Fifty consecutive patients with newly diagnosed head and neck squamous cell cancer were enrolled in this FDG PET study. The study group was composed of 37 men and 13 women with a mean age of 68 years (range, 51–89 years). In addition to a routine pretreatment physical examination and panendoscopy, all patients underwent head and neck MRI as well as FDG PET. Each case was staged according to the 1997 International Union Against Cancer (UICC) TNM staging system [7]. Patients presenting with distant metastatic disease (M1) were excluded. The distribution of TNM stages in the 50 patients was as follows: stage I in 11 patients, stage II in six, stage III in five, and stage IV in 28. All patients provided written informed consent for participation in the PET study, which had been approved by our institutional review board.

For the 50 patients, all treatment consisted of definitive radiotherapy with or without chemotherapy or radical surgery with or without adjuvant chemoradiation therapy. Thirty-four patients were scheduled to receive definitive radiotherapy at doses of 60.0–70.0 Gy; we refer to these patients as the “radiotherapy group.” Of these, 17 patients received concomitant chemotherapy, generally using docetaxel and carboplatin or cisplatin (or both) with or without a continuous infusion of 5-fluorouracil. Sixteen patients underwent radical surgery and corresponding neck dissections, and we refer to these patients as the “surgery group.” Of these patients, 12 received adjuvant chemotherapy with or without radiotherapy.

All patients fasted for at least 5 hours before PET studies. Serum glucose levels measured at the time of FDG injection were < 150 mg/dL in all patients. A whole-body PET scanner (SHR22000, Hamamatsu Photonics) was used [8]. The SHR22000 scanner permits simultaneous acquisition of 63 transverse planes of 3.6-mm thickness encompassing a 23.0-cm axial field of view. Whole-body emission scans were obtained with five bed positions starting 60 minutes after IV administration of FDG (400–500 MBq) [9]. Acquisition time was 6 minutes per bed position. A transmission scan for segmented attenuation correction was then acquired with a ⁶⁸Ge external ring source for 3 minutes per bed position. To minimize the accumulation of FDG activities in the urinary bladder, patients were asked to void just before emission scan. Transaxial, coronal, and sagittal images were reconstructed by means of the ordered-subset expectation maximization (OSEM) method with 8 subsets and 3 iterations. The average reconstructed x–y spatial resolution was about 4.0-mm full width at half maximum (FWHM) in plane.

PET images were visually interpreted and analyzed in comparison with the corresponding MR images. Any foci of FDG uptake that were increased relative to the surrounding normal soft tissue and were not located in areas of physiologically increased uptake were considered to be positive for primary or metastatic lesions. For semiquantitative analysis, the SUV was calculated for the primary tumor in each patient using the following equation: SUV = tracer activity / injected dose normalized to body weight. A computerized semiautomated algorithm was used to eliminate interobserver discrepancy [8]. This method helps to define the maximal uptake in a small square 1.0 × 1.0 cm region of interest (3 × 3 pixels) placed within the tumor that was biopsied. To minimize the partial volume effect, the maximal SUV value within a pixel was used to represent tumor FDG uptake.

The Kaplan-Meier method was applied to calculate local control and disease-free survival rates. We determined a statistically significant SUV cutoff value for the survival analysis using the log-rank test. Tumor persistence or recurrence was documented by at least two different examinations (endoscopy, MRI, or PET). Persistent or recurrent tumor at the initial primary site was considered as an event in determining local control rate; for disease-free survival, nodal metastases as well as distant metastases were also taken into account. The time interval for the end points was calculated from the first day of treatment until the date of an event or the last follow-up. The log-rank test was applied to assess the correlation of these end points with the clinical risk factors (primary tumor SUV, T stage, N stage, and patient age) and therapeutic variables (treatment groups). Bivariate Cox proportional hazards analyses were performed for primary tumor SUV alone (unadjusted) or adjusted for each of the other tested risk factors.

We repeated this proportional hazards analysis using primary tumor SUV as a continuous variable. The Mann-Whitney U test was used to compare primary tumor SUV in different subgroups. Comparison of subgroups for the probability of recurrence was performed with Fisher's exact test. A two-sided p value of < 0.05 was considered significant.

FDG PET correctly detected the primary tumor in 46 patients, but did not show the primary lesion in four patients with stage T1 disease, including laryngeal cancer (n = 3) and tongue cancer (n = 1). MRI findings for three of the four patients were also false-negative. All four patients had N0 disease. In semiquantitative analysis, tumor SUV for these four patients with false-negative FDG uptake was defined as 2.0 to reflect typical baseline SUV of normal head and neck soft tissues.

The median tumor SUV for all 50 patients was 10.53 (range, 2.0–32.34). It was 8.99 and 11.28 for the radiotherapy and surgery groups, respectively (p = 0.78). When tumor SUV was correlated with T stage, T1–T2 tumors had a significantly lower median SUV than T3–T4 tumors (4.77 vs 12.54, respectively; p = 0.001). There was a significant difference of the median SUV between N0 and N1–N3 diseases (4.77 vs 11.40, p = 0.027).

At the last follow-up, 41 patients were alive and nine patients had died. The median follow-up for the surviving patients was 15 months (range, 2–56 months). Of the 41 patients, seven presented with persistent or recurrent local or regional lesion, whereas the remaining 34 patients were free from disease. The discriminative value of various cutoff SUVs for FDG uptake of primary tumor was analyzed in the context of local control and disease-free survival. The most discriminate cutoff SUV for prognosis proved to be 7.0, although dichotomization with a broad range of SUVs of between 5 and 10 gave significantly discriminative log-rank p values. Therefore, an SUV of 7.0 was used as the cutoff value in the following analyses.

In the log-rank test, patients with low SUVs (≤ 7.0) had significantly higher rates of local control (91% vs 55%, respectively; p = 0.0067; Fig. 1) and disease-free survival (91% vs 55%, p = 0.0051, Fig. 2) at 2 years than patients with high SUVs (> 7.0) (Table 1). Patients with early T stage (T1–T2) also showed significantly higher rates of local control (87% vs 45%, p = 0.0008) and disease-free survival (87% vs 45%, p = 0.0006) at 2 years than those with advanced T stage (T3–T4), whereas N stage, patient age, and treatment strategy did not predict for the disease outcome.

In the Cox proportional hazards analysis, tumor SUV and T stage were significant and independent predictors of local control and disease-free survival at 2 years, whereas N stage, patient age, and treatment strategy were not identified as significant prognostic factors for either local control or disease-free survival (Table 2). The local control and disease-free survival hazard ratios for tumor SUV remained significant when adjusted for N stage and patient age. When we repeated this proportional hazards analysis using primary tumor SUV as a continuous variable, tumor SUV was significantly predictive of both local control (p = 0.0043) and disease-free survival (p = 0.0032). The hazard ratios of local control and disease-free survival for continuous tumor SUV remained significant when adjusted for N stage and patient age.

In Table 3, the group of patients with a high tumor SUV (14/29 relapses) were associated with poorer outcome than those with a low tumor SUV (2/21 relapses) (p = 0.0052). High T stage (12/20 relapses) was also associated with poorer outcome than low T stage (4/30 relapses) (p = 0.0014), whereas N stage, patient age, and treatment strategy did not predict disease outcome. Table 4 indicates that, in the group of patients with a high tumor SUV, a high T stage (12/19 relapses) was more associated with recurrence than a low T stage (2/10 relapses) (p = 0.0502), although such a relationship was not observed in the group of patients with a low tumor SUV.