*3.3. Interrelationships of Histology and PET/MRI Parameters*

LNM demonstrated a higher SUV, larger diameters, higher RI, and ΔSUV than benign LN, as detailed in Supplementary Table S2.

Moreover, these differences were amplified by the grade of the primary tumor, as presented in Figure 2 and Supplementary Table S2. In particular, LNM from G3 tumors presented with significantly higher SUV and FDG dynamics between early and delayed scan measured with RI-SUVavg (*p* = 0.03) and ΔSUVavg (*p* = 0.02) compared with LNM from G2 tumors (*p* < 0.01; Supplementary Table S3). Furthermore, G3 LNM presented with a greater short-axis diameter vs. G2 LNM (*p* < 0.01) and a slight increase in sphericity (*p* = 0.08), while ADC revealed no significant difference.

**Figure 2.** Boxplots presenting [18F]FDG PET (**A**,**B**) and MRI (**C**,**D**) parameters of lymph nodes dependent on the tumor grade of the primary tumor derived from biopsy before PET/MRI. No LNM were present in G1 carcinomas.

LN short-axis diameter correlated significantly with SUVe, SUVd, BPCSUVe, BPCSUVd, and ΔSUVpeak (*p* < 0.01, r: 0.477–0.716) but not with RI-SUVpeak (r = 0.085) or ADC (r = 0.241).

G3 LNM revealed an increase in [18F]FDG uptake between early and delayed scans compared with benign LN (RI-SUVpeak and ΔSUVpeak: *p* < 0.01 and 0.02), as presented for representative cases in Figure 3a,b. A similar trend was observed for RI-SUVpeak in G2 LNM, though not reaching significance (*p* = 0.19).

**Figure 3.** (**a**). Case of a 49-year-old patient with pT1b2 G3 cervical cancer. Focal [18F]FDG uptake (arrow) of the right interiliac LN decreased by 33% between early (60 min, SUVavg 1.8) and delayed PET scan (88 min, SUVavg 1.2) and was histologically confirmed as lymphofollicular hyperplasia. (**b**). Case of a 41-year-old patient with pT2b G3 cervical cancer. The left iliac extern LNM (arrow) presents an ongoing [18F]FDG trapping between the early (60 min, SUVavg 2.1) and delayed scan (82 min, SUVavg 2.5) and a slight decrease in blood pool activity.

#### *3.4. PET/MRI Parameter Evaluation*

PET demonstrated high accuracy in differentiating between LNM and benign LN using an SUV-based quantification with an AUC of up to 0.809 (Figure 4 and Supplementary Table S2) without significant differences between the SUV quantification parameters SUVemax, SUVepeak, and SUVemean (*p* ≥ 0.54).

**Figure 4.** ROC analysis for the detection of lymph node metastases of selected [18F]FDG PET/MRI parameters for G 1-3 tumors (**A**) and (**B**) as well es G2 tumors (**C**) and G3 tumors (**D**) separately.

The delayed PET scan did not result in a significantly higher AUC than the early PET scan (*p* ≥ 0.55). Blood pool correction improved the AUC in the delayed PET slightly but nonsignificantly (SUVeavg: 0.784 vs. 0.766; SUVdavg: 0.741 vs. 0.767, *p* = 0.73).

The primary tumor grade crucially impacted the accuracy of LNM detection in PET with a significant decrease in discriminatory power in G2 versus G3 tumors (SUVeavg G2: 0.673; G3: 0.901, *p* < 0.01). The error rate (ER = false-positive + false-negative rate = 1-accuracy) was more than twice as high for G2 LNM (65.5%) as for G3 LNM (30.4%) at their individual optimal SUVeavg cut-off (Supplementary Figure S1), while the prevalence was comparable (G2: 17.5% vs. G3: 23.0%).

Dual-time-point kinetics calculated with RI and ΔSUV significantly correlated with malignancy, especially in G3 tumors with an AUC up to 0.791 (*p* < 0.01). The SUVpeak quantification method achieved the highest AUCs but required blood pool correction. Overall, the ΔSUV calculation method was comparable to the RI-SUV but performed slightly and nonsignificantly better in G3 tumors (G3 SUVavg: 0.791 vs. 0.718, *p* = 0.48).

LN diameters revealed a significant discriminatory power for short-axis (0.741) and long-axis (0.777) measurements and performed best in LNM from G3 tumors (AUC: 0.904 and 0.881). LN sphericity was not a significant stand-alone predictor of LNM, neither in G2 nor G3 tumors (*p* ≥ 0.269).

ADC presented a borderline significant discriminatory power (AUC 0.600, *p* = 0.05), with a significantly lower AUC compared with the SUVavg and short-axis diameter (*p* < 0.01 and *p* = 0.03, *n* = 162).

#### *3.5. Multiparametric Approach*

The parameters ADC, sphericity, bpcSUVeavg, and tumor grade of the primary tumor were identified as independent predictors of LNM and were included in the calculation of the MS, as described above. The response variable of the model were the probabilities of being malignant predicted by the model, calculated as a sum of the predictor values weighted according to their (fixed effect) regression coefficients. After listwise exclusion of cases with missing parameters, the sample size was 171 LN with 21.1% prevalence of metastases.

Using MS resulted in a high discriminatory power between malignant and benign LN (AUC: 0.820, 95% CI: 0.736–0.879). At the optimal cut-off value (Youden optimum: 0.042), the MS improved sensitivity from 63.5% to 72.2% compared with SUVeavg at a specificity of 80.7%.

Furthermore, error rates could be lowered (47.0%) and kept constant over a wider cut-off range compared with the best single parameter SUVeavg (52.7%), as presented in Supplementary Figure S1.

Further subgroup analysis focusing on the grade of the primary tumor revealed a significantly (*p* < 0.01) better prediction of LNM in G3 tumors (AUC 0.850, 95% CI: 0.755–0.945) compared with G2 tumors (AUC 0.695, 95% CI: 0.526–0.863). In particular, the parameter SUVe showed a markedly different predictivity for LNM in G2 compared with G3 tumors (log-odds: SUV: 1.5/17.7, *p* = 0.01).
