*3.2. Combining US Features with Cytology for Diagnostication*

The diagnostic performance of cytology alone, using the TIR3A/III class as the threshold, showed a sensitivity of 100% (34/34, 95% CI = 89.7–100%), a specificity of 91% (334/367, 95% CI = 87.6–93.7%), a PPV of 50.7% (34/67, 95% CI = 38.2–63.2%), and an NPV of 100% (334/334, 95% CI = 98.9–100%) (Table 4).

As expected, combining the TIRADS and cytology classes did not improve their performance. However, a decision tree applied to cytology and single US features of the two TIRADS systems identified two independent characteristics, i.e., echogenic foci and irregular margins, able to contribute to diagnostication, especially in the setting of undetermined nodules (Figure 1).

Indeed, this approach adequately detected four cases of carcinoma, otherwise classified as TIR3A/III (*n* = 1) and TIR3B/IV (*n* = 3). Moreover, three indeterminate nodules (one TIR3A/III and two TIR3B/IV) were wrongly classified as benign by the application of this approach and corresponded to an encapsulated follicular variant of papillary thyroid carcinoma (EFVPTC), whose malignant potential is still debated. Similarly, among the eight cases misclassified as malignant by the decision tree, four were diagnosed as follicular adenomas on the final histology (two TIR3A/III and two TIR3B/IV), and, although these lesions are benign, they usually require the complete histological assessment of the final surgical specimen for the exclusion of carcinoma. Finally, the adoption of this combined US/cytology approach led to a significant increase in the specificity (97.5% vs. 91%, *p* < 0.0001) and PPV (77.5% vs. 50.8%, *p* < 0.0001), with still excellent values of sensitivity (91.2% vs. 100%, *p* = 0.0833) and NPV (99.2% vs. 100%, *p* = 0.0820), as compared to cytology alone (Table 4).

**Figure 1.** Nodules' risk of malignancy (ROM, based on histology or follow-up) using a combination of cytology with ultrasound ACR features of the echogenic foci and margins.
