Gray zone possibility in repeat Bethesda III cytology: a retrospective cohort study of malignancy and multifocality in thyroid nodules Authors: Nevin Sakoglu, Ferhat Ozden Published: 2026 ABSTRACT PurposeThyroid cancer is the most common endocrine cancer, developing in an average of 5%–7% of all nodules. In this study, we aimed to estimate the true risk, including multifocality and microcarcinomas, in patients who underwent total thyroidectomy after repeat Bethesda III cytology and in the other Bethesda groups.MethodsThree hundred patients participated in the study. Eighty-three (27.7%) of the patients were male and 217 (72.3%) were female. In the study, the diagnostic performances of the R-TIRADS (Revised Thyroid Imaging Reporting and Data System) classification used in the diagnosis of malignancy in thyroid nodules were compared with the Bethesda cytological classification. In order to compare the malignancy prediction powers, receiver operating characteristic curves were created for each system and area under the curve values were calculated. In all statistical analyses, a P-value below 0.05 was interpreted as statistically significant. A total of 150 cases (50.0%) with atypia of unknown significance (AUS) or follicular lesions of unknown significance (FLUS) were retrospectively evaluated.ResultsHistopathology results were consistent with malignancy in 103 patients. Of these patients, 27 had multifocality (26.2%), 26 had microcarcinomas (25.2%), and 22 had both multifocality and microcarcinoma (21.3%). The majority of the carcinomas (46.6%) were seen to arise in AUS with cellular and nuclear atypias.ConclusionPrevalence of Bethesda category III cytologies is significantly higher than reported. A high rate of multifocality and incidental microcarcinomas in the definitive histopathology reports shows that the Bethesda classification falls short in the accuracy of risk in AUS/FLUS. INTRODUCTION Thyroid nodules are very common all over the world. The incidence with ultrasound (US) is 68% [1]. It is more common in female than male with increasing age [2]. Due to the risk of malignancy, care must be taken during the management of thyroid nodules. US-guided fine-needle aspiration (FNA) has become the gold standard for evaluation of thyroid nodules and is now an initial test for detecting thyroid cancer [3]. After FNA, pathologic specimens are evaluated according to The Bethesda System for Reporting Thyroid Cytopathology (TBSRTC), which was developed in 2007 and revised in 2017 [4]. This system has improved the quality of FNA reporting by standardizing the results into 6 definitive diagnostic categories and reducing the overall rate of further unnecessary thyroid interventions. While 60%–70% of biopsied nodules are evaluated as benign and 10% as malignant, the remaining 30% classified as Bethesda category III cannot be diagnosed based on cytological features alone. Bethesda category III including a follicular lesion of undetermined significance (FLUS) and an atypia of undetermined significance (AUS) is a heterogeneous class that fits most of these undefined cases. In the beginning, this category was designed and used by many cytopathologists as a last resort, with only 7%–10% of the FNAs receiving this diagnosis [56]. It has a malignancy rate of approximately 5%–15% depending on the type of atypia detected in the aspirate according to recent studies [578]. The cytological findings of AUS/FLUS are not sufficient to categorize the lesion as benign or malignant, thus creating a gray zone. However, some recent reports have shown higher malignancy rates associated with AUS/FLUS as in our study (32%), suggesting different interpretations among cytopathologists, endocrinologists, and surgeons [91011]. According to a classification by the World Health Organization, a thyroid carcinoma smaller than 10 mm in diameter is called as a ‘microcarcinoma’ and a papillary thyroid carcinoma smaller than 10 mm in diameter as a ‘papillary thyroid microcarcinoma (PTMC).’ There are few reports evaluating the use of a US-FNA biopsy for thyroid nodules smaller than 10 mm in diameter [12]. In the present study, we aimed to investigate the malignancy rate specifically in patients with thyroid nodules reported as Bethesda category III (AUS/FLUS) on both initial and repeat FNA and in other Bethesda groups, and to correlate the results with US features and the final histopathological report. METHODS Study population This retrospective study included 395 patients who underwent total thyroidectomy due to multinodular goiter causing pressure and difficulty during swallowing, and whose cytological examinations were in Bethesda categories I, II, III, IV, and V groups between April 1, 2019 and April 1, 2025. Ninety-five patients in Bethesda group I were not included in the statistical calculations because no malignancy was detected. A total of 395 patients were initially evaluated. Ninety-five Bethesda I cases were excluded because no malignancy was detected histologically or during follow-up, which would have diluted diagnostic accuracy metrics. If included, the malignancy rate would be 26.0%, but the AUC values for Revised Thyroid Imaging Reporting and Data System (R-TIRADS; 0.79) and Bethesda (0.70) would remain unchanged. Surgical indications were standardized across the Bethesda/R-TIRADS categories based on compressive symptoms (dysphagia, dyspnea, cosmetic enlargement), ultrasonographic suspicion (R-TIRADS ≥4a), or patient preference after multidisciplinary discussion, ensuring a uniform surgical policy across all diagnostic strata. The remaining 300 patients were included in the pathology and radiology databases. All data were retrospectively queried with Bethesda III thyroid nodules that underwent a second FNA verification. The data were accessed for research purposes starting on April 2, 2025. We did not include patients with solitary thyroid nodules in the study to avoid confusion regarding surgical indications and to indirectly emphasize the importance of FNA examination in these patients. Data collection All patients gave informed written consent for the use of data from their medical records in the research. Parents signed the consent form and clinical trial form for our 17-year-old patient. Demographics, surgical outcome, cyto- and histopathologic correlations, malignancy rates, and histopathologic subtypes of carcinomas were recorded in Microsoft Excel 2017. US findings with R-TIRADS scores and pathology records were reviewed for each case. All medical records were fully anonymized before we accessed them. The authors have access to the data. Ultrasound evaluation In radiologic examination, US is commonly used to assess thyroid nodules [13]. All thyroid nodules were evaluated preoperatively by high-resolution ultrasonography using a linear transducer (LOGIQ P9, 6–15 MHz; GE Healthcare). Horvath and colleagues established the first Thyroid Imaging Reporting and Data System (TIRADS) in 2009, based on US features of thyroid nodules and, as of 2020, there are 8 versions of TIRADS worldwide [14]. Each risk stratification system has been recommended for different racial populations. The literature has shown that TIRADS versions achieve similar results in assessing the risk of malignancy of thyroid nodules [1516]. In this study, all nodules were evaluated according to the R-TIRADS. For statistical comparison, the TR4 category was subdivided into TR4a (moderate suspicion) and TR4b (high intermediate suspicion) following previously published analytic adaptations used to improve risk stratification. Pathologic evaluation US-guided FNAs were performed by the same interventional radiologist in our radiology department using a 20- to 23-gauge needle that allowed multidirectional passages through the nodule. All samples were immediately immersed into 10% buffered formalin solution for fixation. The cytological results were evaluated as per the Bethesda classification system (I. Nondiagnostic, II. Benign, III. Atypia/FLUS, IV. Follicular neoplasm or suspicious for follicular neoplasm [FLUS], VI. Suspicious for malignancy, V. Malignancy). Routine diagnostic criteria for interpretation of the FNA results were used [1718]. All patients categorized as Bethesda category III in their first FNA report were re-biopsied to confirm the diagnosis, and all patients underwent a total thyroidectomy procedure according to our clinic’s protocol and were operated on by the same endocrine surgeons. For patients with multiple nodules, only the dominant or most suspicious nodule (the one undergoing FNA) was included for analysis. Accordingly, all diagnostic performance metrics were calculated on a per-nodule basis (n = 300), ensuring direct correspondence between cytologic, radiologic, and histopathologic data. Surgical procedure All patients were hospitalized on the day of surgery. A standard transverse cervical incision 2 cm above the sternal notch and just below the cricoid cartilage was used to explore the thyroid gland. Intraoperative nerve monitoring (Nim Response 3.0, Medtronic) and the Harmonic Focus shears instrument (Ethicon) were used in all cases. Standard total thyroidectomy was performed after identification and electrical stimulation of the recurrent laryngeal nerve bilaterally. They were discharged on the first postoperative day and patients with a final pathology report showing malignancy were referred to nuclear medicine department for further treatment. Statistical analysis All statistical analyses were performed using R software ver. 4.4.1 (R Foundation for Statistical Computing) and JASP (ver. 0.19.0, University of Amsterdam). The normality assumption for quantitative variables was tested with the Kolmogorov-Smirnov and Shapiro-Wilk tests. Descriptive statistics of variables are given as mean ± standard deviation and median (range), and frequencies are given as number (%). Two-group comparisons were made using the independent t-test and Mann-Whitney U-tests depending on the assumptions. Homogeneity of variances was tested with the Levene homogeneity test. Categorical data were analyzed using the chi-square test and Fisher-Freeman-Halton exact test, considering the number of categories and expected values. Univariate logistic regression analysis was applied to determine the effective risk factors in determining malignancy. In this study, the diagnostic performances of the R-TIRADS classification used in the diagnosis of malignancy in thyroid nodules were compared with the Bethesda cytological classification. The cytological evaluation results of the cases included in the study were grouped according to the Bethesda system; the TIRADS score was determined by ultrasonographic imaging. Histopathological examination was accepted as the gold standard. In order to compare the malignancy prediction power of the TIRADS and Bethesda systems, receiver operating characteristic (ROC) curves were created for each system and AUC values were calculated. In all statistical analyses, cases with a P-value below 0.05 were interpreted as statistically significant. Ethics and consent All procedures performed in this study involving human participants were in accordance with the ethics standards of the institutional and/or national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. The study was approved by the Ethics Committee of Medipol University in Istanbul, Türkiye (No. E-10840098-202.3.02-4542). Parents signed the consent form and clinical trial form for our 17-year-old patient. All patients gave informed written consent for the use of data from their medical records in the research. This study does not include minors other than one 17-year-old patient. RESULTS Demographic and clinical characteristics Demographic characteristics of the patients and radiological and histopathological comparisons between malignant and benign groups are shown in Table 1. Three hundred patients participated in the study. Eighty-three of the patients (27.7%) were male and 217 (72.3%) were female. There was no statistically significant difference in sex distribution between the 2 groups (P = 0.483). The median age was 43.50 years, the youngest patient was 17 years old, and the oldest patient was 69 years old. The mean age of the patients was 44.46 ± 11.35 years. Although the mean age of the benign group was higher than that of the malignant group, this difference between the groups was not statistically significant (P = 0.401). The nodule diameter was significantly smaller in the malignant group (P = 0.006) (Table 1). Multifocality and microcarcinoma Multifocality was observed in 47.6% of the malignant cases, but not in any of the benign cases (P < 0.001). Similarly, microcarcinoma was detected in 46.6% of the malignant group and not detected in the benign group (P < 0.001) (Table 1). In the Bethesda III group, the tumor was multifocal in 11 patients, only microcarcinoma in 18, and multifocal microcarcinoma in 12. In the remaining 7 patients, the tumor was <2 cm and within the thyroid tissue. In the Bethesda II group, the tumor was multifocal in 4 patients, microcarcinoma in 7 patients, and both multifocal and microcarcinoma tumors were present in 7 patients. In contrast, in the Bethesda IV–V groups, the tumor was multifocal in 12 patients, microcarcinoma in 1 patient, and multifocal + microcarcinoma in 3 patients. Distribution of R-TIRADS classification According to R-TIRADS classification, 48 of 103 malignant cases (46.6%) were in the TIRADS 3 group, 3 (2.91%) in the TIRADS 2 group, 17 (16.5%) in the TIRADS 4a group, 18 (17.5%) in the TIRADS 4b group, and 17 (16.5%) in the TIRADS 5 group, while 122 of benign cases (61.9%) were in the TIRADS 3 group. This difference was found to be statistically significant (P < 0.001). The distribution of Benign/Malignant according to TIRADS categories is shown in Fig. 1. Cross-tabulation analysis performed on all nodules (n = 300) revealed a statistically strong relationship between TIRADS and Bethesda classifications (P < 0.001) (Table 2). When the distribution of patients according to TIRADS classification was examined, 63 patients (21.0%) were in TIRADS 2, 170 patients (56.7%) in TIRADS 3, 29 patients (9.7%) in TIRADS 4a, 21 patients (7.0%) in TIRADS 4b, and 17 patients (5.7%) in TIRADS 5 category. Distribution of the Bethesda classification As a result of cytological evaluation, 113 patients (37.7%) were in Bethesda II, 150 (50.0%) in Bethesda III, 12 (4.0%) in Bethesda IV, and 25 (8.3%) in Bethesda V class. The majority of patients in the TIRADS 2 group (61.9%) were in the Bethesda III and 21.2% were in the Bethesda II category. No case was diagnosed with Bethesda IV–V in this group. In the TIRADS 3 group, 55.3% of the patients were in the Bethesda III category, and 44.7% were in the Bethesda II category. The distribution in the TIRADS 4a group was as follows; 34.5% was Bethesda II, 27.6% was Bethesda III, and 29.7% was Bethesda IV–V. In the TIRADS 4b group, 35.1% of the patients were in the Bethesda IV–V, 23.8% was Bethesda III, and 14.3% was in the Bethesda II class. In the TIRADS 5 group, 35.1% of the patients were in the Bethesda IV–V and 23.5% was in the Bethesda III category. The relationship between TIRADS scores and Bethesda cytological classifications was found to be statistically significant (P < 0.001). The findings show that low TIRADS scores (TR2–TR3) are generally associated with benign (Bethesda II) or uncertain (Bethesda III) cytological diagnoses, while high TIRADS scores (TR4–TR5) are more often matched with suspicious (Bethesda IV) or malignant (Bethesda V) cytological findings. This supports the strong agreement between ultrasonographic risk classification and cytological evaluation. The fact that there were no cases in the Bethesda II category in the TIRADS 5 group indicates that cytological accuracy is high in patients with a preliminary diagnosis of malignancy. Distribution of malignant nodules In the chi-square analysis performed to evaluate the distribution of R-TIRADS and Bethesda classifications of malignant nodules, a statistically significant relationship was found between the variables (P < 0.001). Almost half of the total of 103 malignant nodules were found in the TIRADS 3 category (48 nodules, 46.6%). This was followed by TIRADS 4a (17 nodules, 16.5%), TIRADS 4b (18 nodules, 17.5%), and TIRADS 5 (17 nodules, 16.5%) categories (Table 3). In general, it was observed that the cytological diagnoses of malignant nodules shifted towards higher Bethesda categories as TIRADS scores increased. It was determined that the cytological results of malignant nodules in the low-risk TIRADS categories (TIRADS 2 and 3) were mostly in the lower-level Bethesda categories (II–III). In particular, the cytology results of 48 nodules in the TIRADS 3 group were reported as Bethesda III (type of undetermined significance or follicular lesion), while histopathology results were evaluated as compatible with malignancy. In addition, cytology was detected as benign (Bethesda II) in all 3 malignant nodules in the TIRADS 2 category. When evaluated in general, the most frequently observed TIRADS–Bethesda match in malignant cases was seen to be the TIRADS 4–5 and Bethesda IV–V combination. In malignant nodules with high TIRADS scores, cytological evaluation was generally found to be compatible with higher-level Bethesda categories. In TIRADS 4 and 5 nodules in the intermediate-high risk group, cytological results were also distributed in higher Bethesda categories. While more than half of the malignant nodules were reported as Bethesda IV–V in TIRADS 5 (13/17, 76.4%), this rate was found to be 57.1% (20/35) in the TIRADS 4a–4b group. In the study, the effectiveness of TIRADS scores in predicting the risk of malignancy in thyroid nodules was evaluated with univariate logistic regression analysis. The analysis results are summarized in Table 3, and the graphical probability estimates are summarized in Fig. 2. When the model outputs were examined, the relative risk of malignancy in cases with a TIRADS score of 3 was significantly increased compared to the reference TIRADS 2 group (OR, 7.724; 95% CI, 2.305–25.883; P < 0.001). Similarly, an increased risk was observed for TIRADS 4. In particular, the probability of malignancy was found to be 46.667 times (95% CI, 11.260–193.401; P < 0.001) higher in TIRADS 4a coded nodules and 177.333 times (95% CI, 27.514–1,142.941; P < 0.001) higher in TIRADS 4b coded nodules. These findings show that the risk of malignancy increases significantly with the increase in TIRADS score. The classifier performance of the model was also found to be strong. The AUC value was calculated as 0.802; the sensitivity of the model was found to be 52.3%, the specificity was 94.5%, the positive predictive value (PPV) was found to be 84.8%, and the negative predictive value (NPV) was found to be 77.1%. These data reveal that the model has a high success rate, especially in excluding benign cases. The distribution of 103 malignant nodules according to both TIRADS and Bethesda classification is given in Table 3. The presence of 48 malignant nodules in the TIRADS 3 class shows that this group contains a high rate of malignancy despite appearing to be low risk in terms of visual assessment. Similarly, the fact that nodules in Bethesda II and III categories contain malignancy reveals the weakness of these categories in terms of specificity. This distribution is also consistent with the results of ROC analysis and supports the superior performance of the TIRADS system in distinguishing malignancy. The discrimination levels of TIRADS and the Bethesda systems in the diagnosis of malignancy are compared in the ROC curve given in Fig. 3. The AUC value of the R-TIRADS system is 0.79, and the AUC value of the Bethesda system is calculated as 0.70. The graph clearly shows that the diagnostic success of the TIRADS system is higher. The distribution of a total of 103 histopathologically malignant nodules included in the study according to the R-TIRADS classification is as follows: 3 nodules (2.9%) TIRADS 2; 48 nodules (46.6%) TIRADS 3; 17 nodules (16.5%) TIRADS 4a; 18 nodules (17.5%) TIRADS 4b; and 17 nodules (16.5%) TIRADS 5. The distribution of the same nodules according to the Bethesda classification is as follows: 22 nodules (21.4%) Bethesda II; 48 nodules (46.6%) Bethesda III; and 33 nodules (32.0%) Bethesda IV–V. According to the ROC analysis results, the diagnostic performance criteria of the systems according to the threshold values of TIRADS ≥4a, Bethesda ≥III, and Bethesda ≥IV are presented in Table 4. It was observed that the TIRADS system has higher sensitivity and AUC values and offers a more balanced diagnostic profile compared to the Bethesda system. The Bethesda system shows higher specificity and PPV. In the analysis performed for the TIRADS ≥4a threshold value, the sensitivity of the diagnostic test was found to be 52.0% and the specificity was found to be 94.0%. The PPV was calculated as 84.0%, the NPV as 78.0%, and the overall accuracy rate as 78.0%. At this threshold level, 52 cases (17.3%) were classified as true positive, 15 cases (5.0%) as false positive, 182 cases (60.7%) as true negative, and 51 cases (17.0%) as false negative. These findings show that the TIRADS system has high discrimination, especially in terms of specificity. In the evaluation performed based on the Bethesda ≥III category, the sensitivity of the system was found to be 78.6%, the specificity was found to be 46.2%, the PPV was found to be 43.3%, the NPV was found to be 80.5%, and the overall accuracy rate was found to be 57.3%. At this threshold, 81 cases (27.0%) were determined as true positive, 106 cases (35.3%) as false positive, 91 cases (30.3%) as true negative, and 22 cases (7.3%) as false negative. Despite the high sensitivity, the low specificity level shows that Bethesda ≥III is insufficient to exclude malignancy but can be used as a sensitive screening tool. In the analysis performed according to the Bethesda ≥IV classification, sensitivity was calculated as 32.0%, specificity as 98.0%, PPV as 89.2%, NPV as 73.4%, and overall accuracy rate as 75.3%. In this classification, 33 cases (11.0%) were evaluated as true positive, 4 cases (1.3%) as false positive, 193 cases (64.3%) as true negative, and 70 cases (23.3%) as false negative. These results show that the Bethesda ≥IV category is reliable, especially in positive diagnosis with high specificity and PPV values, but many malignant cases may be missed due to its low sensitivity. These results show that the TIRADS system provides higher diagnostic accuracy, especially in nodules evaluated as gray zones. Our postoperative complications were few in number. Fifteen patients had transient hypocalcemia (5%) and 12 had transient vocal cord paralysis (4%), after surgery. There was no other important morbidity, except for surgical site infections seen in 3 patients (1%). Interpretation of pathology results Our pathology results were evaluated in detail. The median diameter of benign and malignant nodules was 2.5 cm (range, 0.6–5.0 cm) and 1.80 cm (range, 0.50–4.0 cm), respectively (P < 0.05). There were 103 carcinomas detected out of 300 patients (34.3%). According to the Bethesda classification, 22 of the malignant cases were in Bethesda II, 48 were in Bethesda III, and 33 were in Bethesda IV–V groups (Table 3). When the relationship between TIRADS classification and cytological subgroups (AUS and FLUS) was examined, the most frequent cases were observed in the TIRADS 3 category in both groups. Of the 48 patients, 31 had AUS and 17 had FLUS. While 67.7% of the cases (n = 21) were in the TIRADS 3 category in the AUS group, this rate was determined as 82.4% (n = 14) in the FLUS group. In the TIRADS 4 category, 19.4% of distribution (n = 6) was observed in the AUS group and 17.6% (n = 3) in the FLUS group. In TIRADS 5, which is a higher risk category, only 12.9% of nodules (n = 4) were detected in the AUS group. No statistically significant difference was found between the AUS and FLUS groups in terms of distribution according to TIRADS classification (P = 0.432) (Table 5). When the histopathological diagnoses of malignant cases were examined, the most common subtype was papillary carcinoma-classic variant with a rate of 69.9% (n = 72) (Table 6). This was followed by papillary carcinoma-follicular variant with 19.4% (n = 20). Hürthle cell variant was detected at 4.9% (n = 5), while follicular carcinoma and papillary carcinoma-follicular + classic variant were detected at equal rates (2.9%, n = 3). Cytology-histopathology discordance We reevaluated cases with cytology and histopathology discordance in the Bethesda II and III groups (Fig. 4). FLUS was defined as a small follicular pattern with poorly attached cells, and a small amount of colloid, and AUS was characterized by the presence of cellular and nuclear atypias, regardless of the distribution pattern, which could be non-characteristic, follicular, or papillary pattern. The majority of carcinomas were seen to arise in AUS with cellular and nuclear atypias. Follow-up of our patients continues without any problem. DISCUSSION In detecting thyroid nodules, US examination is the first option. During US examination, if nodules have solidity, marked hypoechogenicity, microcalcifications, a taller-than-wide shape, increased blood supply, spicular appearance, irregular margins, metastatic lymph nodes, and capsule invasion, biopsy is indicated because of these signs of malignancy. R-TIRADS was correlated with Bethesda system. Bethesda category III is the most heterogeneous category on TBSRTC. Bethesda category III diagnosis in the initial FNA is very difficult, thus requiring either repeat FNA or molecular testing to confirm the diagnosis, and planning further treatment [9192021]. In our study, we repeated FNA only for the Bethesda category III. Although FNA has been reported to have a sensitivity of 65%–99% and a specificity of 72%–100%, in practice it can diagnose 30%–40% of nodules as AUS (nuclear atypia in a follicular or non-follicular pattern)/FLUS (follicular lesion without nuclear atypia) [22]. Despite a second FNA, making treatment decisions is difficult for clinicians concerned with the optimal follow-up periods or oncologic risk necessitating the surgical management [1123]. Therefore, in 2023, clinical management algorithms were revised for each category, taking into account the risk of malignancy by adding new data to the decisions made in 2017. For different risks of malignancy, molecular tests are provided to detect thyroid tumor-specific mutations or gene expression profiles to rule out malignancy. The risk of malignancy increases gradually in the Bethesda classification. It is <3% in the Bethesda II group, 13%–30% in Bethesda III, and 74% in Bethesda V. Several multi-gene expression tests have been proposed to overcome these problems. In papillary thyroid cancer, BRAF mutations can be seen in 29%–69%, RET rearrangement in 13%–43%, NTRK1 rearrangement in 5%–13%, and RAS mutations in 0%–21% [24]. While RAS mutations are more likely to be seen in noninvasive follicular thyroid neoplasm with papillary-like nuclear features and invasive encapsulated follicular variants of papillary cancer, BRAF mutations are more common in infiltrating follicular variants of papillary cancer [24]. In follicular thyroid carcinoma, RAS mutations can be seen in 40%–53% and peroxisome proliferatoractivated receptor gamma rearrangements in 25%–63% [24]. However, the high cost, their unavailability in most countries, and questionable diagnostic accuracy have not completely prevented the search for other solutions [252627]. Recent studies have attempted to estimate the risk of malignancy by combining the evaluation of cytological results and suspicious sonographic findings [528]. We did not perform molecular testing on our patients because it was very expensive. We performed a second FNA only in the Bethesda III group. We evaluated ultrasonographic and cytological findings. Gweon et al. [29] have evaluated the feasibility and the role of thyroid US in predicting malignancy for Bethesda III nodules, and they suggested that when the US assessment for the thyroid nodules is suspicious for malignancy, repeat FNA may be unnecessary and surgery should be considered. Yoo et al. [30] categorized AUS/FLUS nodules using the US risk stratification system to calculate the malignancy rate and identify the patients most suitable for surgical treatment. In their study, the malignancy rates were found to be 60%–67% for the high-suspicion category, 32%–36% for the intermediate-suspicion category, and 12%–16% for the low-suspicion category. The overall malignancy rate for thyroid nodules with AUS/FLUS was 38.7%. We found a similar rate in our Bethesda III category (32%). Additionally, the malignancy rate in our study was 19.4% for the Bethesda II and 89.1% for the Bethesda IV–V group. The discrimination levels of R-TIRADS and the Bethesda systems in the diagnosis of malignancy are compared in the ROC curve given in Fig. 3. The AUC value of the TIRADS system is 0.79, and the AUC value of the Bethesda system is calculated as 0.70. The graph clearly shows that the diagnostic success of the TIRADS system is higher. Although TBSRTC has improved the quality of FNA reporting, Bethesda category III has generated a new controversy due to its heterogeneous pattern. At first, this category was designed as a last resort to include only 7%–10% of the FNAs with a roughly 5%–15% malignancy risk [48]. However, in recent publications, higher malignancy rates (over 30%) associated with AUS/FLUS have been reported, similar to the findings in our study [9101131]. Ho et al. [31] presented a range of the true prevalence of malignancy, lying between a lower-bound estimate of 26%, which included all AUS/FLUS nodules (assuming all observed nodules were benign), and an upper-bound estimate of 37% risk of malignancy, which was calculated based only on the AUS/FLUS nodules selected to undergo surgery after initial or repeated Bethesda III cytology. In other studies, the risk of malignancy was found to be 35%–59% [1132]. In the present study, the ratio of our patients with the Bethesda III category was 37.9% and approximately 4 times more than the expected 7%–10% of all Bethesda categories (I–VI). Moreover, all patients in the Bethesda III group had a repeat FNA confirming AUS/FLUS, and all suspicious nodules revealed by US have already been biopsied with inconclusive results. A total of 46.6% of malignant nodules were radiologically located in TIRADS 3, 16.5% in TIRADS 4a and TIRADS 5, and 17.5% in TIRADS 4b. It is observed that the risk of malignancy increases significantly with increasing TIRADS score (Table 3). Several large studies with more than 500 FNAs reported a high rate of malignancy for the AUS/FLUS category, as in our study, raising the question of whether the Bethesda system should be revised specifically for the categories of indeterminate thyroid lesions, as the Bethesda III category includes a variety of heterogeneous cases that are neither clearly benign nor sufficiently atypical to be included in the other categories. Recently, several authors have divided Bethesda III nodules into those with low cellularity with a microfollicular pattern and those showing nuclear atypia, with some studies suggesting that nuclear atypia is an independent risk factor [933]. Kelman et al. [34] found that up to 60% of thyroid nodules with nuclear atypia were related to malignant disease. Similarly, Shrestha and Hennessey [35] showed a 36.8% malignancy rate for patients with nuclear atypia compared to the rate of 10.8% in those without nuclear atypia. Another recent study analyzing results from second FNAs in patients with a prior Bethesda III diagnosis showed a malignancy rate of 41.5% in patients with nuclear atypia, while those with architectural atypia had a malignancy rate of 15.5% [36]. Similarly, in our study, the majority of carcinomas were seen to arise in AUS with cellular and nuclear atypias (64.5%). We have shared pathology images of 2 of our cases here to illustrate and discuss the interpretation and sampling difficulties in the Bethesda Thyroid Cytopathology Reporting System. In the first case, the cytological diagnosis was Bethesda category III (AUS/FLUS), based on a small suspicious cell cluster observed in the cell block. However, the area was not seen in deeper sections, preventing further immunohistochemical evaluation. Given the limited material and absence of definitive malignant features, the AUS/FLUS classification was considered appropriate. The final histopathological diagnosis was PTMC. Although the outcome was malignant, the cytological findings had already raised concern, and the initial interpretation reflected the diagnostic uncertainty due to insufficient cellularity and subtle atypia. Since the sample limitations were noted in the original report, there was no significant discordance between the cytologic and histologic findings. The second patient, however, was cytologically classified as category II (Benign), with features suggesting Hashimoto thyroiditis—such as lymphocyte-rich follicular cells, Hürthle cell change, and focal reactive atypia. No signs of malignancy were identified, but the report advised repeat aspiration if clinical suspicion remained. Following surgery, histology revealed PTMC, indicating a false-negative cytologic result. This case highlights how a Hashimoto background and Hürthle cells can obscure or mimic neoplastic features, complicating cytologic interpretation. The recommendation for follow-up suggests that the cytopathologist was aware of potential limitations in the aspirate. Together, these cases show the gray areas and limitations of the Bethesda system. Issues like limited sample quality or background inflammation with Hürthle cells can affect diagnostic accuracy. While category III already includes some uncertainty, even category II can hide malignancy, despite its estimated risk being only 0%–3% [18]. This highlights the importance of combining cytology with clinical and radiologic findings, especially when suspicion persists. This study is a retrospective cross-sectional diagnostic accuracy study. Potential sources of bias include selection bias due to including only surgically confirmed multinodular cases and confirmation bias due to analyzing only nodules confirmed by histopathology. Furthermore, the study’s retrospective design may limit observer variability in imaging and cytology classifications. These methodological limitations are clearly noted in the discussion section. In conclusion, this study particularly highlights the need to question Bethesda category III, as most malignancies are diagnosed as microcarcinomas at an earlier stage and half are multifocal. The high rate of multifocality and incidental microcarcinomas in definitive histopathology reports and the fact that most cancers originate from AUS atypia (64.5% in our series) may influence the surgical decision toward total thyroidectomy. In our opinion, the Bethesda category III needs an updated revision, and a new international algorithm should be developed to use less invasive protocols.