{"paper_id":"b494b86c-8481-4d92-b60d-0cc85fd817f1","body_text":"Ovarian cancer has the worst prognosis among all gynaecological cancers and is often diagnosed at advanced, less curable stages (stage 3 or 4). It carries significant morbidity and mortality with the 5-year survival for stage 4 disease being 13.4% [ 1 ]. The most common histological subtype is epithelial ovarian cancer and these account for approximately 80–85% of all ovarian cancers [ 2 ]. When the disease is diagnosed at stage 1, the 5-year survival rate is greater than 90%, however a standardized screening tool for ovarian cancer that significantly reduces death from the disease does not currently exist within the UK [ 1 ,  3 ]. It is therefore essential that there is accurate pre-operative characterisation of ovarian tumours to improve patient’s outcomes [ 4 ].\nAccurate assessment of ovarian masses by ultrasound is of great importance. It is often the first line imaging modality to assess adnexal masses and subjective assessment by expert ultrasound examiners has demonstrated excellent results in distinguishing between benign and malignant masses [ 4 – 7 ]. The diagnostic accuracy of adnexal masses is operator dependent, however there are several prediction algorithms, with excellent results in practice to aid with triaging and diagnosis [ 8 – 11 ].\nAlthough Magnetic Resonance Imaging (MRI) is a more expensive and not as readily available as ultrasound, it can add further information to those ovarian masses that are defined indeterminate by ultrasound [ 12 ,  13 ]. In previous studies, MRI has been shown to correctly identify malignant ovarian masses with an accuracy of 83–93% [ 14 – 16 ]. Different MRI sequences can add a different element of characterisation of ovarian masses, for example T2 weighted imaging has high soft tissue contrast resolution which can help in teratoma diagnosis. Scoring systems can further aid characterisation [ 17 ]. The ovarian adnexal reporting and data system (ORADS) MRI scoring system is frequently used as a malignancy risk stratification system with a sensitivity of 93.5% and specificity of 96.6% [ 18 ].\nThere are however continued challenges to the pre-operative diagnosis of ovarian masses and frozen sections can help to intraoperatively determine the malignant potential of an ovarian mass [ 19 ]. They can help determine and guide the operating surgeon on the extent of a surgical procedure including staging procedures [ 20 ]. This can avoid overtreatment in patients, which is of particular importance in cases of fertility preservation and can also avoid extensive surgical staging procedures which can carry a higher risk of visceral and vascular injury 13 .\nFrozen section diagnosis is considered a reliable diagnostic method for assessing ovarian tumours with sensitivities reported between 90–96% [ 19 ,  21 ,  22 ].\nThe purpose of this retrospective study was to compare the diagnostic performance of ultrasound, MRI, and frozen section in the diagnosis of ovarian masses relative to the final histology of the masses.\n\nWe conducted a retrospective observational study, between January 2018 to December 2021 and cases with known frozen section and final histology at Guy’s and St Thomas hospital, London were selected. All decisions for frozen section were made and discussed at our gynaeoncology multidisciplinary team (MDT) meeting. The decision for frozen section was made in cases where there was thought to be a possibility that there could be overtreatment due to unnecessary interventions or undertreatment that would require further treatment based on imaging alone.\nThe age and parity of all the patients were recorded. The carbohydrate antigen- 125 (ca-125) tumour marker data and the stage of the tumour was collected for the borderline and malignant cases.\nAll the specimens were processed and diagnosed by a consultant gynaecological pathologist; however, this was not the same pathologist for all cases. The results were reported back to the operating surgeon during the operation within 30 min of receiving the specimen.\nAll frozen section diagnosis were categorized as benign, borderline, and malignant, with the specific histology commented on if possible.\nThe MRI and US data that were performed at our tertiary gynaeoncology centre were used. Pre-operative MRI diagnosis was undertaken in 129 cases and ultrasound diagnosis in 63 cases.\nThe US, MRI and frozen section diagnosis was compared with the final paraffin histological section diagnosis in all cases.\nFor the US cases, the level of sonographer who performed the scan was recorded and grouped into level 1–3 examiners [ 23 ].\nApproval of the study as a service improvement was granted by the audit department at Guys and St Thomas NHS trust. Formal ethical approval was not required.\nStatistical analysis was performed using Stata MP v17.0 software (USA, 2023) [ 24 ]. To calculate the diagnostic performances of US, MRI, and frozen section in determining the benign, borderline and malignant status of an ovarian mass, performance measures including sensitivity, specificity, positive and negative likelihood rations were calculated according to the standard 2×2 method. Agreement was assessed using the Kappa co-efficient. The 95% confidence intervals were calculated for the sensitivities, specificities, and positive and negative predictive values.\n\nFrom January 2017 – December 2021, a total of one hundred and fifty-six ovarian masses were examined by frozen section and by final histopathological examination. The median age of the patients in the study was 51, range 62 (80-18). The parity of the patients were varied with 56/156 nulliparous and 75/156 multiparous. Parity was not stated in 25/156.\nIn the final histopathological examination, 123/156 (78.8%) of tumours were epithelial tumours. Out of these cases, 55 were benign, 38 were borderline and 30 were malignant.\nSex cord stromal tumours accounted for 20/156 (12.8%) of tumours and 9 were benign and 11 were malignant. Benign germ cell tumours accounted for 10/156 (6.4%) of tumours. There were 3 tumours which were metastases (1 appendiceal, 2 colorectal). All the final histopathological examinations can be seen in Table  1 . Table 1 Distribution of all histological types according to final diagnosis. Benign epithelial tumours Simple serous cyst 1 Serous cystadenoma 7 Serous cystadenofibroma 9 Serous fibroadenomas 2 Mucinous cystadenoma 21 Mucinous cystadenofibroma 1 Mucinous fibroadenoma 1 Mucinous cystadenoma and Brenner tumour 2 Seromucinous cystadenofibroma 2 Endometrioma 3 Endometriotic cystadenofibroma 1 Benign epithelial inclusion cyst 1 Benign corpus luteal cyst 2 Benign cyst, unclassified 1 Endometriosis with mucinous metaplasia 1 Borderline epithelial tumours Serous borderline tumour 17 Seromucinous borderline tumour 6 Mucinous borderline tumour 13 Borderline mucinous tumour with benign Brenner 2 Malignant epithelial tumours High grade serous carcinoma 7 Seromucinous carcinoma 1 Mucinous carcinoma 9 Mucinous carcinoma with serous mucinous features 1 Endometrioid ovarian ca 5 Clear cell carcinoma 6 Intra-epithelial carcinoma arising within borderline mucinous tumour 1 Benign Sex cord stromal tumours Fibroma 5 Fibrothecoma 3 Thecoma 1 Malignant sex cord stromal tumours Adult type granulosa cell tumour 9 Leydig cell tumour 1 Sertoli - leydig cell tumour 1 Benign germ cell tumours Mature cystic teratoma 4 Struma ovarii 5 Mature cystic teratoma with benign mucinous cystadenoma 1 Mestatases Appendiceal 1 Colorectal 2\nDistribution of all histological types according to final diagnosis.\nFor the 38 borderline cases, 14 of the cases did not have pre-operative ca-125 data. Of the remaining 24 cases, the median value for ca-125 was 15 u/ml and interquartile range was 38 (48-10). The Federation of Gynaecology and Obstetrics (FIGO) staging for the borderline tumours were as follows: stage 1 A were 26 cases, stage 1B were 2 cases, stage 1 C were 7 cases, stage 2 was 1 case, stage 3 were 2 cases and there were no cases for stage 4 disease.\nFor the 44 malignant cases, 11 of the cases did not have pre-operative ca-125 data. Of the remaining 33 cases, the median value for ca-125 was 44 u/ml and the interquartile range was 73 (85.5-12.5). The FIGO staging for the malignant cases were as follows: Stage 1 A were 21 cases, stage 1B was 1 case, stage 1 C were 8 cases, stage 2 were 3 cases, stage 3 were 7 cases and stage 4 was 1 case. 3 cases were metastases from other organs and therefore were not given FIGO staging.\nPre-operative ultrasound diagnosis was made in 63/156 patients (40.4%). Level 1 sonographers performed 41/63 (65.1%) of cases, Level 2 sonographers 8/63 (12.6%) cases and level 3 sonographers 14/63 (22.2%) cases. Ultrasound had an overall sensitivity and specificity of 95.2% (95% CI: 76.2–99.9) and 61.9% (95% CI: 45.6–76.4) for benign masses, 20% (95% CI: 4.33–48.1) and 79.2% (95% CI: 65–89.5) for borderline masses and 57.1% (95% CI: 28.9- 82.3) and 87.8% (95% CI: 75.2–95.4) for malignant masses (Table  2 ). Ultrasound showed poor agreement with histological cell type interpretation with a kappa coefficient of 0.29. Table 2 Diagnostic performance of ultrasound, MRI and Frozen Section. Sensitivity (% (95% CI)) Specificity (% (95% CI)) Positive Predictive Value, PPV (95% CI) Negative Predictive Value, NPV (95% CI) Ultrasound diagnosis Benign 95.2 (76.2–99.9) 61.9 (45.6–76.4) 55.6 (38.1–72.1) 96.3 (81–99.9) Borderline 20 (4.33–48.1) 79.2 (65–89.5) 23.1 (5.04–53.8) 76 (61.8–86.9) Malignant 57.1 (28.9–82.3) 87.8 (75.2–95.4) 57.1 (28.9–82.3) 87.8 (75.2–95.4) MRI diagnosis Benign 100 (80.5–100) 61.3 (51.5–70.4) 29.5 (18.5–42.6) 100 (94.7–100) Borderline 31.5 (19.5–45.6) 81.3 (70.7–89.4) 54.8 (36–72.7) 62.2 (51.9–71.8) Malignant 61.5 (44.6–76.6) 85.6 (76.6–92.1) 64.9 (47.5–79.8) 83.7 (74.5–90.6) Frozen section diagnosis Benign 90.8 (81.9–96.2) 93.8 (86–97.9) 93.2 (84.9–97.8) 91.5 (83.2–96.5) Borderline 86.8 (71.9–95.6) 97.5 (92.7–99.5) 91.7 (77.5–98.2) 95.8 (90.5–98.6) Malignant 97.6 (87.4–99.9) 95.6 (90.1–98.6) 89.1 (76.4–96.4) 99.1 (95–100)\nDiagnostic performance of ultrasound, MRI and Frozen Section.\nThe interpretation and the final histology of the 14 ovarian tumours that were classified by level 3 sonographers are shown in Table  3 . There were 13/63 cases that were classified as uncertain on ultrasound by subjective impression. The histology for these can be seen in Table  4 . Table 3 This table demonstrates the subjective assessment of level 3 sonographers for the 14 ovarian tumours classified by them and the results of the final histology for the 14 ovarian tumours. Ultrasound Final histology Benign Borderline Malignant Benign 4 0 0 Borderline 2 1 2 Malignant 0 2 1 Uncertain 0 1 1 Table 4 Specific histology for ultrasound classified as uncertain by subjective impression. Histology Number (n) Benign  Mature cystic teratoma with benign mucinous cystadenoma 1  Mucinous cystadenoma and Brenner tumour 1  Benign corpus luteal cyst 1  Serous cystadenoma 1 Borderline  Serous borderline tumour 3  Seromucinous borderline tumour 1  Mucinous borderline tumour 1 Malignant  Endometrioid carcinoma of the ovary 1  Granulosa cell tumour 2  Seromucinous carcinoma 1\nThis table demonstrates the subjective assessment of level 3 sonographers for the 14 ovarian tumours classified by them and the results of the final histology for the 14 ovarian tumours.\nSpecific histology for ultrasound classified as uncertain by subjective impression.\nMRI was used to make a pre-operative impression in 129 patients. The sensitivity and specificity for benign tumours was 100% (95% CI: 80.5–100) and 61.3% (95% CI: 51.5–70.4) for benign tumours;31.5% (95% CI: 19.5–45.6) and 81.3% (95% CI: 70.7–89.4) for borderline tumours and for 61.5% (95% CI: 44.6–76.6) and 85.6% (95% CI: 76.6–92.1) for malignant tumours (Table  2 ). MRI showed agreement with histological cell type interpretation with a kappa coefficient of 0.34.\nThere were 18/129 cases that were classified as uncertain on MRI by subjective impression. The histology for these can be seen in Table  5 . Table 5 Specific histology for MRI classified as uncertain by subjective impression. Histology Number (n) Benign  Serous cystadenofibroma 1  Endometriotic cystadenofibroma 1  Serous cystadenoma 1  Serous fibroadenoma 1  Seromucinous cystadenofibroma 1  Serous cystadenofibroma 1  Mature cystic teratoma 1  Mucinous cystadenoma and Brenner tumour 1 Borderline  Serous borderline tumour 1  Seromucinous borderline tumour 2  Mucinous borderline tumour 2 Malignant  Serous carcinoma 2  Mucinous carcinoma 3\nSpecific histology for MRI classified as uncertain by subjective impression.\nFrozen section was conducted on all 156 patients included in the study. Of these, there was discordance between the frozen section diagnosis and the final histopathological in 13/156 (8.3%) cases.\nThe sensitivity and specificity values were as follows: benign tumours:90.8% (95% CI: 81.9–96.2) and 93.8% (95% CI: 86–97.9); borderline tumours: 86.8% (95% CI: 71.9–95.6) and 97.5% (95% CI: 92.7–99.5) and malignant tumours: 97.6% (95% CI: 87.4–99.9) and 95.6% (95% CI: 90.1–98.6) (Table  2 ). Frozen section showed good agreement with histological cell type interpretation with a kappa coefficient of 0.88.\nWhen borderline tumours were classified as malignant, ultrasound showed a sensitivity and specificity for malignant tumours of 58.6% (95%CI:38.9–76.5) and 70.6% (95%CI:52.5–84.9) respectively. The corresponding values for MRI were 62.4% (95% CI: 51.7–72.2) and 72.2 (95% CI: 54.8–85.8); and for frozen section were 93.8% (95% CI: 86–97.9) and 90.8% (95% CI: 81.9–96.2). All figures can be seen in Table  6 . Table 6 Diagnostic performance of each modality when borderline tumours are classed as malignant. Malignant Sensitivity (% (95% CI)) Specificity (% (95% CI)) Positive Predictive Value, PPV (95% CI) Negative Predictive Value, NPV (95% CI) Ultrasound 58.6 (38.9–76.5) 70.6 (52.5–84.9) 63 (42.4–80.6) 66.7 (49–81.4) MRI 62.4 (51.7–72.2) 72.2 (54.8–85.8) 85.3 (74.6–92.7) 42.6 (30–55.9) Frozen Section 93.8 (86–97.9) 90.8 (81.9–96.2) 91.5 (83.2–96.5) 93.2 (84.9–97.8)\nDiagnostic performance of each modality when borderline tumours are classed as malignant.\n\nAccurate characterization of ovarian masses, both pre-operatively with US and MRI and intraoperatively with frozen section is essential for precise management. Frozen section is vital for determining the malignant potential of an ovarian mass in a timely manner and is important for operative planning.\nWe performed diagnostic studies on the accuracy of ultrasound, MRI and frozen sections within our cohort of patients who had an intraoperative frozen section over a 5-year period.\nOur results demonstrate that frozen section is an accurate method to appropriately characterise ovarian masses as benign, borderline and malignant intraoperatively with good diagnostic performance demonstrated throughout.\nMRI and Ultrasound demonstrated good sensitivity for benign tumours, however ultrasound had poor sensitivity for malignant (57.1%) and borderline (20%) ovarian tumours.\nA systematic review in 2011 demonstrated that if FS was benign or malignant, in 94% and 99% of the cases respectively, the histological diagnosis would remain the same, which is consistent with our study findings [ 25 ]. The accuracy of frozen sections for borderline tumours remains a major diagnostic challenge and studies have reported sensitivities as low as 0–50%. In all these studies, mucinous borderline tumours caused the greatest disagreement [ 26 ,  27 ].\nIt has been suggested that mucinous tumours are difficult to diagnose on frozen section as they are larger tumours with a mean tumour size 12 cm [ 28 ,  29 ]. In addition to this, they are more heterogenous masses with different histological subtypes and few areas of invasion in one mass when compared to their serous counterparts [ 30 ]. This would then require multiple samples to reach an accurate diagnosis. Misdiagnosis of borderline tumours is also related to <10% of borderline component and the experience of the pathologists [ 31 ].\nIn more recent studies, the rate of correlation between borderline tumours at frozen section and final histopathological diagnosis was 90.6% and this did not differ among different hospital settings or pathologists with different subspecialities [ 32 ].\nThis is similar to our dataset and could be due to the similarity in the distribution of serous and mucinous borderline tumours in both studies. Mucinous borderline tumours accounted for just under 50% of the borderline tumours analysed in this study.\nIn this study, 7 tumours were reported as benign at FS and later determined to be BOT or malignant. These were 3 borderline mucinous tumours and 4 malignant tumours (an appendiceal metastasis, ovarian intra-epithelial carcinoma, granulosa cell tumour and endometrioid carcinoma). The malignant cases were all stage 1 disease except for the metastases and 2 of these cases required further surgical treatment to stage the disease.\nThere was a false positive in one case. However, this patient was 70 years of age and had concurrent simple endometrial hyperplasia for which an MDT decision was made for a total abdominal hysterectomy, BSO and omental sampling. No other patients received extensive staging surgery unnecessarily.\nMRI is a good preoperative tool for diagnosing ovarian masses and has particular use in determining sonographically indeterminate masses, however, its cost and lower availability prevents it’s use as the primary imaging modality for ovarian tumours [ 33 ,  34 ].\nOur results for the sensitivity of MRI for diagnosis of benign tumours is consistent with the literature [ 35 – 37 ].\nMRI has high contrast resolution and is helpful in detecting fat tissue and haemorrhagic areas as well as good sensitivity at detecting solid components and tumour septations [ 38 ]. This proves essential when assessing indeterminate tumours.\nIn the literature, borderline ovarian tumours are often misclassified on MRI and there is a range of sensitivities reported [ 39 ]. Bazot et al. reported the sensitivity and specificity of MRI for diagnosing borderline tumours as 45.5% and 96.1% respectively [ 37 ]. The sensitivity for serous vs mucinous BOT was 33.3% and 93.3% [ 37 ]. In this study a lower sensitivity was found for borderline and malignant ovarian masses, 31.5% and 63.2% respectively, in our dataset.\nA systematic review has shown that MRI had a sensitivity of 92% and specificity of 85% for detecting borderline and invasive tumours [ 27 ]. For detecting malignancy alone, a sensitivity of 93% has been reported [ 14 – 16 ,  36 ,  40 ].\nThere have been several factors reported that can impact the diagnostic ability of an MRI. Vegetations and multilocularity were two factors could help determine serous from mucinous borderline tumours on MRI [ 41 ]. The size of the tumour is a significant factor in detection of tumours on MRI and it has been shown that there has been difficulty detecting lesions smaller than 2 cm [ 16 ].\nIt has further been reported that certain MRI techniques, such as dynamic and gadolinium contrast enhancement provides better resolution and characterization of these ovarian masses, improving sensitivity for detection [ 16 ,  42 ].\nThe heterogeneity between MRI imaging techniques, level of clinician reporting the scan, distribution of serous vs mucinous borderline tumours within the dataset and size of the tumours present could account for the variation in sensitivity, specificity and accuracy among different studies.\nUltrasound produces high resolution imaging of the ovaries and can distinguish simple ovarian cysts from complex ovarian masses [ 34 ]. Subjective assessment has an excellent sensitivity, as high as 96.7% for distinguishing between benign and malignant ovarian masses [ 7 ,  43 – 45 ]. Our study had good sensitivity for benign tumours but lower than expected sensitivity for malignant tumours.\nOf the 63 cases that had a pre-operative US in our cohort, 14 of these had a malignant histopathological diagnosis. These included clear cell carcinoma, granulosa cell carcinoma and endometrioid carcinoma. In a large study describing the ultrasound features of clear cell carcinomas, it was reported that there was no specific ultrasound pattern for these tumours [ 46 ]. In other studies, granulosa cell tumours have been reported to have multiple sonographic appearances and endometrioid carcinoma are reported to be a rare diagnosis [ 47 ,  48 ]. Given the distribution of malignancies and taking into consideration that all pre-operative ultrasounds were not performed by only level 2 or 3 experts as in previous studies, this most likely accounts for the difference in sensitivities.\nSpecifically, when analysing the data for the level 3 examiners, the confirmed histological borderline and malignant tumours were only classified as borderline and malignant on US by the level 3 examiners. In clinical practice, all borderline and malignant ovarian tumours are triaged to a gynae-oncology cancer centre and so would receive the appropriate care. This is important to highlight as outcomes for ovarian cancer are better for patients when it is provided by trained gynaeoncologists in specialist centres [ 49 ]. Unfortunately, in this study it was difficult to directly compare the specific diagnostic performances at each level due to the small numbers.\nAccurate preoperative ultrasound assessment of borderline tumours is imperative given the impact on patients’ fertility journeys; however, BOTs continue to pose a difficulty and the accuracy of ultrasound diagnosis is reported to be lower than for benign and malignant tumours [ 50 ].\nPrevious studies have reported a sensitivity of 44% and 69% for the subjective impression of BOTs on ultrasound diagnosis and a recent systematic review reports a sensitivity of 66% with a low false positive rate [ 50 – 52 ].\nIn this study, we reported a sensitivity of 20% for the ultrasound subjective assessment of ultrasound diagnosis. In addition to this, 5/13 unclassified tumours were BOTs. This is lower than reported in previous studies and this could be attributed to the expertise level of clinicians performing and interpreting the scans and the heterogeneity of the BOTs included in this study. This data does however support that sensitivity is lower than for benign or malignant tumours.\nTo our knowledge, this is the first study that has compared the diagnostic performance of frozen section, MRI and US in the same cohort of patients, however it did have some limitations.\nIt was retrospective in nature, and this could be considered a weakness of the study. In addition to this, dissimilar from previous studies, the level of clinicians performing and interpreting the pre-operative imaging, in particular US, varied and was not solely level 3 examiners. This could have influenced the subjective impression reported; however results are generalizable to clinical practice where not all patients are scanned by level 3 clinicians.\nWe also acknowledge that there are a smaller number of preoperative ultrasound cases than there are MRI cases in this study. As a tertiary referral gynaeoncology centre, prior to referral to our unit, ultrasounds may be conducted at the patient’s base hospital. In our study, we only included ultrasounds which were performed within our trust. For future research and in comparisons of such data, an equal weighting of each imaging modality would be desirable.\nWe had chosen to report and include cases where the pre-operative imaging was unclassified. In practice, occasionally tumours may be subjectively described and given such diagnosis which had prompted further imaging and investigation.\nTo further improve early diagnosis of ovarian cancer on imaging modalities and improve outcomes for patients, further widespread training on ultrasound features of tumours for all grades of examiners is pivotal. Access to level 3 examiners in all hospital trust can be limited and upskilling all examiners is important.\nFrom a research perspective, larger prospective studies including those comparing subjective impression of ultrasound of level 3 examiners to final histology are required to further build on this data.\nIn addition to this, further research is required to improve the diagnosis accuracy of pre-operative imaging modalities and in the era of increased research on computer assisted diagnosis in the diagnosis of ovarian tumours, this could reduce misdiagnosis and reduce intervention for many patients.\n\nTo conclude, the results from frozen section and MRI remain good and there is still an important place for frozen section among pre-operative imaging. However, the diagnosis of BOTs remains challenging in all modalities.","source_license":"CC-BY-4.0","license_restricted":false}