Introduction
Neoadjuvant chemotherapy—treatment administered prior to surgical intervention—is a critical component of modern oncology. However, its efficacy varies significantly across patient populations. Many individuals undergo intensive regimens without achieving a pathologic complete response (pCR), exposing them to unnecessary toxicity and delaying alternative treatments. The primary obstacle to personalized care is the lack of reliable, biologically interpretable biomarkers that can predict response from standard-of-care imaging.
In this study, we introduce Quantitative Tumor-Associated Vasculature (QuanTAV), a novel class of radiomic features designed to characterize the morphology and organization of the peritumoral vascular network. Unlike traditional "black box" machine learning approaches, QuanTAV is grounded in the physiological reality that tumor-associated vessels are structurally chaotic, which hinders drug delivery and promotes resistance.
The QuanTAV Framework
The QuanTAV pipeline extracts mathematical descriptors of vessel architecture from 3D clinical imaging (CT or contrast-enhanced MRI). The process involves three primary stages:
- Vessel Segmentation and Skeletonization: The 3D structure of vessels within a 5cm radius of the tumor is isolated and reduced to 1D "centerlines" to identify individual branches.
- Morphological Feature Extraction: 61 features quantify the 3D shape of these vessels, focusing specifically on torsion (the degree of twisting across a branch) and curvature (local deviations from a straight path).
- Spatial Organization Analysis: 30 features utilize 2D projections and Hough transforms to measure the global "chaos" or heterogeneity of vessel orientations relative to the tumor mass.
These 91 features provide a comprehensive profile of the structural integrity of the tumor’s blood supply—a direct proxy for the efficiency of the vascular bed in delivering therapeutic agents.
Experimental Design and Datasets
The study evaluated QuanTAV across a diverse cohort of 558 patients representing two major cancer types and multiple treatment modalities:
- Breast Cancer (MRI): Two cohorts (n=371) receiving either anthracycline-based chemotherapy (BRCA-ACT) or targeted therapy plus chemotherapy (BRCA-TCHP).
- Non-Small Cell Lung Cancer (NSCLC, CT): Two cohorts (n=187) receiving either platinum-based chemotherapy (NSCLC-PLAT) or a "trimodality" regimen of chemoradiation followed by surgery (NSCLC-TRI).
Data was sourced from both private clinical institutions and public repositories like The Cancer Imaging Archive (TCIA).
Key Insights and Results
Predictive Performance
Across all four treatment strategies, QuanTAV classifiers successfully predicted treatment response in independent testing sets, achieving Area Under the Curve (AUC) values between 0.63 and 0.71. Crucially, the addition of QuanTAV features to models based solely on clinical variables (such as age and tumor size) resulted in a significant performance boost, increasing the AUC by 0.06 to 0.12.
Prognostic Value
Beyond immediate response, QuanTAV proved to be a robust prognostic indicator for long-term survival:
- In the BRCA-ACT cohort, the QuanTAV risk score significantly predicted recurrence-free survival (HR=1.25, p=0.002, C-index=0.66).
- In the NSCLC-PLAT cohort, high-risk groups identified by the model faced a significantly higher risk of disease progression (HR=2.29, p=0.034, C-index=0.62).
- In the NSCLC-TRI cohort, the risk score was similarly prognostic (HR=1.28, p=0.039, C-index=0.66).
Biological Interpretability and Robustness
A central finding of this work is the correlation between vascular "chaos" and poor outcomes. High vessel torsion and disorganized spatial arrangements were consistently associated with treatment failure. Biologically, these twisted vessels create high interstitial fluid pressure and sluggish blood flow, physically obstructing the penetration of chemotherapy into the tumor bed. Furthermore, the signal remained consistent across both CT and MRI, demonstrating that the underlying biological phenomenon of abnormal angiogenesis is robust to different imaging modalities. The models also remained stable when researchers introduced noise into the segmentations, proving the signature's resilience.
Limitations and Future Directions
While the results are promising, several steps remain before clinical implementation. The current workflow involves semi-automated segmentation, which may be too slow for high-volume clinical settings. Additionally, as the data was analyzed retrospectively, prospective clinical trials are necessary to validate that QuanTAV-guided decisions can improve patient outcomes in real-time. Future research will also investigate whether QuanTAV can identify patients specifically suited for anti-angiogenic therapies designed to "normalize" tumor vasculature.
Conclusion
QuanTAV represents a significant advancement in radiomics by providing a biologically grounded, modality-independent biomarker of treatment response. By quantifying the structural irregularities of tumor-associated vasculature, clinicians may eventually be able to better stratify patients, sparing those unlikely to respond from the side effects of ineffective chemotherapy while identifying those who require more aggressive or alternative interventions.
Summary updated