Baseline demographics of study participants

Table 1 details the baseline profiles of all patients. Our initial sample size included 522 patients with 699 nodules, recruited from patients who received conventional US, SMI mode US, and FNA at our institution between January 2020 and December 2021. Of these, 5 nodules were eliminated owing to a prior history of FNA (n = 2) or ablation (n = 3). Additionally, 56 nodules, particularly, 39 Bethesda I, 15 Bethesda III, and 7 Bethesda IV nodules, were eliminated due to unavailable pathological reference standards (Fig. 1). In the final training cohort, there were 471 participants (median age, 42 years [IQR, 34–54 years]), among which 103 were males (median age, 41 years [IQR, 34–53 years]) and 368 were females (median age, 42 years [IQR, 34–54 years]). Moreover, the total nodules were 643 (median size 10.9 mm, 7.5–13.9 mm), among which, 298 were benign thyroid nodules (median size 10.8 mm, [IQR, 7.2–15.8 mm]) and 345 were malignant thyroid nodules (median size 10.9 mm, [IQR, 7.9–12.8 mm]).

Table 1 Baseline data of training set and validation setsFig. 1

Flow diagram of the included patients and number of thyroid nodules. SMI, superb microvascular imaging; FNA, fine-needle aspiration; n, number of thyroid nodules

Between the period of January 2022 and June 2022, 159 thyroid nodules from 144 patients (median age, 42 years [IQR, 34–56 years]) were selected from our internal validation cohort, among which, 120 were females (median age, 43 years, [IQR, 34–56 years]) and 24 were males (median age, 40 years, [IQR, 35–49 years]). Additionally, between January 2021 and December 2021, as our external validation cohort, we selected 111 thyroid nodules from 106 patients (median age, 46 years [IQR, 36–56 years], among which, 81 were females (median age, 49 years, [IQR, 36–56 years]) and 25 were males (median age, 42 years, [IQR, 34–49 years]). Table 1 summarizes the baseline profiles of internal and external validation cohorts.

Development of the SMI TI-RADS estimation model

According to ACR TI-RADS, we established SMI TI-RADS for thyroid nodule malignancy prediction and FNA indication assessment. The univariable (Table S1 [online]) and multivariable logistic regression analyses indicated that nodule shape, margins, echogenicity, echogenic foci, extrathyroidal extension, vascularity, ring-SMI pattern, penetrating vascularity, flow-signal enlarged were strong indicators of thyroid carcinoma, and therefore, these parameters were employed for establishment of the SMI TI-RADS estimation model (Tables 2 and S2 [online]). Within this model, hyperechoic or isoechoic, shapes that were wider rather than tall, smooth margins, a lack of calcification or large comet tail artifacts, absence of an extrathyroidal extension, absent or mixed type of vascularity, smooth ring-SMI pattern, none too rare penetrating vascularity, and absent of flow-signal enlarged were given 0 points, while nodules that were either hypoechoic or highly hypoechoic, margins that were lobulated or irregular, peripheral or macrocalcifications, extrathyroidal extensions, central or perinodular type vascularity, absent of ring-SMI patterns, and present of flow-signal enlarged were given 1 point, and punctate echogenic foci, showing greater height than width, punctate echogenic foci, multiple penetrating vascularity, and cloud ring-SMI pattern were given 2 points (Figs. 2–5 and S1 [online]).

Table 2 Association between thyroid nodule malignancy and various SMI featuresFig. 2

Monochrome SMI and conventional US images show features of vascularity

Fig. 3

Monochrome SMI and conventional US images show features of ring-SMI patterns

Fig. 4

Monochrome SMI and conventional US images show features of penetrating vascularity

Fig. 5

Monochrome SMI and conventional US images show features of the area enlarged

The total points determined the nodule SMI TI-RADS category (referred to as TR hereafter): 0 points represented TR1 (benign thyroid nodules; fitted probabilities [FP]: 0%), 1 point indicated TR2 (not suspicious; FP: 1%), 2 points meant TR3 (mildly suspicious; FP: 2–5%), 3 points represented TR4A (moderately suspicious; FP: 9–17%), 4 points indicated TR4B (moderately suspicious; FP: 24–44%), 5 points meant TR4C (moderately suspicious; FP: 56–75%), and 6–11 points represented TR5 (highly suspicious; FP: > 90%) (Fig. S2 [online]). The indication for FNA was dependent on an integration of nodule maximum diameter and classification (Fig. 6).

Fig. 6

Chart shows the five categories of the SMI TI-RADS, with FP and indications for FNA

Comparison between SMI TI-RADS and other risk classification systems

In all, we employed nine RSSs (namely, C TI-RADS, ACR TI-RADS, EU TI-RADS, KSThR TI-RADS, AACE TI-RADS, ATA TI-RADS, and SMI TI-RADS) to assess the risk of malignancy in the classifications of these guidelines. The application of the RSS was generally within the recommended range of malignancy, apart from the moderate and mild suspicion and not suspicious or benign classifications of ACR TI-RADS, the intermediate and low-risk classifications of EU TI-RADS, and the intermediate, low, and very low suspicion classifications of ATA TI-RADS. Beyond this, approximately 6.5% of the nodules (42/643) did not meet any of the classification standards in ATA TI-RADS, and showed a malignant risk of 57% (24 of 42 nodules). The other two RSSs (namely, BTA TI-RADS and FSE TI-RADS) did not propose malignancy risk estimates for each category (Table S3 [online]).

To test the efficiencies of the RSS diagnoses and evaluations, we compared the sensitivities, specificities, and positive and negative predictive values at an optimal diagnostic cut-off, as summarized in Table 3. The area under the receiver operating characteristic curve (AUC) of SMI TI-RADS (0.94 [95% CI: 0.92, 0.96]) was relatively enhanced, compared to that of C TI-RADS (0.86 [95% CI: 0.83, 0.89], p < 0.001), EU TI-RADS (0.84 [95% CI: 0.81, 0.86], p < 0.001), ACR TI-RADS (0.84 [95% CI: 0.81, 0.87], p < 0.001), KSThR TI-RADS (0.80 [95% CI: 0.77, 0.83], p < 0.001), AACE TI-RADS (0.86 [95% CI: 0.83, 0.88], p < 0.001), ATA TI-RADS (0.84 [95% CI: 0.81, 0.87], p < 0.001), FSE TI-RADS (0.82 [95% CI: 0.79, 0.85], p < 0.001), and BTA TI-RADS (0.76 [95% CI: 0.73, 0.80], p < 0.001) (Fig. S3 [online]). The SMI TI-RADS exhibited the largest malignancy biopsy yield at 79% (189/240 nodules), and was closely followed by ACR TI-RADS at 74% (190/258 nodules), AACE TI-RADS at 72% (197/273 nodules), FSE TI-RADS at 68% (191/280 nodules), EU TI-RADS at 66% (200/302 nodules), C TI-RADS at 64% (211/332 nodules), ATA TI-RADS at 61% (210/342 nodules), KSThR TI-RADS at 59% (211/355 nodules), and BTA TI-RADS at 59% (328/557 nodules) (Fig. S4 [online]).

Table 3 Comparison of the diagnostic performances of TI-RADS

The SMI TI-RADS also displayed the least UBR at 21% (51/240 nodules), closely accompanied by ACR TI-RADS at 26% (68/258 nodules), AACE TI-RADS at 28% (76/273 nodules), FSE TI-RADS at 32% (89/280 nodules), EU TI-RADS at 34% (102/302 nodules), C TI-RADS at 36% (121/332 nodules), ATA TI-RADS at 39% (132/342 nodules), KSThR TI-RADS at 41% (144/355 nodules), and BTA TI-RADS at 41% (229/557 nodules). For the purposes of this study, nodules with indications for biopsy were considered positive and those without indications for biopsy were considered negative, their sensitivities, specificities, and positive and negative predictive values are presented in Table 4.

Table 4 Comparison of the biopsy performance of TI-RADSValidation

In the internal validation cohort, the AUC, sensitivity, specificity, BYM, and UBR of the SMI TI-RADS were 0.89 (95% CI: 0.83, 0.94), 0.83 (95% CI: 0.76,0.91), 0.85 (95% CI: 0.75, 0.94), 79% (59/75 nodules), and 21% (16/75 nodules), respectively. In the external validation cohort, the AUC, sensitivity, specificity, BYM, and UBR of the SMI TI-RADS were 0.91 (95% CI: 0.85, 0.96), 0.89 (95% CI: 0.81, 0.97), 0.86 (95% CI: 0.76, 0.96), 72% (31/43 nodules), and 28% (12/43 nodules), respectively.

Evaluation of interobserver variability

Based on our interobserver variability assessment, the κ value was 0.63 for echogenicity, 0.72 for shape, 0.75 for margin, 0.52 for echogenic foci, 0.66 for extrathyroidal extension, 0.46 for vascularity at SMI mode, 0.55 for ring-SMI patterns, 0.70 for penetrating vascularity, and 0.60 for flow-signal enlarged at SMI, respectively (Table S4 [online]).