The Diagnostic Value of PET/CT and MRI in Head and Neck Cancer of Unknown Primary

ABSTRACT

Objective

Imaging plays a crucial role in the work‐up of patients with head and neck cancer of unknown primary (HNCUP). The aim of the present study was to assess the diagnostic value of PET/CT and MRI in patients with clinical HNCUP.

Methods

Consecutive patients with clinical HNCUP admitted to Aarhus University Hospital, Denmark, for diagnostic work‐up between 2012 and 2017 were included and followed for 5 years. Experts in nuclear medicine and radiology conducted blinded evaluations of PET/CT and MRI scans which were compared to histopathology obtained during primary work‐up or subsequent assessment in cases of an emerging primary. A subgroup analysis of patients with clinical HNCUP with lymph node squamous cell carcinoma (HNSCCUP) was performed.

Results

A total of 295 clinical HNCUP patients were included. Histopathological diagnoses were squamous cell carcinoma (60%), adenocarcinoma (19%), lymphoma (13%), other cancers (6%), and benign (2%). Detection rates for PET/CT and MRI in clinical HNCUP were 43% and 36% with a sensitivity of 74% and 66%, p = 0.04. Similarly for clinical HNSCCUP, detection rates were 52% and 43% with a sensitivity of 84% and 75%, p = 0.1. Combined PET/CT and MRI successfully identified a primary tumor in 51% of clinical HNCUP patients and 58% of clinical HNSCCUP patients.

Conclusion

Our findings support the complementary roles of PET/CT and MRI in the diagnostic work‐up of clinical HNCUP. PET/CT tended to detect more primary tumors and may be preferred as the initial imaging modality when only one can be performed, while MRI not only adds value for staging and surgical planning but also helps identify additional primaries, enhancing overall diagnostic precision.

Evidence Level

2b.

PET/CT demonstrates a higher diagnostic yield than MRI in detecting primary tumors in patients undergoing diagnostic work‐up for head and neck cancer of unknown primary origin. However, combining PET/CT and MRI improves overall detection rates and enhances soft tissue evaluation.

Abbreviations

DAHANCA

Danish Head and Neck Cancer Group

HNCUP

head and neck cancer of unknown primary

HNSCC

head and neck squamous cell carcinoma

HNSCCUP

head and neck squamous cell carcinoma of unknown primary

MDT

multi‐disciplinary team

MRI

magnetic resonance imaging

NPV

negative predictive value

PET/CT

18‐fluorodeoxyglucose positron emission tomography–computed tomography

PPV

positive predictive value

SCC

squamous cell carcinoma

1. Introduction

Head and neck cancer of unknown primary (HNCUP) presents a significant diagnostic challenge. The diagnosis of clinical HNCUP applies to patients presenting with a suspicious lymph node in the neck with no obvious primary tumor following a physical examination and corresponds to what is often referred to as tentative HNCUP in early diagnostic settings. The most common histological type of clinical HNCUP is squamous cell carcinoma (SCC) [1], and the most frequent location of the primary tumor is the head and neck region [2]. In Denmark, patients with clinical HNCUP are evaluated according to national guidelines established by the Danish Head and Neck Cancer Group (DAHANCA) and Danish Society for Head and Neck Oncology [3]. The diagnostic program includes a thorough examination of the head and neck region, including ultrasound‐guided fine needle aspiration cytology and diagnostic imaging [3, 4, 5]. In this context, full‐body 18‐fluorodeoxyglucose positron emission tomography–computed tomography (PET/CT) is mandatory in Denmark. However, the diagnostic value of applied imaging modalities in clinical HNCUP remains a matter of debate. Although PET/CT is often presented as superior to magnetic resonance imaging (MRI), most studies examining the application of these imaging modalities in the diagnostic work‐up of clinical HNCUP have limitations, that is, exclusively including patients with SCC or small sample size [6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16]. In Denmark, all patients suspected of cancer are entitled to a full diagnostic work‐up within 15 days of referral as mandated by the national cancer patient pathways. As part of this accelerated process, both PET/CT and MRI are typically requested early in the course of evaluation, often at the patients first hospital visit when HNCUP is suspected. This study therefore reflects “real‐world data” generated under these standardized clinical conditions [17].

Therefore, we aimed to compare the diagnostic effectiveness of full‐body PET/CT to head‐and‐neck MRI in clinical HNCUP patients, hypothesizing that PET/CT is superior to MRI in terms of identifying the primary tumor. Analyses included subgroup assessment of HNCUP with lymph node squamous cell carcinoma (HNSCCUP) as well as overall positive and negative predictive values (NPV) for primary tumor identification.

2. Materials and Methods

2.1. Study Design and Population

This retrospective study included consecutive patients undergoing diagnostic evaluation for clinical HNCUP at Aarhus University Hospital, Denmark, a tertiary care center. Patients with previous head and neck cancer were excluded from the study. Prior to inclusion, all patients underwent a comprehensive clinical assessment, which involved a detailed ear, nose, and throat (ENT) evaluation, including flexible video pharyngo‐laryngoscopy and neck ultrasound with ultrasound‐guided fine‐needle aspiration, none of which revealed a primary tumor. In accordance with Danish national guidelines [3] additional diagnostic procedures were conducted: PET/CT, endoscopic evaluation under general anesthesia with biopsies of any suspicious sites, bilateral palatine tonsillectomy, and random biopsies from base of tongue or total base of tongue mucosectomy. As an extension of the guidelines, MRI was performed concurrently with PET/CT in all patients, before any interpretation of either modality was undertaken. The inclusion period extended from 01.01.2012 to 31.12.2017. All patients were monitored for 5 years following treatment to identify any emerging primary tumors, resulting in a study period spanning from 2012 to 2022.

Blinded evaluation of the PET/CT and MRI scans was conducted by experts in nuclear medicine and radiology. The diagnostic reference standard was histopathology, which ultimately reflected the diagnosis given by the multidisciplinary team conference decision (MDT) following the diagnostic work‐up. Primary tumors emerging within 5 years after inclusion were registered. The criteria for defining an emerging primary, as opposed to a new primary, were recurrence within 5 years and histopathological concordance in tumor subtype (e.g., SCC, adenocarcinoma) and, where available, p16 immunostaining status. Relevant data were retrieved from the patients' electronic medical records and final diagnoses of head and neck cancer were crosschecked with the DAHANCA database.

Tumors were staged using the 9th edition of the TNM classification of the AJCC.

2.2. PET/CT

PET/CT imaging was conducted following standard protocols on fasting patients (6 h fast), approximately 60 min after the administration of a dose of ¹⁸F‐FDG. Imaging was performed using the Siemens Biograph TruePoint 40 (Siemens Healthcare GMbH, Erlangen, Germany) or the Discovery PET/CT 690 (GE Healthcare, Waukesha, WI, USA). The standard scanning protocol underwent minor modifications over the 5‐year period and comprised the following components: (1) a low‐dose CT for attenuation correction (50 Ref mAs, 120 kV) spanning from the vertex of the skull to the mid‐femur in a craniocaudal direction, and (2) a WB PET scan with a 3 min per bed position acquisition time and (3) a contrast‐enhanced arterial CT scan (100 Ref mAs, 120 kV) of the neck, thorax, and abdomen and from 2014 an additional venous phase liver scan (Ref mAs 175, 120 kV).

Following the scan, the images were reconstructed using the TrueX algorithm from Siemens (21 subsets and 3 iterations) in a matrix of 168 × 168 and post‐filtered with a 3.0 mm FWHM Gaussian filter.

¹⁸F‐FDG PET/CT images were inspected using Hermes Gold Client (Hermes Medical Solution AB, Stockholm, Sweden). PET scan and low‐dose CT combined were utilized for re‐evaluation. The maximum standardized uptake value within the region of interest was assessed using a volume‐of‐interest tool. PET/CT was performed as a full‐body scan.

2.3. MRI

Patients were scanned using either a 1.5 T MRI system (Siemens Avanto, Erlangen, Germany) or a 3 T MRI system (Siemens Skyra, Erlangen, Germany or a Phillips Medical System Archiva, Best, The Netherlands) with a standard head and neck coil. MRI protocols included 4 mm T2 STIR/Dixon and T1 FS/Dixon with and without contrast in the coronal plane, and T2 and T1 FS/Dixon with contrast in the axial plane. Only the head and neck area was included in MRI scans.

2.4. Blinded Evaluation of PET/CT and MRI Scans

Diagnostic PET/CT and MRI scans underwent blinded evaluation by two experienced nuclear medicine consultants (PET/CT) and one experienced radiologist (MRI). No clinical context was provided, except that scans were performed as part of a clinical HNCUP assessment. For both PET/CT and MRI, the identification of a primary tumor was registered using a dichotomous scale, yes or no, along with location and relation to anatomical structures and topography, and affected cervical lymph nodes.

In addition, PET/CT‐scans were evaluated based on a grading scale for ¹⁸F‐FDG‐uptake. A site with no or physiological ¹⁸F‐FDG‐uptake was Grade I, while an ¹⁸F‐FDG‐uptake greater than that of the blood pool was Grade II, an ¹⁸F‐FDG‐uptake greater than that of the liver was Grade III, and an ¹⁸F‐FDG‐uptake twice that of the liver was Grade IV.

Based on ¹⁸F‐FDG uptake, pattern, and location, sites were categorized by the nuclear medicine consultants as being most probably benign, most probably malignant, or equivocal.

2.5. Statistical Analysis

Findings were categorized as true positive if the imaging read positive for the same site as per final diagnosis of the pathology report. Likewise, if the final diagnosis was true HNCUP (i.e., no primary tumor detected), lymphoma, or benign etiology, findings were categorized as true negative if imaging read negative (no suspicious primary tumor on imaging). If the primary tumor was found outside the confinements of the MRI scan (i.e., lung, intestines, etc.) a negative read on the MRI was categorized as true negative. A positive image read in patients in whom a primary tumor was not clinically confirmed was categorized as false positive. Likewise, if an image read positive in one location, but the primary tumor was located in another location, this was categorized as false positive. Absence of suspicion in an anatomical site where a primary tumor was detected based on histopathology was categorized as false negative.

Sensitivity and specificity were compared using the McNemar test. Detection rates, defined as numbers identified per number scanned, were compared using Z‐tests. Positive predictive values (PPV) and NPV were compared by their 95% confidence interval. A p value < 0.05 was considered statistically significant.

The study was approved by The Scientific Ethics Committees for the Central Jutland Region, case no. 1‐10‐72‐225‐22.

3. Results

3.1. Patient Demographics

A total of 295 consecutive clinical HNCUP patients were included in the study. On referral, the majority of patients had a fine needle aspiration cytology with confirmed or suspected carcinoma in one or more lymph nodes in the neck.

In total, following diagnostic work‐up according to national guidelines [3], a primary tumor was identified in 187 (63%) of the patients. In the HNSCCUP subgroup, a primary tumor was identified in 119 of 178 patients (67%). Of these, 89 (75%) were p16‐positive, and the majority presented with nodal disease involving levels II and III (n = 111, 93%).

Among the patients with clinical HNSCCUP where a primary tumor was not identified (n = 59), four presented with an emerging primary during the five‐year follow‐up period. None of the emerging primary sites were labeled as suspicious on initial PET/CT and MRI. Final histology is shown in Table 1.

TABLE 1.

Characteristics of patients with tentative diagnosis HNCUP.

Variable	n = 295
Age a	62.7 (12.2)
Female sex	97 (33%)
Smoking status
Never smoked	96 (33%)
Previously smoked	107 (36%)
Currently smoking	92 (31%)
Verified histology
SCC
Mucosal SCC	167 (57%)
Cutaneous SCC	11 (4%)
Adenocarcinoma	55 (19%)
Lymphoma	38 (13%)
Neuroendocrine tumors	6 (2%)
Benign	7 (2%)
SCLC	2 (1%)
Poorly differentiated carcinoma	2 (1%)
Adenoid cystic carcinoma	2 (1%)
Sarcoma	2 (1%)
Malignant melanoma	2 (1%)
Anaplastic thyroid carcinoma	1 (< 1%)
p16 status b
Positive	118 (66%)
Negative	49 (28%)
Unknown	11 (6%)

^a Presented as mean years (SD).

^b Of SCC tumors, proven by cytology or histology.

3.2. Detection of Unknown Primary

Based on the entire population of clinical HNCUP population, PET/CT labeled a lesion as “suspicious” in 171 patients (58%). Of these, malignancy was confirmed in 126 patients (43%), resulting in a PPV of 74% and a sensitivity of 74%. Similarly, MRI identified a “suspicious” lesion in 160 patients, of which 106 (36%) were confirmed as malignant based on histopathology, equaling a PPV of 66% and a sensitivity of 66% (Figure 1A). Based on these findings, PET/CT appeared more sensitive than MRI, p = 0.04, although no obvious differences in PPV and NPV were found (Table 2).

FIGURE 1.

Number of primary tumors detected by PET/CT (left ellipse), MRI (right ellipse) and by both modalities (overlapping ellipse, middle). (A) For all tentative HNCUP patients, n = 295. (B) For tentative HNSCCUP patients, n = 178. (C) For all patients with a diagnosed primary tumor in the head and neck, n = 146.

TABLE 2.

Diagnostic metrics of PET/CT and MRI in tentative HNCUP (n = 295) and tentative HNSCCUP (n = 178) patients with 95% confidence interval.

Modality	Patient group	Sensitivity	Specificity	PPV	NPV
PET/CT	HNCUP	74% (67%–80%)	64% (56%–72%)	74% (67%–80%)	64% (56%–72%)
	HNSCCUP	84% (76%–89%)	50% (39%–61%)	70% (61%–77%)	69% (56%–79%)
MRI	HNCUP	66% (58%–73%)	59% (51%–67%)	66% (58%–73%)	59% (51%–67%)
	HNSCCUP	75% (66%–83%)	50% (40%–60%)	65% (56%–73%)	63% (51%–73%)

In 45 patients (15%), a primary tumor was identified by PET/CT but not by MRI, and in 25 patients (8%) a primary tumor was identified by MRI, but not by PET/CT. Combining MRI and PET/CT, the overall diagnostic imaging detection rate was 51%.

3.3. Subgroup Analysis of HNSCCUP Patients

A subgroup analysis of patients with clinical HNSCCUP (n = 178) revealed that PET/CT correctly identified a primary tumor in 92 patients (52%), whereas MRI identified a primary tumor in 76 patients (43%); the difference between modalities was not statistically significant (p = 0.9). The PPV of PET/CT and MRI was 70% and 65%, respectively, whereas the sensitivity was 84% for PET/CT and 75% for MRI, p = 0.1, although no obvious differences in PPV and NPV were found (Table 2). When stratified by HPV status, PET/CT demonstrated significantly higher sensitivity in the HPV‐positive group compared to the HPV‐negative group (89% vs. 67%, p = 0.019) and a higher detection rate (62% vs. 30%, p = 0.0008). For MRI, the detection rate was also significantly greater in HPV‐positive patients (49% vs. 25%, p = 0.009), whereas sensitivity did not differ significantly between the two groups. Across both imaging modalities, HPV status had no significant impact on specificity, PPV, or NPV. A primary SCC was identified in 27 patients (15%) solely on PET/CT and in 11 patients (6%) exclusively on MRI (Figure 1B). When combining PET/CT and MRI, the detection rate was 58%.

The primary tumors solely identified by PET/CT were located in the palatine tonsils (44%), base of tongue (37%), other oropharynx (7%), and other locations (4%). The majority (89%) of these were staged as T1 or T2. The primary tumors solely identified by MRI were located in the skin in relation to the parotid gland (40%), base of tongue (30%), hypopharynx (10%), nasopharynx (10%), and palatine tonsils (10%). All of these were staged as T1 or T2.

3.4. Subgroup Analysis of Patients With Primary Tumor Within the Head and Neck Area

A total of 146 patients were diagnosed with a primary tumor within the head and neck region, corresponding to 78% of the overall number of identified primary tumors. Of these, 103 (71%) were identified by PET/CT and 91 (62%) were identified by MRI, p = 0.1, whereas 120 (82%) of the diagnosed head and neck primary tumors were identified when combining the two imaging modalities (Figure 1C).

4. Discussion

To our knowledge, this study is the first to assess the individual and combined diagnostic value of PET/CT and MRI in HNCUP, encompassing diverse histological subtypes representative of real‐world clinical scenarios. In addition, it is the largest study to date to explore this aspect within the HNSCCUP subgroup.

In patients with clinical HNCUP, detection rates for PET/CT and MRI were 43% and 36%, with detection rates rising to 52% and 43%, respectively, in patients with HNSCCUP. In combination, the two imaging modalities identified a primary tumor in 51% of clinical HNCUP patients rising to 58% of clinical HNSCCUP patients. Based on the total population of clinical HNCUP patients, PET/CT was found to be more sensitive than MRI, with detection rates of 74% and 66%, p = 0.04, respectively. Accordingly, omission of PET/CT, that is, subjecting the patients to only MRI, would reduce the number of detected primary tumors by 30% based on imaging. Inversely, if only PET/CT was to be utilized, 17% of the tumors would remain occult on imaging.

A subgroup analysis of patients with clinical HNSCCUP, comprising 60% of our cohort, found that omitting PET/CT would reduce the number of detected primary tumors by 26% on imaging. Likewise, omitting MRI would reduce the number of detected primary tumors by 11%.

Inherently, the use of whole‐body PET/CT allows for detection of primary tumors located outside the head and neck, extending beyond the confines of MRI. In the present study, PET/CT and MRI were both found to be efficient in the identification of primary tumors within the head and neck area. However, as more than 20% of the diagnosed primary tumors were located outside this region, the additional diagnostic gain of full‐body PET/CT is obvious. Nevertheless, MRI still identified a considerable number of tumors not detected by PET/CT. In addition, MRI gains by providing high‐contrast resolution with superior soft tissue visualization, generally making MRI the preferred imaging modality for detailed evaluation of local and regional disease.

To some degree, our results align with those previously reported in studies by Gődény et al. [6] and Noij et al. [7] comparing the use of PET/CT and MRI in the diagnostic work‐up of clinical HNSCCUP patients. In the study by Gődény et al., scans were subject to blind re‐assessment by two radiologists and one nuclear medicine consultant, whereby their results were based on a clinical read of the scans. Noij et al. applied a Likert‐Scale to gauge 18F‐FDG uptake on PET/CT, which shared similarities with the scale used in our study. The studies reported a significantly higher sensitivity (94.4% and 93.8%) of PET/CT compared to MRI (88.2% and 81.3%). However, both studies were based on markedly smaller populations than our study (n = 14 and n = 31). The authors concluded that PET/CT should be the primary imaging modality in the diagnostic work‐up of patients with clinical HNSCCUP.

In comparison, we reported a sensitivity of PET/CT and MRI of 84% and 75%, respectively, when applied on patients with clinical HNSCCUP, not reaching statistical significance, p = 0.1. Although MRI demonstrated a significantly higher detection rate in HPV‐positive patients, the difference in sensitivity did not reach statistical significance. This discrepancy may reflect methodological and biological factors. Sensitivity is calculated only among patients with confirmed primaries, a smaller subset, which reduces statistical power and may mask clinically relevant trends. Furthermore, HPV‐positive tumors often present as small, deeply located lesions in the oropharynx, which can be challenging for MRI despite its superior soft tissue contrast. Thus, while HPV status appears to influence overall detectability, MRI's inherent limitations and the restricted sample size likely contributed to the lack of statistical significance in sensitivity. Future studies with larger cohorts are warranted to clarify whether this represents a true absence of effect or a power‐related issue.

Unlike previous studies, we incorporated data on emerging primary tumors identified during a five‐year follow‐up in patients initially diagnosed with true HNSCCUP. This approach minimizes potential bias in sensitivity and specificity estimates and strengthens our findings by capturing additional primaries that would otherwise remain undetected. All PET/CT and MRI scans conducted in this period were protocoled. This approach may have led to an underestimation of detected primary tumors since relevant clinical information was withheld from the imaging specialists. Nevertheless, this was necessary to evaluate the diagnostic value of each modality independently. Furthermore, as the scans were evaluated only once by one of the radiologists or nuclear medicine consultants, we have no data on inter‐reader nor intra‐reader agreement for either scanning modality.

While the single‐center design may appear limiting, all patients in the Central Denmark Region with suspected head and neck cancer are referred to the regional oncological center at Aarhus University Hospital, as per organizational guidelines. This ensures that our cohort of 295 patients is consecutive, representing a catchment area of approximately 1.3 million inhabitants. Thus, the risk of selection bias is minimal despite the retrospective design.

A key point put forward by Gődény et al. [6] is that although PET/CT may be superior in several diagnostic aspects, MRI outperforms in the staging of especially small and/or superficial tumors. Likewise, as PET/CT is based on uptake of ¹⁸F‐FDG, interpretation may be challenged in areas with active lymphoid tissue, muscular activity or inflammation, as is often seen in the head and neck region. Thus, subclinical superficial primary tumors embedded in such tissue may remain unidentified if only examined using PET/CT.

Reinholdt et al. [18] employed SUV‐max to assess incidental 18F‐FDG‐avid foci in the palatine tonsils. Thus, rather than disregarding 18F‐FDG grading entirely, it should be considered as one component of a broader, multifaceted diagnostic assessment.

Taken collectively, our findings indicate that PET/CT yields more favorable outcomes than MRI in cases of HNCUP and HNSCCUP, partly due to its broader field of view. However, MRI remains highly valuable in the head and neck region due to its complementary advantages over PET/CT and the detailed assessment of both nodal disease and primary tumors. Notably, the primary tumor is identified in the majority of HNCUP cases, further underscoring the importance of MRI in these evaluations. Given the clinical preference for MRI and the low number of patients needed to scan for the identification of a primary tumor otherwise undetected on PET/CT, both imaging modalities are recommended in the diagnostic work‐up of clinical HNCUP.

5. Conclusion

Our study supports the complementary roles of PET/CT and MRI in the diagnostic work‐up of clinical HNCUP. While both modalities contribute valuable information, PET/CT tended to identify more primary tumors and may be considered the preferred initial imaging modality when only one can be performed. MRI remains important for staging and surgical planning, and its use alongside PET/CT may enhance overall diagnostic precision and treatment planning.

Funding

The authors have nothing to report.

Ethics Statement

The study was approved by The Scientific Ethics Committees for the Central Jutland Region, case no. 1‐10‐72‐225‐22.

Consent

The granted approval from The Scientific Ethics Committees for the Central Jutland Region waived the need for consent to publication from the participants. Publication is, however, only allowed in an anonymous form, and sharing of raw data outside study center is not permitted.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

The full dataset generated and analyzed during the current study is not publicly available. The granted approval only allows for publication of anonymized data.