How Lymph Node Levels Are Mapped in Radiology

Lymph node levels in radiology are standardized anatomic zones that divide the body’s lymph node chains into numbered or named stations, giving radiologists, surgeons, and oncologists a shared language for pinpointing where disease has spread. In the neck alone there are at least seven major levels (I through VII, with several subdivisions), each defined by imaging landmarks such as specific muscles, bones, and blood vessels rather than by what a surgeon can feel during a physical exam. Similar station maps exist for the chest, abdomen, and pelvis, each tailored to the cancers most likely to drain to those regions.

How the Cervical Node Classification Came About

Surgeons have grouped neck lymph nodes into levels since the mid-twentieth century, relying on landmarks they could see and touch in the operating room. As CT and MRI became routine for treatment planning, a gap emerged: the surgical landmarks did not always translate neatly to what a radiologist saw on cross-sectional images. In the late 1990s, a team at Memorial Sloan Kettering proposed an imaging-based classification that matched the clinical systems of the American Joint Committee on Cancer (AJCC) and the American Academy of Otolaryngology-Head and Neck Surgery but used anatomic landmarks visible on CT and MRI to define each level’s borders.¹ The goal was to eliminate the confusion that arose when different readers placed the same node in different levels, and to let the radiologist and surgeon speak the same dialect when discussing staging.

That early framework was refined into formal consensus guidelines. In 2003, representatives from major cooperative groups in Europe and North America, including DAHANCA, EORTC, GORTEC, NCIC, and RTOG, published CT-based delineation rules for the node-negative neck, giving radiation oncologists precise boundaries for drawing treatment volumes on planning scans.² A 2013 update expanded the guidelines, added contributors from Asia-Pacific groups, and fine-tuned sublevel boundaries that had caused persistent disagreement among readers.³

The Seven Cervical Levels and What They Contain

Each cervical level occupies a distinct zone in the neck, defined by landmarks you can identify on an axial CT or MRI slice. The system runs roughly from the chin downward and from the midline outward.

Level Ia: A midline space between the front bellies of the two digastric muscles, containing the submental nodes. It is continuous across the midline, so there is no true left-right border.⁴
Level Ib: Lateral to Ia, sitting between the inner surface of the mandible and the digastric muscle, housing the submandibular nodes.⁵
Level II: The upper jugular nodes, clustered around the top third of the internal jugular vein and upper spinal accessory nerve. Level II is commonly subdivided into IIa (anterior to the vein) and IIb (posterior to it).⁶
Level III: The mid-jugular nodes, along the middle third of the internal jugular vein, roughly from the hyoid bone down to the cricoid cartilage.
Level IV: The lower jugular nodes, extending from the cricoid to the clavicle.
Level V: The posterior triangle group, subdivided into Va (along the spinal accessory nerve) and Vb (along the transverse cervical vessels near the clavicle).
Level VI: The central compartment, containing pretracheal, paratracheal, and prelaryngeal (Delphian) nodes between the carotid sheaths.
Level VII: The superior mediastinal nodes below the top of the manubrium, often relevant in thyroid and esophageal cancers.

Remembering every boundary is less important for a general reader than understanding why the system exists: when a report says “there is a pathologic node at level IIb on the right,” the surgeon, radiation oncologist, and medical oncologist all know exactly which cluster of tissue is involved and what primary tumor sites most often drain there. That single phrase replaces a paragraph of anatomic description.

Size Thresholds and What Makes a Node “Abnormal”

Size is the quickest screening criterion a radiologist uses, but it is far from perfect. The most widely cited rule of thumb for cervical nodes comes from work in the early 1990s: the minimum diameter measured on the axial plane (the short axis) is the most accurate single size criterion. A short-axis diameter of 10 mm was found to be the best cutoff, except in the subdigastric region (level II), where 11 mm performed better.⁷ In children, the thresholds are somewhat larger for level II: nodes with a short-axis diameter above 15 mm at level II and above 10 mm at all other cervical levels are uncommon in otherwise healthy children, suggesting those cutoffs as a reasonable upper limit of normal in pediatric imaging.⁸

Size alone misses plenty. A 7 mm node that has lost its normal fatty hilum, shows internal necrosis, or has irregular margins can be malignant, while a 12 mm node with a preserved hilum in a patient fighting a viral infection may be entirely reactive. This is why radiologists also look at shape (round is more suspicious than oval), internal architecture, and the relationship of the node to surrounding fat planes. Central necrosis within a node has been shown to have a pooled sensitivity of about 81 percent for detecting extranodal extension in head and neck cancers, while infiltration of the adjacent fat planes carries a pooled specificity of about 94 percent.⁹ In practice, a combination of features matters more than any single measurement.

Imaging Modalities for Node Assessment

Different imaging tools have different strengths when it comes to evaluating lymph node levels, and the choice often depends on the clinical question and the body region.

CT and PET/CT

Contrast-enhanced CT is the workhorse for initial staging of head and neck cancers because it is fast, widely available, and provides clear anatomic detail. Its main weakness is reliance on size and morphologic criteria; a normal-sized node harboring microscopic disease will look unremarkable. Adding metabolic information with FDG-PET/CT substantially improves detection. In one study of head and neck squamous cell carcinoma, PET/CT achieved 96 percent sensitivity and roughly 99 percent specificity at the nodal-level, and correctly identified nodal disease in two patients whose CT alone had been read as node-negative.¹⁰ Agreement with surgical pathology was stronger for PET/CT than CT alone.

More recently, radiolabeled FAPI PET/CT has been explored as an alternative tracer. Across a pooled analysis of eight studies in head and neck cancer patients, FAPI PET/CT showed a sensitivity of about 89 percent and a specificity of 93 percent for lymph node metastases. By comparison, FDG PET/CT had a similar sensitivity of 91 percent but much lower specificity of 50 percent in the same analysis, meaning FDG lit up many more benign nodes as false positives.¹¹ If those specificity numbers hold up in larger trials, FAPI tracers could reduce unnecessary biopsies and spare patients unneeded surgery.

Ultrasound and Ultrasound-Guided Biopsy

Ultrasound excels at depicting the internal architecture of superficial lymph nodes. Grey-scale imaging reveals the shape, hilum, and any internal necrosis or calcification, while power Doppler maps the vascular pattern, since malignant nodes tend to show disordered peripheral vascularity rather than a tidy central hilum vessel.¹² In a head-to-head comparison with CT for differentiating benign from malignant cervical nodes, sonography significantly outperformed CT, largely because it could better depict the internal architecture of each node.¹³

The real power of ultrasound, though, lies in pairing it with fine-needle aspiration. Ultrasound-guided FNA achieves a specificity of about 98 percent, because instead of relying purely on how a node looks, the radiologist can sample it in real time and send cells for cytologic analysis.¹⁴ The trade-off is that ultrasound cannot see deep nodes well, so it is most useful in the accessible neck and axilla rather than in the mediastinum or retroperitoneum.

MRI and Diffusion-Weighted Imaging

Conventional MRI offers excellent soft-tissue contrast and is particularly useful in areas where CT struggles, such as the skull base or in patients who cannot receive iodinated contrast. Diffusion-weighted MRI measures how freely water molecules move within tissue, which is restricted in densely packed tumor cells. This difference allows the technique to distinguish between malignant and benign or necrotic tissue without contrast injection.¹⁵ In the pelvis, combining diffusion-weighted imaging with iron-oxide nanoparticle contrast agents has shown promise for detecting metastases in normal-sized pelvic nodes in bladder and prostate cancer patients, catching disease that standard imaging would miss entirely.¹⁶

Beyond the Neck: Chest, Abdomen, and Pelvis

The concept of numbered node levels is not unique to the head and neck. In the chest, the International Association for the Study of Lung Cancer (IASLC) lymph node map is now the standard. It superseded all earlier mediastinal maps and is used alongside the TNM staging system for lung cancer.¹⁷ The IASLC map divides thoracic nodes into stations numbered 1 through 14, grouped into supraclavicular, upper mediastinal, aortic, subcarinal, lower mediastinal, and hilar-interlobar zones. The distinction between, say, station 4R (right lower paratracheal) and station 7 (subcarinal) determines whether a lung cancer is potentially operable or has crossed into a higher stage.

In the abdomen and pelvis, node stations are described by their relationship to nearby vessels and organs. Retroperitoneal nodes cluster around the aorta, inferior vena cava, and the space between them (interaortocaval). Gastric and hepatic nodes sit within the gastrohepatic and hepatoduodenal ligaments. Pancreaticoduodenal nodes lie between the duodenum and pancreas. Pelvic nodes are grouped along the common, external, and internal iliac vessels.¹⁸ A structured CT-based reporting template has been proposed to standardize how radiologists describe these stations, aiming to bring the same level of consistency to abdominal node reporting that the cervical level system brought to the neck.¹⁹

The axillary node system matters most in breast cancer. Axillary nodes are traditionally divided into three levels based on their relationship to the pectoralis minor muscle: level I lies lateral to it, level II behind it, and level III medial to it. Sentinel lymph node biopsy has reduced the need for full axillary dissection in many patients, but the level system still guides both surgical planning and radiology reporting when imaging the axilla.

Why Node Levels Matter for Treatment Planning

The practical reason radiologists spend so much effort categorizing nodes by level is that treatment decisions hinge on exactly which levels are involved. In head and neck cancers treated with radiation, the choice of which node levels to include in the radiation field is guided by the primary tumor site and the current nodal stage. Updated consensus guidelines published in 2019 provide level-by-level target volume recommendations for oral cavity, oropharynx, hypopharynx, larynx, nasopharynx, paranasal sinus, and nasal cavity cancers, as well as for carcinoma of unknown primary, all stratified by the eighth-edition UICC nodal staging system.²⁰ Getting the level assignment wrong could mean either undertreating (missing a node station that harbors microscopic disease) or overtreating (irradiating normal tissue unnecessarily).

In thyroid cancer, CT can play a complementary role to ultrasound for determining the extent of surgery, because CT’s strength at a per-level analysis can reveal disease in compartments that ultrasound may not reach.²¹ For salivary gland carcinomas, PET/CT combined with histologic grade has been shown to be useful for detecting cervical lymph node metastases and guiding the decision of whether and how extensively to dissect the neck.²²

Common Pitfalls in Node Level Interpretation

Reading lymph nodes on imaging is not as straightforward as measuring a diameter and checking a box. Several pitfalls trip up even experienced readers.

One is confusing a non-nodal mass for a lymph node. Nerve sheath tumors, for example, can mimic pathologic lymph nodes on both ultrasound and CT, appearing as solid, hypoechoic masses without a fatty hilum. Features that point toward a schwannoma rather than a node include its solitary nature, a fusiform shape aligned along a nerve course, and the absence of the multiplicity and bilaterality typical of pathologic lymphadenopathy.²³

The post-treatment neck presents its own headaches. After surgery, radiation, or both, the normal anatomic landmarks that define node levels can be distorted or obliterated. Scar tissue, edema, and radiation-induced inflammation can all mimic residual or recurrent disease on CT and MRI. Even on PET/CT, post-treatment inflammation can cause metabolic activity that looks worryingly similar to active cancer.²⁴ Interpreting studies of the irradiated neck is widely recognized as one of the more challenging tasks in head and neck radiology, because the tissue changes created by treatment generate a whole spectrum of findings that can mimic disease.²⁵

Another common source of error is assuming that “normal-sized” means “normal.” As noted earlier, metastases can live in small nodes. In one study of bladder and prostate cancer patients who underwent pelvic lymph node dissection, histopathology revealed metastases in about 3 percent of all nodes examined, and combined USPIO-enhanced and diffusion-weighted MRI caught 92 percent of them. The two that were missed were micrometastases under 1 mm in diameter, essentially invisible to any current imaging technique.²⁶ A follow-up study with a larger cohort confirmed the value of the technique but showed per-patient sensitivity ranging from 65 to 75 percent across readers, a reminder that even advanced methods leave room for improvement.²⁷

Artificial Intelligence and the Future of Node Detection

Identifying and measuring small lymph nodes on hundreds of CT slices is tedious, time-consuming work, and it is exactly the kind of task that deep learning algorithms are being trained to handle. A recent study developed an automated segmentation model trained on over 25,000 CT slices from 221 neck CT scans, focusing specifically on the hardest category: nodes between 5 and 10 mm. The algorithm achieved a Dice score of about 0.81, a measure of how well the computer-generated outline matches the expert’s manual outline, where 1.0 would be a perfect overlap.²⁸ That is a promising start, especially for small nodes that are easy for a human reader to overlook when scrolling through a scan quickly.

Where this technology could matter most is in radiation planning. Contouring every relevant lymph node level on a treatment-planning CT currently takes considerable time and is subject to inter-observer variability. If an algorithm could accurately pre-contour the node levels and flag suspicious nodes, the radiation oncologist could review and refine rather than build from scratch, potentially reducing both time and inconsistency. The algorithms are not yet reliable enough to replace a trained eye, but the trajectory is clear: the combination of standardized level definitions and machine learning is likely to make node staging faster and more reproducible in the coming years.