**4. Results**

The demographic and clinicopathological characteristics of the 56 patients included in the study and their previous treatments are reported in Table 1. The median age was 39.1 years (range 24–72), and the population was mainly Italian (92.9%). In total, 23 cancers (41%) were squamous cell carcinomas, 23 (41%) adenocarcinomas, and 10 other histotypes (18%), including 4 adenosquamous tumors, 3 clear cell variants, 2 neuroendocrine carcinomas, and 1 adenoid basal carcinoma. Invasive cervical cancer tumors were confirmed in 55 patients, while no residual tumor was found in one case. None of our patients reported any discomfort during the procedure.

Forty-two (75%) lesions were described as highly vascularized (color score 4—CS 4), 12 (21.4%) as moderately vascularized (CS 3), and 2 (3.6%) as minimally vascularized. We did not find any correlation between vascularization and histotype, but we found a statistically significant correlation between vascularization and the largest dimension of the lesion, with a mean of 26.9 mm in CS 4 and 12.4 mm in CS 3 (*p* < 0.001).

Out of the six patients with prior conization and residual disease, MRI detected the lesion only in three cases, while VGS detected all six cases. In Figure 1, it is possible to observe one of these cases.

Squamous tumors were predominantly hypoechoic (14/23; 60.1%), while adenocarcinomas were mainly hyperechoic (16/23; 69.5%) (*p* = 0.005). Figure 2 shows an example of two echogenicities, with and without VGS.

When the stromal infiltration was analyzed, there was a good concordance between histology and VGS, with a sensitivity of 89% (95% CI 0.44–0.83), a specificity of 89% (95% CI 0.52–0.86), and a Cohen's kappa of 0.78. When the stromal invasion was >2/3, the histological exam identified five of six (83.3%) cases with positive lymph nodes (6), while VGS predicted four cases (66.6%).

Although we did not compare the fornix infiltration to the gold standard because of the early stage of the tumors, we observed eight false-positive cases of fornix infiltration with MRI and two with VGS, with a specificity of 85.7% and 96.4% and a negative predictive value (NPV) of 88.9% and 100%, respectively.

VGS significantly overestimated the lesion size by 2.7 mm compared to the gold standard. On the other hand, MRI underestimated it by 1.9 mm (Table 2). Figure 3 is an example of the small difference in lesion size by the methods compared with the final specimen.


**Table 1.** Patients' characteristics and treatments, N = 56.

*<sup>a</sup>* Statistics are: N (%) for categorical variables, mean (SD) otherwise; SD = standard deviation; *<sup>b</sup>* NACHT = neoadjuvant chemotherapy; *<sup>c</sup>* adenosquamous N = 4, clear cell spinocellular variant N = 3, adenoid basal carcinoma N = 1, large cell neuroendocrine carcinoma N = 2; *<sup>d</sup>* min = 24, max = 72; *<sup>e</sup>* N = 49; *<sup>f</sup>* N = 54. VGS = vaginosonography; SCC = squamous cells carcinoma; MRI = magnetic resonance imaging.

**Table 2.** Maximum diameter (mm) methods comparison with histology (gold standard) and summary statistics.


*<sup>a</sup>* Two-sided paired t-test: lower limit of agreement: *<sup>b</sup>* <sup>−</sup>9.5 mm, *<sup>c</sup>* <sup>−</sup>18.8 mm; cases under limit: *<sup>b</sup>* N = 2 (3.6%), *<sup>c</sup>* N = 1 (1.8%); upper limit of agreement: *<sup>b</sup>* 15.0 mm, *<sup>c</sup>* 15.1 mm; cases over limit: *<sup>b</sup>* N = 3 (5.4%), *<sup>c</sup>* N = 2 (3.6%).

**Figure 1.** Small tumor after conization. (**A**) Not detectable disease by MRI. (**B**) Detectable disease by TVUS, measuring 10.3 × 8.3 mm. (**C**) Lesion vascularization by power Doppler. (**D**) Disease detectable by VGS, measuring 13.3 × 6.9 mm. Squamous tumors were predominantly hypoechoic (14/23; 60.1%), while adenocarci-**Figure 1.** Small tumor after conization. (**A**) Not detectable disease by MRI. (**B**) Detectable disease by TVUS, measuring 10.3 × 8.3 mm. (**C**) Lesion vascularization by power Doppler. (**D**) Disease detectable by VGS, measuring 13.3 × 6.9 mm. *Diagnostics* **2022**, *12*, x FOR PEER REVIEW 6 of 12

**Figure 2.** Echogenicity by TVUS and VGS. (**A**) Small adenocarcinoma (arrows) at TVUS; (**B**) the same tumor at VGS, with better definitions of the margins and more hyperechoic echogenicity (arrows). (**C**) Squamous tumors (arrows) at TVUS; (**D**) the same tumor at VGS, with better definitions of the margins and more hypoechoic echogenicity (arrows). When the stromal infiltration was analyzed, there was a good concordance between **Figure 2.** Echogenicity by TVUS and VGS. (**A**) Small adenocarcinoma (arrows) at TVUS; (**B**) the same tumor at VGS, with better definitions of the margins and more hyperechoic echogenicity (arrows). (**C**) Squamous tumors (arrows) at TVUS; (**D**) the same tumor at VGS, with better definitions of the margins and more hypoechoic echogenicity (arrows).

tive value (NPV) of 88.9% and 100%, respectively.

**Method N Mean (SD) 95% CI** *p***-Value** *<sup>a</sup>* Histology 56 23.4 (11.2) (20.4, 26.4) - VGS 56 26.1 (9.8) (23.5, 28.7) - MRI 56 21.5 (12.5) (18.2, 24.9) - Bias VGS—Histology *<sup>b</sup>* 56 2.7 (6.3) (1.1, 4.4) 0.002 Bias MRI—Histology *<sup>c</sup>* 56 −1.9 (8.5) (−4.2, 0.4) 0.11

histology and VGS, with a sensitivity of 89% (95% CI 0.44–0.83), a specificity of 89% (95% CI 0.52–0.86), and a Cohen's kappa of 0.78. When the stromal invasion was >2/3, the his-

Although we did not compare the fornix infiltration to the gold standard because of the early stage of the tumors, we observed eight false-positive cases of fornix infiltration with MRI and two with VGS, with a specificity of 85.7% and 96.4% and a negative predic-

VGS significantly overestimated the lesion size by 2.7 mm compared to the gold standard. On the other hand, MRI underestimated it by 1.9 mm (Table 2). Figure 3 is an example of the small difference in lesion size by the methods compared with the final

**Table 2.** Maximum diameter (mm) methods comparison with histology (gold standard) and sum-

specimen.

mary statistics.

*<sup>a</sup>* Two-sided paired t-test: lower limit of agreement: *<sup>b</sup>* −9.5 mm, *<sup>c</sup>* −18.8 mm; cases under limit: *<sup>b</sup>* N = 2 (3.6%), *<sup>c</sup>* N = 1 (1.8%); upper limit of agreement: *<sup>b</sup>* 15.0 mm, *<sup>c</sup>* 15.1 mm; cases over limit: *<sup>b</sup>* N = 3 (5.4%),

**Figure 3.** Dimensions with both techniques compared with final histology. VGS image of IB2 N+ squamous tumor (**A**,**B**), MRI (**C**), and macroscopic appearance (**D**). **Figure 3.** Dimensions with both techniques compared with final histology. VGS image of IB2 N+ squamous tumor (**A**,**B**), MRI (**C**), and macroscopic appearance (**D**).

When comparing the size of the lesions, both MRI and VGS had a good concordance with the pathology report (Table 3), showing a Cohen's kappa of 0.73 and 0.81, respectively. If the tumor was <2 cm, VGS identified 16 cases, while MRI identified 18 cases out of 21. However, in the latter group, MRI missed eight cases whose dimensions were recorded as 0 mm (Figure 2D), corresponding to a false-negative rate of 38.1% (95% CI: 18.1– 61.6%). This explains the apparently higher sensitivity of MRI with respect to VGS for tumors smaller than 2 cm (Figure 4) When comparing the size of the lesions, both MRI and VGS had a good concordance with the pathology report (Table 3), showing a Cohen's kappa of 0.73 and 0.81, respectively. If the tumor was <2 cm, VGS identified 16 cases, while MRI identified 18 cases out of 21. However, in the latter group, MRI missed eight cases whose dimensions were recorded as 0 mm (Figure 2D), corresponding to a false-negative rate of 38.1% (95% CI: 18.1–61.6%). This explains the apparently higher sensitivity of MRI with respect to VGS for tumors smaller than 2 cm (Figure 4).


**Table 3.** Methods comparison with the gold standard by tumor size (maximum diameter, mm). **Table 3.** Methods comparison with the gold standard by tumor size (maximum diameter, mm).

*<sup>a</sup>*McNemar's test. *<sup>a</sup>* McNemar's test.

*<sup>c</sup>* N = 2 (3.6%).

Figure 5A shows the Bland–Altman plot of the maximum diameters in VGS and histopathology, while Figure 5B shows the diameter in MRI and histopathology. Visual inspection of the scatterplots suggests a good distribution of the data with both methods. The magnitude of the differences did not change with the mean of the two measurements. Overall, Figure 5C,D show no differences despite the diameter of the tumors, except for the cluster of eight tumors missed by the MRI (Figure 5B,D). Lin's correlation was strong for VGS (0.8) and moderate for MRI (0.7).

*Diagnostics* **2022**, *12*, x FOR PEER REVIEW 8 of 12

**Figure 4.** Detection of small tumors with VGS. (**A**,**B**) Small tumor not detectable at TVUS and MRI; (**C,D**) the same tumor detected by adding VGS and PD. Figure 5A shows the Bland–Altman plot of the maximum diameters in VGS and his-**Figure 4.** Detection of small tumors with VGS. (**A**,**B**) Small tumor not detectable at TVUS and MRI; (**C,D**) the same tumor detected by adding VGS and PD. Overall, Figure 5C,D show no differences despite the diameter of the tumors, except for the cluster of eight tumors missed by the MRI (Figure 5B,D). Lin's correlation was strong for VGS (0.8) and moderate for MRI (0.7)

(C) Passing–Bablok regression line for VGS and histology agreement

5. Discussion

selected patients.

(**A**) VGS and histology Bland–Altman agreement plot (**B**) MRI and histology Bland–Altman agreement plot

(D) Passing–Bablok regression line for MRI and histology agreement

Figure 5. Bland–Altman plot of both methods compared with the gold standard. **Figure 5.** Bland–Altman plot of both methods compared with the gold standard.

The role of ultrasound in the pre-treatment evaluation of cervical cancer is well established [2,16–18]. However, some factors hinder the ultrasound evaluation of some tumor characteristics, such as lesion margins (especially the small exophytic ones), fornix

VGS in the assessment of early-stage cervical cancer. VGS is an inexpensive and welltolerated ultrasound technique that can increase the image quality of cervical tumors. The acoustic windows created by saline solution results in a distance of the probe to the lesion, the distension of vaginal walls, and the exclusion of hindering factors, including vaginal

Compared to histology (used as the gold standard), VGS and MRI assess the dimension of the lesion with similar accuracy. However, we found that VGS tended to overestimate the dimensions, whereas MRI underestimated them. Previous studies have reported that ultrasounds overestimate tumor dimensions [7,17]. Despite avoiding confounding factors, such as adjacent normal tissues, by using liquid distension in order to create an acoustic window, our data showed that the small difference between VGS and histology was statistically significant, while the difference between MRI and histology was not. Even if overestimating the tumor size could be better than underestimating it in oncological cases, it may compromise the opportunity for fertility-sparing surgery in

The value of Lin's concordance in the assessment of the lesion dimensions was better between VGS and histology than MRI and histology. This is because MRI did not detect some small tumors, recorded as 0 mm in our study. Indeed, when the tumor was <2 cm, MRI was not able to identify 38.1% of the tumors. Therefore, the real detection rate of MRI was 47% (10 out of 21) versus 76.1% of VGS (16 out of 21). Hence, VGS is more sensitive than MRI in identifying small lesions. Prior studies did not find that TVUS was better than MRI in the detection of lesions <2 cm [9,20]. Only one previous study showed that TRUS

As our study focuses on early-stage cervical cancers, we have limited data on fornix infiltration. Nevertheless, we found that VGS was more sensitive than MRI in excluding fornix infiltration, with an NPV of 100% and 88.9%, respectively. Recently, similar results have been reported by comparing the accuracy of histology to TVUS, MRI, and clinical examination under anesthesia [21]. TVUS can be a good method to exclude but not to predict vaginal infiltration, which is often overestimated. Probably, VGS could play a role in the assessment of more advanced stages of cervical cancer with vaginal involvement.

was superior to MRI in the identification of very small lesions (<1 cm) [7].

collapse, bleeding, and/or mucus secretions.
