Clinical data regarding age, gender, duration of MD, vestibular drop attacks, cochlear signs (hearing fluctuations, tinnitus, fullness), vestibular signs (vertigo attacks, average frequency during the last 6 months, duration) were systematically recorded.

Audiometry comprised pure-tone average (PTA, 0.5–4 kHz), and dissyllabic (French Fournier lists) word discrimination score (WDS). Audiometries (AC40^®, Interacoustics, Middelfart, Danmark) were conducted in a standard audiometry booth with a headphone. Electrocochleography (ECoG) was measured by an extratympanic electrode (Elios^®, Echodia^®, Clermont-Ferrand, France) and averaging the response to 1,000 clicks at 85 dB HL. Action (AP) and summation potential (SP) amplitudes were recorded. SP/AP > 0.33 was considered in favor of a hydrops (9). Bithermal caloric vestibular testing vestibular test was performed by water irrigation in a calibrated setting at 30 and 44°C and automatic videonystagmography (Visual Eyes Spectrum^®, Interacoustics, Middelfart, Danmark). For patients with perforated tympanic membrane or tympanostomy tubes, caloric tests were performed with air and not with water in both ears. A caloric weakness was defined by a response asymmetry above 20% based on the maximal slow-phase velocity of the nystagmus according to Jongkees (23).

Cervical vestibular evoked myogenic potentials (cVEMP) were measured by 500 Hz tone bursts at 90–100 dB HL with the rate of 7.1 stimuli/s transmitted via inserted headphones (Eclipse^®, Interacoustics^®). 100 stimuli were applied to each ear. The rise/fall time was 1 ms and plateau time 2 ms. cVEMP amplitudes were corrected for muscular pre-tension of the sternocleidomastoid muscle and patients with an air-bone gap at 500 Hz were excluded from cVEMP measurements. Four electrodes were required for the cVEMP recording: a negative electrode at the forehead, two positive electrodes at the junction of the upper and middle thirds of the sternocleidomastoid muscles and one ground electrode at the sternal manubrium. Contraction of the sternocleidomastoid muscle on the homolateral side was required. The patient was then asked to turn his head to the side opposite to the sound stimulation. P13 et N23 wave amplitudes (A, μV) were recorded and compared to the contralateral side. Asymmetry ratio was calculated by: (A_Unaffected – A_Affected) / (A_Affected + A_Unaffected). An absent response or an asymmetry > 0.35 was considered as a dysfunction (24). Other pathological test values were absent or increased latencies of early waves: a P13 wave appearing after 10 ms and/or an N23 wave arising after 19 ms. Video Head-Impulse test (vHIT, Otometrics^®, Hoerskaetten, Danemark) provided vestibulo-ocular reflex (VOR) gains for each semicircular canal. Gains < 0.8 and the presence of corrective saccades were considered to indicate hypofunction of the respective canal (25). The disease side was defined by the cochlear signs including fluctuating hearing loss, fullness, and tinnitus. MD was considered as bilateral when cochlear signs involved both ears.

High-resolution T2- weighted images (3D SPACE) of both inner ears were obtained in all 76 cases (Siemens 3 Tesla^® SPECTRA) with a head antenna. In addition, 3D-FLAIR images 4H after intravenous injection of gadolinium (0,2 ml/kg of gadoteric acid, Dotarem^® at 0,5 mmol/ml) were available in 68 patients (same day as audiovestibular tests in 54 and a different day in 14 cases). T2-sequence-weighted images had the following characteristics: slice thickness 0.3 mm, 1,200 ms relaxation time, 145 ms echo time, number of excitations: 2, 320x356 Matrix, 120° Flip angle, acquisition time:12,269 ms. The saccule was visualized on the coronal views (Figure 1) and its maximal height and width were measured (Osirix v.5.6^®, Pixmeo^®, Bernex, Suisse). Hydrops was defined by a height > 1.6 mm and/or a height/width ratio < 1.14 with a tendency toward a round saccule (saccule height and width criteria) (11). All 90 ears had a T2-weight CISS sequence. However, the fuzzy contours of the saccule in this sequence did not allow to measure all parameters with certainty. Consequently, 80 or 88 cases depending on the parameter could be reported.

3D-FLAIR images were obtained 4H after gadolinium with the following parameters: slice thickness 0.3 mm, 8,000 ms relaxation time, 498 ms echo time, 2,350 ms inversion time, number of excitations: 2, 320x320 Matrix, variable Flip angle, and 8:58 min acquisition time. On an axial view at the level of the lateral semicircular canal, the entire surface of the vestibular endolymphatic compartment (saccule+utricle) and entire vestibular surface (perilymphatic+endolymphatic compartments) were estimated by manual contouring and the endolymphatic surface/entire vestibule surface was deduced (Figure 2). The contouring was performed on only one slice at the level of the lateral semicircular canal. Vestibular hydrops (VH) was graded as 0 (no hydrops) for a ratio < 0.33, as I (moderate) for ratios 0.33–0.49, and II (significant) for ratios 0.5–1.0 according to Nakashima et al. (10). On a view at the level of modiolus, we graded the cochlear hydrops (CH) as 0 (no hydrops) when no dilatation of the cochlear canal was detected, as I (moderate) when a moderate dilatation of the cochlear canal coexisted with a visible vestibular scala, and as II (significant) when the dilatation of the cochlear canal totally masked the vestibular ramp (axial surface ratio criteria) (17). On the coronal plane, the saccule was measured similarly to T2-view analysis, and the same diagnostic criteria were applied. Finally, on the sagittal plane, we identified both the utricle and the saccule in their greatest diameters on the same view. We estimated their surfaces by manual contouring and deduced the saccule/utricle ratio. A ratio > 1 was considered as a saccule/utricle ratio inversion (SURI) indicative of an EH (Figure 3) (18, 26).

MRI scans were obtained in two centers (39 in center 1, and 37 in center 2) and analyzed by one of the radiologists for each patient (ED for center 1, and JLB for center 2). Measurements and gradings were conducted according to the preestablished protocol. Radiologists were aware of the suspected diagnosis but were blinded regarding the side. Age, sex ratio, duration of symptoms before investigation, DMD/PMD ratio did not differ between the two centers (mean age: 51.2 ± 15.26 years, median: 51.0, range 21–78 vs. 57.4 ± 14.23, median: 59.0, range: 27–78, p = 0.08; sex ratio 0.44 vs. 0.85, p = 0.24; duration of symptoms 71.7 ± 71.94 months, median: 49.0, range: 0–272 vs. 89.0 ± 92.29, median: 62, range: 2–368, p = 0.36, Mann-Whitney test; DMD/PMD ratio 1.6 vs. 4.3, Chi-2 test, p = 0.08 for centers 1 vs. 2 respectively).

Data was collected and managed by Excel^® (Microsoft 2010, Microsoft Inc. Redmond, VI, USA), and analyzed by Prism^® (v.6, Graphpad Inc., San Diego, CA, USA). Most continuous variables did not pass D'Agostino and Pearson's test for normal distribution. Consequently, they were compared by a Mann-Whitney non-parametric test between two unpaired groups, Kruskal-Wallis test for more than 2 unpaired groups or a Wilcoxon signed-rank test in case of paired comparisons. Nominal variables were compared by a Chi-2 test. Correlations between continuous variables were evaluated by the Spearman coefficient and a F-test. R2 > 0.5 with p < 0.05 were considered as significant. Inter criterion reliability was tested by Cohen's kappa on XLSTAT^® (Addinsoft, New York, NY, USA). Logistic regression analysis was conducted on Statview software^® (v.6, SAS Inc., Cary, NC, USA). Values were presented as means ± standard deviation (SD), n, median and range. Supporting data is available at an online repository: Bozorg Grayeli, Alexis (2022), Confrontation of Endolymphatic Hydrops Diagnosis on 3-Tesla MRI to Clinical and Audiovestibular Findings in Meniere's Disease, Dryad, Dataset, https://doi.org/10.5061/dryad.hqbzkh1j7.

The population comprised 47 women (62 %) and 29 men (38 %). The mean age was 54 ± 1.7 years (range: 21–78). Fifty-four patients (61%) had a DMD, and 22 (29%) had a PMD. Sixty-two patients (81%) had a unilateral MD (UMD, 30 right, 32 left), and 14 (19%) had a bilateral disease (BMD). The disease duration at the time of inclusion was 6.2 ± 6.79 years (median: 4, range: 0.1–30). The average number of vertigo spells per month was 4.4 ± 8.85 (median: 1, range: 0.1–30) and the duration of each episode was estimated as 16.3 ± 23.51 h (median: 5, range: 0.1–96). Nine patients (12%) reported vestibular drop attacks. An ongoing medical treatment was noted at the time of inclusion in 51 patients (67%). The treatment was oral glycerol in 16 (21%), betahistine in 14 (18%), tympanostomy tube in 5 (7%), in situ gentamycine in two cases (3%), acetazolamide in 1 (1%), and other symptomatic vertigo medications in 38 patients (50%). Among these, 22 declared an improvement after the treatment onset (44%). Patients with an ongoing treatment at inclusion were predominantly female (sex ratio: 0.4, n = 49 vs. 1.25, without treatment), but they did not differ from those without a treatment in terms of age (53.2 ± 15.92 years, median 53, range: 21–78 vs. 56.0 ± 13.23, median 58, range: 28–73, respectively, p = 0.46, Mann-Whitney test), frequency of vertigo spells (5.6 ± 10.02 /month, median: 1, range: 0–30 vs. 2.4 ± 6.02, median: 0.5, range: 0–30, respectively, p = 0.16, Mann-Whitney test), duration of vertigo spells (14.2 ± 19.28 hours, median: 5, range: 0–96, vs. 19.9 ± 29.60, median: 3, range: 0.02–96, respectively, p = 0.72, Mann-Whitney test), proportion of cases with hearing fluctuation in the involved ears (61%, n = 59 vs. 70%, n = 30, respectively, p = 0.40), and PTA in the involved ears (42.5 ± 26.99 dB HL, median: 41.9, range: 5–120, n = 58 vs. 39.2 ± 27.68, median: 33.8, n = 29, respectively, p = 0.55).

In DMD, 63 ears were affected among 108 (58%), while in PMD, 27 ears out of 44 were symptomatic (62%).

The mean age was higher in DMD than in PMD (56.3 ± 14.94 years, median: 56, range: 21–78 vs. 48.9 ± 14.81, median: 49, range: 21–70, respectively, p < 0.01, Mann-Whitney test). The sex ratio was also different between the groups with a male predominance in DMD (0.86 in DMD vs. 0.22 PMD, p = 0.0066, Fisher's exact test). The disease duration at the time of inclusion was significantly longer in DMD than in PMD (7.1 ± 7.05 years, median: 5, range: 0.1–30 vs. 4.2 ± 5.68, median: 2, range: 0.1–21 respectively, p = 0.0043, Mann-Whitney test).

Expectedly, the hearing loss dominated on the involved ear in cases UMD; and it was greater on the right side in our BMD patients (Table 1). As expected, ECoG was bilaterally abnormal in most of BMD, but also in most patients with UMD. Average SP/AP ratios did not differ between ears in both UMD and BMD groups (Table 1).

Caloric tests in UMD and BMD patients were not always in accordance with the disease side, and in UMD patients, even contralateral weakness could be detected (Table 1). Like the caloric test, cVEMP revealed both relative reduction of amplitudes on the ipsi- or the contralateral side involvement in UMD patients. In the BMD group, only a unilateral involvement was reported in 5 cases. vHIT did not show an abnormal gain in this population (average gain in horizontal plane 0.99 ± 0.09, median: 1.0, range: 0.83–1.16 for right and 0.94 ± 0.09, median: 0.96, range: 0.75–1.06 for left in BMD, n = 11, 0.89 ± 0.22, median: 0.95, range: 0.26–1.36 for ipsi- and 0.92 ± 0.22 for contralateral, median: 1.00, range: 0.15–1.37, n = 54 in UMD).

In DMD, audiologic findings and cVEMP indicated a more severe inner ear dysfunction in comparison to PMD (Table 2). In accordance with audiovestibular tests, MRI showed cochlear and vestibular hydrops more frequently in DMD than in PMD (Table 2).

The proportion of VH based on semi quantitative assessment of the same parameters was not different between affected and asymptomatic ears (Table 3) or between sides in patients with UMD (Table 4). However, in UMD, both the vestibular endolymphatic surface and the saccule surface (in sagittal plane) were higher on the disease side (Table 4). The proportion of EH based on height and width were also similar in both groups except for the saccular width (coronal plane) which was higher for symptomatic ears vs. asymptomatic ears and for the ipsilateral ears vs. contralateral ears in patients with UMD (Tables 3, 4). Paradoxically, saccular height/width ratio was not modified between the two ears in patients with UMD (Table 4). In sagittal plane, saccular surfaces were higher for symptomatic ears than asymptomatic ears and utricular surfaces were lower for symptomatic ears than asymptomatic ears (Table 3). In BMD, PTA showed a dominant loss on the right side (Table 1). According to 3D-Flair images results (no difference between right and left sides except the saccule surface in sagittal plane), EH did not seem to influence the PTA in patients with BMD (Table 4).

The frequency of vertigo attacks was similar in VH grades 0 and 1 (3.2 ± 7.03 attacks/month, n = 87 vs. 6.0 ± 10.65, n = 27, respectively, p = 0.03, Mann-Whitney test). The duration of the disease was higher in ears with VH grade I or II than with grade 0 (93.5 ± 81.16 months, for grade I vs. 64.6 ± 72.31, n = 94, for grade 0, p = 0.035, Mann-Whitney test).

The frequency of attacks was not related to CH grades (4.0 ± 8.38 attacks/month, n = 79 for grade 0, vs. 3.2 ± 7.14, n = 38 for grade I, and 11.7 ± 15.95, n = 3 for grade II, respectively, p = 0.14, Kruskal-Wallis test). Also, the disease duration was not related to CH grades (75.0 ± 84.4 months, n = 85 for grade 0, vs. 76.6 ± 70.7, n = 42 for grade I, and 53.7 ± 31.3, n = 3 for grade II, respectively, p = 0.55, Kruskal-Wallis test).

Hearing loss seemed to be influenced by CH grades (PTA=61.3 ± 34.26 dB, n = 3 in grade II vs., PTA=43.3 ± 29.94 dB, n = 44 in grade I vs. 30.2 ± 27.59 dB, n = 84 in grade 0, p = 0.016, Kruskal-Wallis test).

Among selected criteria, SURI seemed to be best related to the side of the disease (Tables 3–5). The disease duration did not differ between ears with and without SURI (75.5 ± 80.89 months, n = 43 vs. 85.0 ± 80.63, n = 34, p = 0.86, Mann-Whitney test), but ears with SURI tended to be associated to more frequent attacks (3.4 ± 7.64 attacks/month, without SURI, n = 16 vs. 4.8 ± 9.12, with SURI, n = 33, p = 0.071, Mann-Whitney test).

There was no significant correspondence between SURI and other MRI criteria based on FLAIR and T2-CISS images (Table 6), suggesting that each criteria evaluates a different aspect of the EH.

The proportion of hydrops according to height and width criteria was similar in clinically involved and unaffected ears except for the height/width ratio which was higher for the asymptomatic ears vs. symptomatic ears and for contralateral ears vs. ipsilateral ears in patients with UMD such as the saccular height in UMD (in the coronal plane, contralateral ears vs. ipsilateral ears) (Tables 3, 7). Similarly, Caloric test, cVEMP and ECoG did not seem related to EH detected on T2- weighted images. Finally, a same-day MRI did not produce a better correspondence between T2-MRI criteria and audiovestibular tests than later imaging expect for the correlation with cVEMP (Table 5).

There was no agreement between saccular height-width criteria and any of the criteria on 3D-FLAIR to indicate the side of the involved ear (Cohen's kappa < 0.1). Similarly, there was no agreement between SURI and other criteria based on 3D-FLAIR views (saccular height and width on coronal views or endolymphatic surface on axial views, Cohen's kappa < 0.1). We observed a minimal accordance between SURI and the presence of CH (Cohen's kappa = 0.23), and between VH and CH grades on 3D-FLAIR views (Cohen's kappa = 0.18). Although total vestibular surfaces were well correlated between T2- weighted and FLAIR-3D sequences (Y=4.61+0.85X, Y: surface on FLAIR, X: surface on T2, R=0.65, p < 0.0001, F test), saccular height and width on T2 were not correlated to those on FLAIR-3D (R² = 0.02, F = 1.47, p = 0.23 for the height and R² = 0.07, F = 4.94, p = 0.03 for the width, n = 67, ANOVA).

Analysis of Receiver Operating Characteristic (ROC) curves showed that among MRI diagnostic criteria, saccule surface on sagittal views, saccule/utricle surface ratio on sagittal 3D-FLAIR views, and saccule height/width ratio on coronal T2 views could discriminate between asymptomatic and affected ears according to clinical criteria (Figure 4). Discrimination capacity of T2 views was low: a ratio < 1.14 had a sensitivity of 88% but a specificity of only 23% (likelihood ratio=1.13). The same parameter on 3D-FLAIR, and endolymphatic/total vestibular surface on axial views could not discriminate between involved and asymptomatic ears. In contrast, Saccule/Utricle surface > 1on sagittal 3D-FLAIR views yielded a sensitivity of 81% and a specificity of 64% (likelihood ratio = 2.2).

By choosing the caloric weakness as the indicator of the affected side instead of clinical criteria, not only height/width on T2 and saccule/utricle ratio on 3D-FLAIR but also relative endolymphatic surface on axial 3D-FLAIR appeared to indicate the affected side (Figure 5). Among these criteria, the saccule/utricle ratio on 3D-FLAIR had the best discrimination ability and detected the weak caloric side with a 74% sensitivity, a 72% specificity (likelihood ratio = 2.6).

By selecting SP/AP > 0.33 on EcoG as an indicator of the disease, no MRI diagnostic criteria could distinguish between involved and normal ears as judged by ROC analysis (Table 8).

Separate ROC analysis of definite and probable unilateral MD showed that saccule and utricle surfaces on sagittal FLAIR views had a higher performance to discriminate affected ears in DMD than in PMD as judged by AUC (Table 9).

A model combining MRI criteria and audiovestibular data was explored to predict the clinically affected ear. Among all possible combinations between MRI criteria, PTA on audiometry, caloric weakness, vHIT, VEMP loss or asymmetry, ECoG (SP/AP > 0.33), the model which best predicted the affected ear based on clinical criteria was the combination of SURI (yes/no) and caloric weakness (yes/no). This model yielded the following characteristics: R² = 0.22, Log likelihood = −66.2, R_SURI = 0.26, R_caloric = 0.15, and likelihood ratio: Chi-2 = 37.74, p < 0.0001. This combination improved the positive likelihood ratio of SURI. With this model, 44 out of 51 (86%) unaffected ears were predicted as normal, and 50 out of 75 affected ears (67%) were predicted as abnormal (specificity: 86%, and sensitivity: 67%).

According to SURI criteria, ears with EH on same-day MRIs had greater hearing loss than those without hydrops, while in the delayed-MRI group, there was no difference in PTA between EH and no EH group (Table 5). In contrast, caloric and cVEMP loss appeared to be related to radiological EH both in the same-day MRI and the delayed-imaging groups if SURI or axial vestibular surface criteria on FLAIR views were considered (Tables 5, 10). Audiovestibular findings did not seem to be influenced by the imaging delay if other hydrops criteria were employed (Tables 5, 10). Interestingly, cVEMP loss appeared to be more frequent in patients no sign of EH according to height/width criteria on T2 sequences than in those with EH in the same-day MRI group (Table 5) underlining the possible dissociations between electrophysiology and MRI data.

It is noteworthy that this transversal study included all patients seen for the first time for a MD with a significant heterogeneity in terms of age, duration of the disease, severity, and ongoing treatment. This heterogeneity is typical in MD and reported in other publications (3, 4). At the time of inclusion, no therapeutic change was imposed, and the type of ongoing treatment was not set as an inclusion criterion because this type of restriction would have reduced the number of inclusions and the generalizability of the observations. Two patients (3% of the population) were treated with gentamicin. It is therefore important to note that the results of audiovestibular test in these patients might introduce a bias due to the potential vestibulotoxic and (to a lesser degree) cochleotoxic effect of the drug. These patients should perhaps not have been included in this study in the first place as their audiovestibular function may have been affected by the in-situ treatment.

ECoG results were bilaterally in most UMD patients. Similar results were found with comparable ratio SP/AP and area ratio of SP to AP between ispsilateral and contralateral ears in UMD patients (27). In contrast, another study comparing healthy and diseased ears in UMD patients found a difference in the SP response at 1 kHz, and the SP/AP ratio between the two ears (9). Also, several research teams describe a low sensitivity of the ECoG to confirm Meniere's disease, especially when using click (which is the case in our study) (9, 28). In our study, we find rather the opposite: as the contralateral ears display pathological results in the UMD group, the sensitivity of the ECoG is rather to high with a low specificity. Another study to investigate the sensitivity and specificity of ECoG to discriminate between healthy and “Meniere's” ears in UMD patients according to the use of clicks or tone burst would be interesting to develop.

Concerning UMD patients, cVEMP revealed relative reduction of amplitudes on the ipsi- or the contralateral side. It's important to precise that a relative reduction of cVEMP amplitudes on the contralateral side does not necessarily indicate a pathology on the contralateral side. In fact, it might also be due to a relative elevation of the cVEMP amplitude on the MD side, as has been observed before particularly in early stages of MD (29).

In contrast to other series (30), our semiquantitative analysis did not show significant hydrops in many cases (for example, VH grading p = 0.79 and CH grading p = 0.46, in UMD, vs. opposite ear, Table 4). However, quantitative analysis such as height and width criteria led to the diagnosis of hydrops in 92% of DMD and in 75% of PMD (Table 2). SURI criterion was also present in nearly 70% of ears with DMD. This discrepancy suggests that semiquantitative criteria (VH and CH grading) are probably less sensitive than quantitative measures and many abnormal ears were classed as grade 0 in the semiquantitative scale. To support this idea, another team enhanced the performances of the semiquantitative vestibular hydrops scale by adding a lower grade between 0 and 1 and proposed a four-grade classification (12).

In the first reports of this imaging technique, an intratympanic injection of gadolinium was employed to opacify only the perilymphatic space and to visualize the endolymphatic sector by contrast (10, 13–15). With this technique, the dilatation of the endolymphatic space can be seen in the vestibular and cochlear regions as black spots in a contrast-enhanced perilymphatic space (10, 13). Although reliable, this technique was progressively abandoned in favor of intravenous gadolinium (16, 18, 26, 31, 32). To reach the perilympahtic space through this route, the contrast agent requires several hours (16), and this delay often complicates the examination protocol in the routine practice. As an alternative, evaluation of the saccular anatomy on T2- weighted images has been proposed to avoid gadolinium (11). With this technique, the saccule could be clearly visualized on coronal slices, and its size and shape quantified. Hydrops was reportedly associated with an enlargement of this structure taking a more circular shape (11). By avoiding the gadolinium injection, this technique could be much easier to organize and would be better integrated in the routine practice. However, the analysis of our series was disappointing by showing no correlation between 3D-FLAIR and T2, revealing no relation between clinical or audiovestibular signs and T2-images, and showing a low discrimination of symptomatic ears on ROC analysis.

The fuzzy limits of the endolymphatic space on T2 sequences might have hampered the exact evaluation of the saccular size. Also, the fact that a small hydrops with no clinical signs may be present on the contralateral ear of a unilateral MD reduces the discrimination between affected and unaffected ears.

One of the most interesting findings of 3T-MRI is the presence of EH in asymptomatic ears of MD patients and even in normal subjects (17, 18). This finding corroborates histopathological studies showing endolymphatic enlargement in subjects with various diseases other than MD (e.g., otosclerosis, idiopathic progressive hearing loss), and even in those with no reported otological symptoms (21, 32). These observations suggest that the pathological mechanisms leading to EH are not fully understood (33, 34). EH seems to be merely an indicator of abnormalities in inner ear fluid homeostasis and it cannot solely explain the symptoms in MD (35). This further complexifies the relation between EH and audiovestibular findings in MD patients. Today, 3T-MRI data suggests that this phenomenon is far from anectodical, and this reduces considerably the specificity of EH in MD diagnosis (17, 18). The clinical and the pathophysiological significances of this phenomenon are not clear, and the crucial question in UMD is whether the presence of endolymphatic hydrops in the contralateral ear augurs poorly, indicating a BMD in the future. To our knowledge, there is no report focusing on this subject. The relatively low number of reported BMD (36) in comparison to the proportion of bilateral or contralateral EH on MRI is reassuring (18), but only a long-term follow-up of patients with UMD and bilateral EH will provide a direct answer (36).

Another interesting question is the reversibility of the endolymphatic space enlargement between two vertigo attacks. Intuitively, reversibility of the symptoms and the audiovestibular signs should be accompanied by a possible regression of the hydrops on imaging. But the kinetics of the electrophysiological alterations and the morphological signs are probably different (37, 38) and the observed lag between a normalized function and a decrease of endolymphatic pressure can explain the absence of correlation between audiovestibular findings and MRI observations (38). Another factor which can disturb such correlations is the extent of irreversible cochleovestibular damage caused by gentamycine, age, long-term hydrops, or comorbidity factors which do not vary with the hydrops. These elements indicate that, MRI should be regarded as one additional diagnostic tool, and not a gold standard for the diagnosis and the treatment of MD. This tool can be useful to support or confirm the diagnosis in clinically atypical cases or in the early stages of the disease with only cochlear or vestibular symptoms (3, 4). It would be interesting to discuss its integration to the international MD diagnostic criteria: MRI signs of EH in a probable MD could lead to reclassify the case into definite MD. In fact, those patients are less likely to present with EH on MRI, as has been shown in the present study (lower rate of EH in PMD than DMD) and previous studies (39).

A common limit in all MRI-based morphological studies of EH is that the threshold at which the diagnosis is established is crucial for both sensitivity and specificity of the MRI. These thresholds are based on relatively small number of cases (n< 70) (10, 11, 17). Moreover, sensitivity and specificity of these tests are based on the clinical diagnosis of MD which is not constantly associated to histopathological deteriorations during EH (20). Alternatively, the use of electrophysiological indicators (e.g., SP/AP on ECoG) for this purpose has been disappointing (40, 41).

One of the limitations of the surface methods to evaluate hydrops probably comes from the fact that the endolymphatic spaces show a significant interindividual variation even in healthy ears (42). Indeed, on a sagittal medial plane, the saccule could be detected in only 15 among 22 examined healthy ears on 3D-FLAIR MRI sequences after gadolinium. The endolymphatic space had also a variable shape in the axial plane, often Y-shaped (77%) but globular in other cases (23%).

Another limit is that MRI approaches its resolution limits in morphological studies of small structures such as the inner ear. Reliability of linear measurements between 1 and 2 mm on MRI is low since they are close to the slice thickness (43, 44). This is probably one of the reasons why criteria such as SURI and semi quantitative classifications of vestibular and cochlear hydrops seem more coherent to clinical data than diameter and surface measurements. Low resolution might also explain false negative cases. Indeed, the dilatation of the cochlear duct, which is difficult to visualize, is a constant histopathological observation in MD while saccular and utricular enlargements are noted on MRI in only 80 and 55% of temporal bones respectively (45). Focal dilatations of the cochlear duct are also frequently observed in MD and other diseases (33) and are probably underestimated on 3T-MRI (18). Our study also suffers from several limitations proper to its design. The heterogeneity of the population in terms of disease duration, age and medication can introduce discrepancies in terms of relation between audiovestibular findings and MRI. Indeed, recent MD will present with smaller hearing loss and little or no caloric weakness but possibly significant EH on MRI while MD progressing for several years may show more sever audiovestibular deficit with similar or smaller EH on MRI. Even though every precaution was taken during the examination and the classification of the MRI images, interpretation of the images by 2 independent radiologists would have added to the precision of the data.

MRI resolution is directly dependent on the signal/noise ratio and this parameter could be physically improved by increasing the magnetic field force (i.e., 7T-MRI used in research field) or acquisition time (46, 47). A specific antenna close to the target can also enhance SNR (48). In future, signal processing algorithms will also better detect signal in noise and improve image quality and resolution (49).

In conclusion, among several EH criteria, SURI had the strongest relation to audiovestibular and clinical MD manifestations. There was no significant relation between SURI and other MRI criteria for EH even in ears with definite MD suggesting that they evaluate different aspects of the EH and a composite criterion might be more suitable to evaluate the EH. A short delay between MRI and audio vestibular testing seemed to improve coherence between these tests and EH diagnosis on MRI.