Recent advances in the field of radiomics—the comprehensive extraction of quantitative features from medical images—offer promising avenues to overcome these challenges. By harnessing high-throughput data derived from computed tomography (CT) scans, radiomics can reveal subtle imaging biomarkers that are often imperceptible to the human eye. These biomarkers, when combined with clinical indicators, may enable more precise and personalized predictions regarding LVI status in patients with invasive LUAD.

A pioneering study published in BMC Cancer has explored the integration of intratumoral and peritumoral CT radiomics features to develop predictive models for LVI in LUAD patients. The investigators analyzed CT images from a cohort of over 600 patients across two institutions, extracting an extensive array of more than 1,200 quantitative radiomic features from distinct tumor regions. This comprehensive approach allowed for a detailed morphological and textural characterization of both the tumor bulk and its surrounding microenvironment, which is critically implicated in tumor invasion dynamics.

The research team divided their patient population into training, internal, and external validation cohorts, enabling robust assessment of the model’s generalizability across diverse clinical settings. Utilizing advanced machine learning techniques, they constructed three distinct radiomics models: one focusing on the gross tumor alone, a second encompassing both gross tumor and peritumoral regions, and a third analyzing the peritumoral area in isolation. These models were evaluated based on their ability to discriminate LVI presence, measured through the area under the receiver operating characteristic curve (AUC).

Among the three approaches, the model incorporating both intratumoral and peritumoral features demonstrated superior predictive performance. This combined gross tumor and peritumoral (GPT) model revealed AUC values of 0.83 in the training set and maintained robust prediction capabilities with AUCs of 0.79 and 0.75 in internal and external validation sets, respectively. These findings underscore the clinical value of assessing not only the tumor itself but also its interface with the surrounding tissue, a region known to harbor critical biological interactions facilitating vascular and lymphatic spread.

In parallel, the study also identified key clinical parameters independently associated with LVI through rigorous statistical analysis. The preoperative carcinoembryonic antigen (CEA) level, tumor diameter, and the presence of spiculation on CT scans emerged as significant predictors. Incorporating these clinical indicators alongside the radiomic signature resulted in a composite predictive model with further enhanced accuracy. The integrated model yielded AUCs of 0.84, 0.82, and 0.77 across the training, internal, and external cohorts, respectively, outperforming models based solely on imaging or clinical data.

This multifaceted approach highlighting the synergy between image-derived radiomic features and conventional clinical factors represents a substantial step forward in the non-invasive preoperative assessment of LUAD. From a clinical perspective, the ability to predict LVI status before surgical intervention could enable thoracic oncologists to stratify patients according to risk, personalize therapeutic regimens, and potentially improve survival outcomes by identifying those who may benefit from more aggressive treatments or closer postoperative surveillance.

The methodology employed in this study involved comprehensive feature extraction from high-resolution CT images, capturing a spectrum of matrix-based texture descriptors and wavelet transformations, which provide deep insights into tumor heterogeneity. Radiomic features related to shape, intensity, and texture likely reflect the complex biological processes underpinning tumor growth and vascular invasion, offering a quantitative surrogate marker unattainable through standard radiological interpretation.

Furthermore, the inclusion of peritumoral radiomics is especially notable, as the tumor microenvironment plays a pivotal role in facilitating cancer progression and metastasis. By extending analysis beyond the tumor boundaries, the researchers tapped into spatial patterns of tissue alterations adjacent to the tumor that may signal early invasion of lymphovascular structures. These pioneering insights highlight the necessity of looking beyond conventional tumor metrics to fully characterize malignant potential.

The clinical applicability of such predictive models holds profound implications for advancing precision medicine in lung cancer care. As lung adenocarcinoma comprises a heterogeneous group of tumors with variable behavior, preoperative LVI prediction via non-invasive imaging biomarkers could inform decisions surrounding surgical resection margins, lymph node dissection extent, and the necessity for neoadjuvant therapies. This stratification may ultimately reduce overtreatment and associated morbidities while ensuring optimal oncologic control for high-risk patients.

Moreover, the study sets a precedent for the integration of big data analytics, artificial intelligence, and clinical oncology, showcasing a translational framework whereby computational tools augment physician capabilities. Radiomics, when validated in large multicenter cohorts as exemplified in this investigation, can become an indispensable component of the oncologic diagnostic arsenal, fostering more nuanced risk assessments and guiding tailored interventions.

Despite the encouraging results, certain challenges remain for the widespread clinical implementation of radiomics models. Standardization of imaging protocols, reproducibility of feature extraction algorithms, and prospective validation in randomized clinical trials are necessary to cement the role of radiomics as a standard diagnostic tool. Additionally, interdisciplinary collaboration among radiologists, oncologists, bioinformaticians, and machine learning experts will be critical to overcome technical and methodological hurdles.

Looking ahead, the integration of radiomics with emerging molecular and genomic biomarkers could further enhance prediction accuracy and provide a holistic view of tumor biology. Combining imaging phenotypes with genetic profiles may unravel novel mechanisms underlying lymphovascular invasion and identify new therapeutic targets. This multimodal approach embodies the future of oncology, leveraging the convergence of data science and molecular medicine.

In conclusion, this innovative study provides compelling evidence that CT radiomics models incorporating intratumoral and peritumoral features, combined with key clinical parameters, offer a powerful non-invasive method for predicting lymphovascular invasion in invasive lung adenocarcinoma. By facilitating early identification of patients at higher risk for poor prognosis, this approach promises to refine risk stratification, tailor treatment strategies, and ultimately improve clinical outcomes. The findings underscore the transformative potential of radiomics in lung cancer management and highlight the importance of ongoing research bridging advanced imaging analytics with pragmatic clinical applications.

Subject of Research: Non-invasive prediction of lymphovascular invasion in invasive lung adenocarcinoma using intratumoral and peritumoral CT radiomics combined with clinical indicators

Article Title: The clinical value of predicting lymphovascular invasion in patients with invasive lung adenocarcinoma based on the intratumoral and peritumoral CT radiomics models

Article References:
Lin, M., Zhao, C., Huang, H. et al. The clinical value of predicting lymphovascular invasion in patients with invasive lung adenocarcinoma based on the intratumoral and peritumoral CT radiomics models. BMC Cancer 25, 1752 (2025). https://doi.org/10.1186/s12885-025-15128-2

Image Credits: Scienmag.com

DOI: 10.1186/s12885-025-15128-2 (Published 12 November 2025)

The clinical challenge addressed by this research lies in the unpredictable nature of high-risk breast lesions. While some of these lesions remain benign, others upgrade to malignancy upon surgical excision, creating a critical need for improved non-invasive diagnostic tools to stratify patient risk effectively. Traditional imaging and biopsy methods have exhibited limitations in accurately forecasting these pathological upgrades, leading to unnecessary surgeries or delayed treatment. The innovative use of radiomics — extracting vast quantitative data from medical images — provides a promising avenue to overcome these constraints.

This study leveraged dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI), a sophisticated imaging modality that highlights blood flow differences within breast tissue, to derive both clinicoradiological characteristics and complex radiomic features. By delineating both intratumoral and peritumoral regions of interest (ROIs), the team hypothesized that combining data from tumor cores and their immediate microenvironment could capture crucial biological interactions indicative of malignancy risk.

Data from 174 patients with biopsy-confirmed high-risk breast lesions were retrospectively analyzed. These patients underwent preoperative MRI scans between 2019 and 2024 at Shenzhen People’s Hospital. To ensure robust model development and validation, the dataset was split into training and test sets at a 7:3 ratio. The high granularity enabled the researchers to build multiple radiomic models focusing on various spatial regions — the lesion itself and peritumoral areas extended outward by 3 mm, 5 mm, and 7 mm respectively.

Particularly striking was the performance of the peritumoral 3 mm radiomics model, which outperformed broader surrounding regions. This suggests that the immediate peritumoral microenvironment harbors critical imaging features that correlate with risk of morphological upgrade. These findings highlight the importance of not only looking within the tumor boundaries but also closely analyzing its proximal tissue milieu, which may reflect early infiltrative or reactive processes preceding malignant transformation.

Beyond isolated radiomic signatures, the study further integrated clinical and conventional imaging features, combining them with both intratumoral and peritumoral radiomics to construct a comprehensive predictive model. This dual-region combined model demonstrated exceptional diagnostic performance, achieving an AUC (area under the curve) of 0.883 in the training cohort and 0.851 in the independent test set. These metrics significantly surpassed the predictive power of models based solely on clinical data or individual radiomic features.

Diagnostic sensitivity, specificity, and accuracy of the combined model were also impressive. In the training group, these metrics were 79.4%, 82.7%, and 81.8% respectively, while the test cohort exhibited 72.7% sensitivity, 85.7% specificity, and 83.0% accuracy. Such balanced performance underscores the model’s potential utility in real-world clinical settings where minimizing both false positives and negatives is paramount to patient outcomes and healthcare resource optimization.

The researchers employed rigorous univariate and multivariate logistic regression analyses to identify independent risk factors for pathological upgrade. These statistical approaches ensured the integration of only the most relevant radiomic and clinical features into the final model, mitigating overfitting and enhancing generalizability. The result is a nuanced risk stratification tool anchored in biologically meaningful data representation.

This research also culminated in the design of a clinically applicable nomogram — a graphical calculation tool that synthesizes multiple predictive factors into an individualized risk score. Such a nomogram can provide oncologists and radiologists with an intuitive interface to estimate upgrade probabilities, guiding personalized treatment decisions, such as whether to proceed with surgical excision or adopt a watchful waiting strategy.

Importantly, the study’s retrospective multicenter framework and relatively large sample bolster confidence in the findings, though prospective validation across diverse populations remains essential before widespread clinical adoption. The methodology, based on automated ROI delineation and multi-scale radiomic feature extraction, lays a replicable foundation for future investigations into other cancer types and lesion-risk assessments.

From a technological standpoint, the dual-region radiomic approach breaks new ground by recognizing the peritumoral environment as a critical player in oncogenesis and tumor progression. This paradigm shift broadens the imaging biomarker landscape and reflects trends in tumor microenvironment research, which increasingly reveal how surrounding stromal and immune components influence cancer behavior.

Given the rapid evolution of artificial intelligence and machine learning algorithms in medical imaging, this study exemplifies how advanced computational analytics can enable precision oncology. By harnessing subtle imaging textures, shape descriptors, and signal intensity variations imperceptible to the human eye, radiomics enhances diagnostic accuracy and unlocks new insights into tumor biology.

The clinical implications are profound. Accurate preoperative risk assessment helps avoid overtreatment in patients with benign high-risk lesions and conversely ensures timely intervention for those on the verge of malignant transformation. Moreover, this approach can reduce patient anxiety, limit unnecessary invasive procedures, and optimize healthcare resource allocation.

In a broader context, such advances contribute to the shifting landscape from “one-size-fits-all” cancer care to individualized management protocols based on precise phenotypic information. Incorporating quantitative radiomic signatures with clinical parameters exemplifies the future of multi-omic integration, potentially paving the way for more personalized, data-driven diagnostic and therapeutic pathways.

As the cancer research community continues to unravel the complexity of tumor heterogeneity and its clinical ramifications, studies like this underscore