCs of only 0.61 and 0.60, respectively, on the same data.
  • Comparison with AI Models: QVT outperformed standard radiomics (AUC 0.73) and deep learning architectures like LeNet (AUC 0.76) and VGG-19.
  • Robustness: Feature stability analysis using the RIDER dataset identified a subset of QVT features, such as maximum curvature, that remain highly reproducible across different scan conditions.

Discussion

The success of QVT is rooted in the biology of angiogenesis. Malignant tumors rapidly recruit new blood vessels, which grow haphazardly and result in more tortuous shapes than those found near benign granulomas. While most radiomic studies focus on the nodule's internal texture, these results suggest that the surrounding environment provides more discriminative information.

The poor performance of human experts highlights that vessel tortuosity is a "sub-visual" feature—quantifiable by AI but difficult for the human eye to assess accurately. However, the method's reliance on high-quality 3D vessel segmentation remains a limitation, as errors in isolating small vessels can lead to unstable feature values. Future work should expand the scope to other benign mimics and validate these findings in prospective clinical trials.

Conclusion

Quantitative vessel tortuosity represents a promising, non-invasive imaging biomarker. By accurately distinguishing adenocarcinomas from granulomas on routine non-contrast CT scans, QVT could significantly reduce unnecessary invasive procedures and improve patient management in lung cancer screening.

Summary updated

Abstract

Abstract Adenocarcinomas and active granulomas can both have a spiculated appearance on computed tomography (CT) and both are often fluorodeoxyglucose (FDG) avid on positron emission tomography (PET) scan, making them difficult to distinguish. Consequently, patients with benign granulomas are often subjected to invasive surgical biopsies or resections. In this study, quantitative vessel tortuosity (QVT), a novel CT imaging biomarker to distinguish between benign granulomas and adenocarcinomas on routine non-contrast lung CT scans is introduced. Our study comprised of CT scans of 290 patients from two different institutions, one cohort for training (N = 145) and the other (N = 145) for independent validation. In conjunction with a machine learning classifier, the top informative and stable QVT features yielded an area under receiver operating characteristic curve (ROC AUC) of 0.85 in the independent validation set. On the same cohort, the corresponding AUCs for two human experts including a radiologist and a pulmonologist were found to be 0.61 and 0.60, respectively. QVT features also outperformed well known shape and textural radiomic features which had a maximum AUC of 0.73 (p-value = 0.002), as well as features learned using a convolutional neural network AUC = 0.76 (p-value = 0.028). Our results suggest that QVT features could potentially serve as a non-invasive imaging biomarker to distinguish granulomas from adenocarcinomas on non-contrast CT scans.
BibTeX
@inproceedings{alilou2018quantitative,
  author = {Mehdi Alilou and Mahdi Orooji and Niha Beig and Prateek Prasanna and Prabhakar Rajiah and Christopher Donatelli and Vamsidhar Velcheti and Sagar Rakshit and Michael Yang and Frank J. Jacono and Robert Gilkeson and Philip A. Linden and Anant Madabhushi},
  title = {{Quantitative vessel tortuosity: A potential CT imaging biomarker for distinguishing lung granulomas from adenocarcinomas}},
  booktitle = {Scientific},
  year = {2018},
  doi = {10.1038/s41598-018-33473-0},
}
Figure