This is, to the best of our knowledge, the largest retrospective study to date analysing the accuracy of MRI-DWI in detecting residual cholesteatoma across multiple surgery types with a generally long FU. Scan outcomes were verified per-operatively and specifically residual disease was noted (as oppose to recidivism in general), increasing accuracy of diagnostic parameters. As this study was done retrospectively, it is a realistic evaluation of the applicability of MRI-DWI in detecting residual cholesteatoma in clinical practice.

Our study revealed a substantial amount of FN and FP MRI-DWI in radiologic FU. This underlines the importance of correlating findings of MRI-DWI with per-operative findings, rather than considering accuracy of imaging based on clinical FU. Causes of FP MRI-DWI findings are in line with other studies, reporting a Silastic™ sheet or accumulated ear wax against a retracted tympanic membrane to potentially interfere with cholesteatoma detection [23, 28, 29]. In our centre, ear wax removal at the outpatient clinic before obtaining MRI-DWI is done regularly and could be considered standard preparation.

Our reported rate of MRI-DWI sensitivity in detecting residual disease is in line with literature [5, 7, 10, 13,14,15,16,17,18,19,20,21,22,23,24]. Compared to a recent meta-analysis of MRI-DWI sensitivity and specificity in detecting recidivism (respectively 93% and 91%) [10], our rates are slightly lower. In most studies, MRI-DWI is performed on basis of clinical suspicion of residual disease [14, 15, 17, 19, 22, 30], resulting in a higher incidence of residual cholesteatoma and introducing a selection bias. We expect reduced selection bias in our data, as the scans in our cohort were performed in all patients as part of routine FU besides clinical suspicion. Moreover, scans are often scored for study purposes [13, 22, 24], improving diagnostic parameters. In this study, scans were read by a group of head- and neck and neuroradiologists with different levels of experience as part of normal clinical routine. Therefore, our results give a more realistic representation of the diagnostic value of MRI-DWI in routine daily practice with a higher external validity.

Even though 1.5T systems seem to be more common in the current literature, the use of 3.0T compared to 1.5T for radiologic FU of cholesteatoma has shown to be indifferent [31]. Furthermore, there were no evident differences in diagnostic parameters before and after the alteration of b-value used.

This study illustrates a greatly varying prognostic value of MRI-DWI over length of FU, undoubtedly contributing to the wide range of reported diagnostic parameters. The benefit of early scanning can be questioned by the limited diagnostic accuracy within 1.5 yrs of surgery. Previous studies report current MRI-DWI are able to detect cholesteatoma at 2–3 mm and suggest radiological diagnosis of residual cholesteatoma is possible at 9–12 months [10]. However, growth rate of residual cholesteatoma varies [32, 33] and therefore imaging obtained rather shortly after primary surgery can lead to a considerable amount of FN. After approximately 3 yrs of FU, sensitivity of MRI-DWI for residual disease surpasses 80% and significantly more TP MRI-DWI were obtained. This is in line with a recent study reporting an average interval between initial surgery and positive MRI-DWI of 3.8 yrs (of which almost all positive scans were confirmed to be cholesteatoma per-operatively) [32].

This study confirms the role of obliteration in reducing residual disease [25, 33,34,35,36]. Multiple theories could explain the latter: prior to obliteration, complete visualization of the anterior epitympanic space is crucial. This could expose missed disease, as this space is prone to residual disease [37, 38]. Removal of the malleus head and incus reduces the risk of leaving cholesteatoma on an intact ossicular chain [39]. Potentially obliteration could terminate proliferation of any residual epithelial cells (“Hinohira effect” [40]). Lastly, the surgeon could be even more vigorous, as revision surgery after obliteration will not be simple.

Obliteration was not associated with more FP MRI-DWI, perhaps due to the obliteration material used (hydroxyapatite granules). Similar to the “Hinohira effect”, obliteration potentially prevents the formation of granulation tissue or cholesterol granuloma in the mastoid cavity, two frequent causes of FP findings [15, 23]. Also, obliteration of the mastoid cavity was associated with less FN MRI-DWI. This could be due to growth of epithelial remnants in a compressed space, facilitating the threshold of detection with MRI-DWI. Residual disease was not found earlier after obliteration and therefore a shortened interval between initial surgery and first MRI-DWI in obliterated ears is not necessary.

Patients who underwent canal wall up surgery had a higher chance of FN MRI-DWI, demanding increased awareness after a negative scan.

Our study confirmed a higher risk of residual disease in cholesteatoma acquired at a young age [25, 35, 41, 42]. MRI-DWI FU is widely applied in the paediatric population [17, 43] and previously it has been suggested to increase frequency of MRI-DWI FU in this group [10, 43]. As we did not find residual cholesteatoma earlier in the paediatric group, shortened intervals between initial surgery and radiologic FU do not seem necessary. As recidivism of paediatric cholesteatoma mostly occurs within 5 yrs [44], a “late” scan to detect delayed residual disease, conform the adult population, is warranted. The higher rate of FN MRI-DWI found in patients < 12 yrs could be due to the specific growth pattern of paediatric cholesteatoma: invasive rather than forming a dense keratin pearl with associated diffusion restriction.

The diagnostic accuracy of MRI-DWI for FU of residual cholesteatoma could be increased by optimizing a standardized FU scheme. If correctly applied, MRI-DWI can replace a standard second-look procedure. In a stable ear, the first routine FU scan (an “early” scan) can be obtained approximately 3 yrs post-operatively. If the result is doubtful, we suggest to repeat it after 12 months. A “late” scan can be obtained after approximately 5 yrs. High risk patients, i.e., patients < 12 yrs or patients after surgery with an intact canal wall and without obliteration, show how higher FN MRI-DWI rates. We therefore suggest an “extra late” scan approximately 9 yrs after initial surgery in this high-risk group, in order to catch previously missed disease. Shortened intervals for these specific patients does not seem necessary.

Detecting cholesteatoma approximately 3 yrs after initial surgery might raise the question whether this is not “too late”. The Covid pandemic has raised similar concerns regarding increased delays between cholesteatoma diagnosis and surgery. Increased waiting times up to 1 yr have not been associated with increased recidivism or major complications defined as facial nerve palsy or intracranial complications [45]. It could be argued early detection and subsequent surgery prevent further damage of the ossicular chain. However, the question remains if this would lead to better audiologic outcomes [46, 47]. Furthermore, doubts concerning dubious outcomes of early scans lead to further wait-and-scan FU, wasting costly resources. Starting radiologic FU approximately 3 yrs after initial surgery increases diagnostic accuracy, reduces unnecessary imaging, misdiagnosis and potentially even prevents unwarranted second-look procedures. This reduces costs whilst also keeping sustainability in mind. In our opinion, FU scans within 1.5 yr of surgery should be limited and solely reserved for unstable ears with clinical suspicion of residual disease, warranting early intervention.

The “late” scan is crucial to reveal delayed growth of residual disease. A scan after 4 or 5 yrs has been suggested previously, as up to 31% of MRI-DWI turn positive after initial negative MRI-DWI, without any clinical signs of disease [11, 32]. Also, long-term residual disease is reported up to a decade after initial surgery and some studies therefore suggest lifelong surveillance [32]. We identified specific patient and surgical characteristics that lead to a higher risk of residual disease and a higher risk of FN MRI-DWI, necessitating a longer period of radiologic FU. A flowchart of our suggested optimized FU scheme is provided in Fig. 5.

Flowchart of MRI-DWI follow-up scheme after cholesteatoma surgery: standard MRI-DWI after approximately 3 and 5 yrs, as well as an MRI-DWI after approximately 9 yrs in patients with specific risk factors. “− ” indicates no cholesteatoma was identified on MRI-DWI, “?” indicates a dubious result of MRI-DWI and “ + ” indicates a cholesteatoma identified on MRI-DWI. FU: follow-up; w/o: without

The retrospective set-up of our study has obvious restrictions, such as the risk of slight selection bias and assessment of imaging by a group of head- and neck and neuroradiologists with different levels of experience. Re-evaluation of FN and FP MRI-DWI by an experienced head- and neck and neuroradiologist yielded 1/4 FP and 4/32 FN MRI-DWI misinterpretations, while 3/4 FP and 12/32 FN MRI-DWI were confirmed. Retrospectively, a focus too small to classify was identified in 8/32 FN MRI-DWI. In another 8/32 FN MRI-DWI (all obtained in 2012–2014), a dubious focus was found. The latter scans were found to be of lesser quality than later scans, most probably due to the use of an older scanner in that period. Recalculations of diagnostic parameters after correction of misinterpreted MRI-DWI did not influence our conclusions significantly.

Furthermore, this study was performed in a tertiary referral centre, potentially influencing type of cholesteatoma and respective surgery performed. An under-coverage bias in our study may result in a slight underestimation of sensitivity and specificity rates; it is expected that a larger proportion of the negative MRI-DWI that are still in FU are true negative rather than false negative. Furthermore, timing of TP MRI-DWI was used to estimate timing of disease manifestation. However, the timing of routine scans followed our standard FU scheme and therefore may not directly reflect the exact timing of disease manifestation. It is also possible incidental FN and FP MRI-DWI were missed in the case of coincidental presence of both FN and FP within one ear, for instance a Silastic™ sheet and a small residual pearl. We are aware of contra-indications of MRI-DWI, such as patients with cochlear implants or severe claustrophobia. This group, however, is expected to be a negligible.

Ideally, a prospective study would obtain MRI-DWI in all patients planned for a second-look procedure in both secondary and tertiary referral centres, based on clinical suspicion or planned routinely at various months to years of FU. In the future, long-term outcomes of reduction of surgical interventions, scans, costs and carbon footprint as well as safety and the occurrence of adverse events in the proposed FU scheme could be analysed. The introduction of fused CT and MRI-DWI, a novelty in our institution, can also further enhance diagnostic accuracy of MRI-DWI FU (supplemental Fig. 1). Also, MRI-DWI FU does not replace out-patient clinic visits for detection of recurrent disease, as a novel retraction pocket can form at any moment after surgery. An optimum FU scheme to detect recurrences was beyond the scope of this study and could be the subject of further research.