Radiomic Profiling of the Peritumoral Microenvironment

111 concepts80 papers1 proposal

In classical medical image analysis, a tumor is often treated as an isolated island. A radiologist or an automated segmentation algorithm draws a sharp, binary boundary around the visible mass, separating the "tumor" from the "healthy tissue." Quantitative features—such as shape, intensity, and texture—are then extracted exclusively from within this binary mask.

However, cancer biology does not respect these clean geometric boundaries. A tumor is an active ecosystem that constantly interacts with its surrounding tissue, known as the tumor microenvironment (TME). Crucial pathological processes, such as microscopic cellular infiltration, localized inflammation, extracellular matrix remodeling, and neoangiogenesis, occur primarily at the tumor-host interface. By discarding the tissue immediately outside the tumor boundary, traditional radiomics suffers from a blind spot.

Peritumoral radiomics fixes this by computationally expanding the region of interest (ROI) beyond the visible tumor core. By extracting high-dimensional mathematical features from the surrounding "buffer zones," this approach captures subvisual patterns of the microenvironment, significantly improving our ability to predict treatment response, genetic expressions, and patient survival.


Computational Strategies for Capturing the Interface

To extract features from the peritumoral space, researchers must define how far into the surrounding tissue the analysis should reach. Because different biological signals manifest at different distances from the tumor shell, several spatial segmentation strategies have emerged.

   [ Tumor Core ] ---> [ Hard Boundary ] ---> [ Peritumoral Buffer Zone ]
         |                                              |
   (Intratumoral)                                (2.5mm - 20mm Ring)
   Texture & Density                             Immune Cells & Edema

Fixed-Width Dilated Margins

The simplest method involves mathematically dilating the primary tumor volume outward by a set distance. For instance, Nathaniel Braman et al. demonstrated this in breast cancer imaging by capturing textural features within a 2.5–5mm radius outside the tumor boundary in Intratumoral and peritumoral radiomics for the pretreatment prediction of pathological complete response to neoadjuvant chemotherapy based on breast DCE-MRI. This narrow window is highly sensitive to localized stromal alterations and vascular density changes immediately adjacent to the malignant mass.

Concentric Annular Rings

Rather than looking at a single expanded zone, advanced models extract features from multiple sequential rings extending outward. In a multicenter lung cancer study, Shaowei Wu et al. utilized concentric annular rings extending up to 20mm into the surrounding lung parenchyma in Pretreatment radiomic biomarker for immunotherapy responder prediction in stage IB–IV NSCLC (LCDigital-IO Study): a multicenter retrospective study. Similarly, Marwa Ismail introduced the R-DepTH descriptor in Radiomic Deformation and Textural Heterogeneity (R-DepTH) Descriptor to Characterize Tumor Field Effect: Application to Survival Prediction in Glioblastoma, which samples texture and physical tissue deformation across a series of concentric rings to quantify the long-range "field effect" of glioblastomas on healthy brain tissue.

Fuzzy Radiomics

A major limitation of standard dilation is its reliance on a hard, human-defined threshold for where the peritumoral zone ends. To overcome this, Marko Grahovac et al. proposed the use of "fuzzy" radiomics in Machine learning predictive performance evaluation of conventional and fuzzy radiomics in clinical cancer imaging cohorts. Instead of binary masks, this technique applies weighted probability masks to the tumor edges. By accounting for the mathematical "gray areas" where tumor cells gradually thin out into normal tissue, fuzzy radiomics reduces data redundancy and improves model stability, particularly for small or poorly defined lesions.


Peritumoral feature extraction works because mathematical texture descriptors—such as entropy, gray-level non-uniformity, and local intensity gradients—correlate directly with microscopic biological phenomena that are invisible to the naked eye on standard CT, MRI, or PET scans.

Mapping the Immune Landscape

When the immune system attempts to fight a tumor, immune cells aggregate at the invasive margin. Traditional imaging cannot visualize individual lymphocytes or macrophages, but their presence alters the local tissue density and heterogeneity.

Weicai Huang et al. leveraged this relationship in Noninvasive imaging of the tumor immune microenvironment correlates with response to immunotherapy in gastric cancer, demonstrating that quantitative features extracted from a narrow peripheral ring around gastric tumors could accurately map the local neutrophil-to-lymphocyte ratio and predict a patient's response to anti-PD-1 immunotherapy. The authors noted that the peritumoral region often contained more significant indicators of immune cell infiltration than the tumor core itself, as the core is frequently necrotic or immunosuppressed.

Detecting Microscopic Infiltration and Edema

In highly aggressive malignancies like glioblastoma, tumor cells migrate along blood vessels and white matter tracts far beyond the visible contrast-enhancing mass. This infiltrative front is typically accompanied by vasogenic edema.

In Radiomics Analysis Based on Magnetic Resonance Imaging for Preoperative Overall Survival Prediction in Isocitrate Dehydrogenase Wild-Type Glioblastoma, Shouchao Wang et al. showed that analyzing the texture of this peritumoral edema region reveals subtle patterns of tumor invasion and heterogeneity that escape standard visual inspection. Capturing these subvisual signatures allows models to differentiate between benign swelling and active, aggressive infiltration, providing critical prognostic information regarding patient survival.

Quantifying Boundary Gradients

The mathematical contrast between the inside and outside of a lesion provides a direct proxy for tumor invasiveness. For example, Alexandre Bône et al. designed a deep learning framework in Guiding the classification of hepatocellular carcinoma on 3D CT-scans using deep and handcrafted radiological features that explicitly calculates energy differences and median intensity variations across the lesion border. A sharp, distinct boundary often signals a well-encapsulated, less aggressive tumor, whereas a chaotic, high-entropy gradient across the boundary indicates an active, infiltrative front.


Building Predictive Synergy

The true power of peritumoral radiomics emerges when it is combined with intratumoral data, creating a multi-regional profile of the disease.

By treating the tumor core and its immediate microenvironment as a unified, multi-layered spatial unit, machine learning models can capture both the internal cellular architecture of the malignancy and its external interaction with the host. Whether predicting the malignancy of lung nodules by analyzing the surrounding lung parenchyma (Yuhao Zhou), or assessing prostate cancer aggressiveness by capturing peritumoral signs of hypoxia and hypervascularity (Chao Zhang et al.), looking outside the box has proven to be a necessity for accurate digital oncology.

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 78 papers
  2. Version 1 · 2026-10-04 · 75 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_peritumoral_feature_extraction_460af016,
  title = {peritumoral feature extraction},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/peritumoral-feature-extraction/rdr_460af016031842be91942599d22f4159},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 2, frozen text):

@misc{lacuna_peritumoral_feature_extraction_460af016_v2,
  title = {peritumoral feature extraction},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/peritumoral-feature-extraction/rdr_460af016031842be91942599d22f4159/v/2},
  note = {Version 2, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake