*3.2. Lesion-Based Analysis*

Finally, a total of 130 lesions (9 in the original thyroid area, 121 in the cervical lymph node) were examined, including 3 malignant and 6 benign thyroid nodules, as well as 74 malignant and 47 benign cervical lymph nodes. Of all lesions, 44 (33.8%) were analyzed by histopathology, including 32 malignant lesions (1 local recurrence and 31 metastatic lymph nodes) and 12 benign lesions (2 benign thyroid nodules and 10 benign lymph nodes). Other lesions were confirmed by strict follow-up in the aforementioned manner. Among 130 lesions, PET/MR detected 119 (91.5%), but missed 1 recurrent thyroid cancer and 10 metastatic lymph nodes; PET/CT detected 105 (80.8%), but missed 2 recurrent thyroid cancers, 21 metastatic lymph nodes, 1 benign thyroid nodule, and 1 benign lymph node. For 74 lymph node metastases, the detection numbers of PET/MR and PET/CT were, respectively, 64 (86.5%) and 53 (71.6%). Table 2 shows the number of thyroid nodules and lymph nodes detected by 18F-FDG PET/MR and PET/CT.


**Table 2.** Number of thyroid nodules and lymph nodes detected by 18F-FDG PET/MR and PET/CT.

The diagnostic performance of two PET modalities (PET/MR and PET/CT) and MR (alone) is displayed in the Table 3 and Figure 4. The diagnostic sensitivities of PET/MR, MR, and PET/CT were significantly different (80.5%, 77.9%, and 61.0%, *p* = 0.012). The paired comparison of PET/MR, MR, and PET/CT showed significant differences of sensitivity between PET/CT and the other two techniques (*p* < 0.001 and *p* = 0.007 respectively). The specificities of the three modalities (84.9%, 83.0%, and 81.1%, *p* = 0.875) showed no significant difference, also in paired comparisons. For metastatic lymph nodes, the sensitivities of PET/MR, MR, and PET/CT were 81.1%, 78.4%, and 62.2% (*p* = 0.018). Paired comparisons also showed significant differences between PET/CT and the other two techniques (*p* = 0.001 and *p* < 0.001 respectively). PPVs, NPVs, and accuracies in diagnosing all lesions (recurrent and malignant cervical lymph nodes) were generally consistent in

tendencies among the three imaging modalities, among which PET/MR yielded the optimal diagnostic performance. Both imaging modalities yielded false-positive results, either for suspicious thyroid nodules or lymph nodes. Based on the available pathological results, misdiagnosed lesions were mainly fibrofatty tissue.


**Table 3.** Diagnostic performance of two modalities.

**Figure 4.** Diagnostic performance of two modalities (ns represents no significance; \* represents the *p*-value < 0.05; \*\* represents the *p*-value < 0.01).

The image clarity scores for each region of the lesion are shown in Table 4. For all included lesions, the PET/MR scores were significantly higher than those of PET/CT (2.74 ± 0.60 vs. 1.9 ± 0.50, *p* < 0.001). The largest score was in level II (2.92 ± 0.33), which almost exceeded PET/CT (1.91 ± 0.45) by 1 point. For malignant lesions, there was no difference between PET/MR and PET/CT in the assessment of suspected thyroid nodules, while there were significant differences in the assessment of metastatic lymph nodes of all levels. The largest differences between PET/MR and PET/CT occurred in level II (2.83 ± 0.48 vs. 1.79 ± 0.58) and level V (2.73 ± 0.67 vs. 1.85 ± 0.46), as presented in Figure 3. For benign lesions, PET/MR had better presentation of thyroid nodules and lymph nodes in other regions, especially in levels II (3.00 ± 0.00 versus 2.02 ± 0.25) and IV (3.00 ± 0.00 versus 2.14 ± 0.36). Image conspicuity agreement was excellent (κ = 1 in thyroid nodules, κ = 0.981 in lymph nodes, *p* < 0.001) between the two physicians.


**Table 4.** Conspicuity score of recurrent or metastatic lesions in relation to location (scoring criteria referring to [26]).

Table 5 lists the SUVmax, SUVmean, and lymph node diameters in the maximum cross-sectional area of lymph nodes measured by PET/MR and PET/CT. The SUVmax of malignant lymph nodes was significantly higher than that of benign lymph nodes on PET/MR (median 2.6 vs. 2.2, *p* = 0.004) and PET/CT (median 2.0 vs. 1.8, *p* = 0.006). However, SUVmax showed an overlap in benign and malignant lymph nodes (Figure S1). Figure 5 presents an isolated malignant lymph node with intense 18F-FDG uptake on PET/MR. The SUVmean, long diameter, and short diameter indicated no significant differences between benign and malignant nodes on both modalities (Figure S1). For all included lymph nodes, SUVmax, SUVmean, and lymph node diameters of PET/MR were higher than those measured on PET/CT by 17.4%, 22.2%, 11%, and 18% (all *p* ≤ 0.001), respectively. The Bland–Altman analysis showed great consistency among modalities between the SUVmax, SUVmean, and diameters (Figure S2). Additionally, there were correlations between the parameters detected on PET/MR and PET/CT (all *p* < 0.001, Figure S3). The correlation coefficient of SUVmax for all lymph nodes was 0.661 (Figure S3).

**Figure 5.** A 51-year-old DTC patient after comprehensive treatment presented with an unexplained increase in Tg (2.88 μg/L). PET/MR depicted an increased uptake of a small lymph node (long diameter, 0.7 cm) in the left level III, with SUVmax 5.1, without other apparently abnormal lymph nodes ((**a**,**b**), red arrows). 30 suspicious lymph nodes were removed by re-operation, and notably, only the lymph node identified on PET/MR was malignant.


**Table 5.** SUVs, diameters of lymph nodes detected by both PET/MR and PET/CT.

<sup>1</sup> The data are uniformly expressed as median (25th–75th %).
