Performance of TGSE BLADE DWI in the diagnosis of cholesteatoma compared with RESOLVE DWI

Background: Based on the high resolution of soft tissue, MRI has gained increasing importance in the evaluation of cholesteatoma, especially diffusion-weighted imaging(DWI). The purpose of this study was to evaluate the role of 2D turbo gradient- and spin-echo (TGSE) diffusion-weighted (DW) pulse sequence with BLADE trajectory technique in the diagnosis of cholesteatoma at 3T and to qualitatively and quantitatively compare the image quality between the TGSE BLADE and RESOLVE methods. Method: A total of 42 patients (23 males, 19 females; age range, 7-65 years; mean, 40.1 years) with surgically conrmed cholesteatoma in the middle ear were enrolled in this study. All patients underwent DWI (both the prototype TGSE BLADE DWI sequence and RESOLVE DWI sequence) using a 3-T scanner with a 64-channel brain coil. Qualitative imaging parameters (imaging sharpness, geometric distortion, ghosting artifacts, and overall imaging quality) and quantitative imaging parameters (apparent diffusion coecient [ADC], signal-to-noise ratio [SNR], contrast, and contrast-to-noise ratio [CNR] for the two diffusion acquisition techniques were assessed by two independent radiologists. Result: Comparison of the qualitative scores indicated that TGSE BLADE DWI produced less geometric distortion and ghosting artifacts (P<0.001) and higher image quality (P<0.001) than RESOLVE DWI. Comparison of the evaluated quantitative image parameters between TGSE and RESOLVE showed that TGSE BLADE DWI produced a signicantly lower SNR (P<0.001) and higher parameter values (both contrast and CNR (P < 0.001)) than RESOLVE DWI. The ADC (P<0.001) measured by TGSE BLADE DWI (0.763×10-3 s/mm2) is signicantly lower than that measured by RESOLVE DWI (0.928×10-3 s/mm2). Conclusion: Comparing with RESOLVE DWI, TGSE BLADE DWI can signicantly improve the image quality of cholesteatoma by reducing magnetic sensitive artifacts, distortion, and blurring. TGSE BLADE DWI is more valuable for the diagnosis of small-sized s/mm 2 ) previously reported in the literature 32-34 . The mean ADC of the cholesteatoma and brainstem on TGSE BLADE DWI was 0.763×10 -3 s/mm 2 and 0.498×10 -3 s/mm 2 , respectively. These ndings demonstrate that the ADC obtained in our study can provide an auxiliary basis for more clinical applications of TGSE BLADE DWI in the future.


Background
Cholesteatoma is a benign, gradually expanding and destructive epithelial lesion of the temporal bone that results in the erosion of adjacent bony structures and may lead to various complications.1 In addition to the clinical features and otoscopic ndings, the early diagnosis of cholesteatoma with imaging examinations, such as high-resolution computed tomography (CT) and magnetic resonance imaging (MRI), is important. From a surgical perspective, highresolution CT remains the primary imaging technique for the diagnosis and characterization of cholesteatoma in the middle ear due to its high spatial resolution and good visualization of bone structures11,12. However, in terms of the disadvantages, it is di cult to distinguish cholesteatoma from granulation tissue, brous tissue, or uid on high-resolution CT2. Based on the high resolution of soft tissue, MRI has gained increasing importance in the evaluation of cholesteatoma. Many studies have shown that MR diffusion-weighted imaging (DWI) has high sensitivity and speci city for identifying the presence of cholesteatoma due to its high keratin content13-15. However, conventional DWI uses an echo-planar imaging (EPI) trajectory to collect k-space data, and the obtained images may suffer from severe susceptibility artifacts at air-bone interfaces. Additionally, the image resolution is limited. Thus, it is di cult to use DWI in cases of a lesion less than 5 mm3,4 from the distortion area. To reduce image distortion, readout-segmented echo-planar imaging (RESOLVE) has been proposed, and this method can signi cantly improve head and neck DWI by reducing the echo spacing. Although RESOLVE DWI has a signi cantly improved image signal-to-noise ratio (SNR) and reduced image distortion, the partial volume effect and T2* blurring effect are not completely eliminated, it is di cult to detect small lesions (<2.5 mm)6-8. Periodically rotated overlapping parallel lines with enhanced reconstruction (PROPELLER) DWI is a non-echo planar fast spin-echo-based DWI sequence, and the central k-space is acquired in a rotating manner. The PROPELLER sequence is free of geometric distortion and susceptibility artifacts, but the scan time is long and imposes a high speci c absorption rate (SAR), especially at high elds28-31. To the best of our knowledge, the BLADE DWI technique has been reported to eliminate susceptibility artifacts by applying the 'blades' acquisition scheme in k-space5. Such a non-EPI technique was further optimized by using a TGSE method to increase the SNR e ciency and both achieve the detection of small (<2.5 mm) cholesteatomas and increase the resolution to decrease susceptibility artifacts, allowing for differentiation from granulation tissue5. Houchun H. et al.18 concluded that TGSE BLADE DWI exhibited less geometric distortion in the brain and reduced magnetic artifacts near the air-tissue interface than conventional SE-EPI. However, the use of TGSE BLADE DWI in studying ear lesions has not yet been reported. L.M.J. Lips et al. 19 found that non-EPI DWI for the detection of residual or recurrent cholesteatoma is better at 3 T than 1.5 T. Hence, the purpose of this study was to evaluate the role of the TGSE BLADE DWI technique in the diagnosis of cholesteatoma at 3 T and to qualitatively and quantitatively compare the image quality between the TGSE BLADE and RESOLVE methods.

Patients
A total of 42 patients (23 males, 19 females; age range, 7-65 years; mean, 40.1 years) with surgically con rmed cholesteatoma were enrolled in this study from October 2018 to April 2019. Clinicopathological results were the gold standard for all patients.

Ethics and Consent to Participate
The study was approved by the Review Committee of Eye & ENT Hospital of Fudan University, and written informed consent was obtained from all patients.

Imaging technique
All patients underwent DWI (both the prototype TGSE BLADE DWI sequence and the commercially available RESOLVE DWI sequence) using a 3-T scanner (MAGNETOM Prisma, Siemens Healthcare, Erlangen, Germany) with a 64-channel brain coil. All images of 42 patients obtained by TGSE or RESOLVE were evaluated by two radiologists with 10 years of experience in ear MRI evaluation.
Each observer randomly and blindly evaluated the images without knowing the type of DWI sequence and compared the two different DWI sequences using the side-by-side display method. A nal decision was made based on mutual consultation when there was a divergence in the assessment results.
Qualitative evaluation of images obtained by TGSE and RESOLVE was performed according to ve criteria: image sharpness, geometric distortion, ghosting artifacts, and overall image quality. Image sharpness was assessed on a scale that ranged from 1 to 3. Both geometric distortion and ghosting artifacts were evaluated on a scale from 1 to 4, and the evaluation of geometric distortion included two parts: the whole image and the lesion in the ear. The overall image quality was also graded on a scale of 1 to 5. The detailed qualitative evaluation criteria are shown in Table 1. In Figure 1, images C and E represent geometric deformation of 4 (no distortion) and 2 (moderate distortion), respectively.

Qualitative analysis of image quality
The SNR, contrast and contrast-to-noise ratio (CNR) were the main evaluation criteria for the quantitative analysis of two sequence images. The apparent diffusion coe cients (ADCs) of the two sequences were compared simultaneously. On the b1000 TGSE and RESOLVE images, the SNR of the lesions in the region of interest (ROI) was de ned as the ratio of the mean signal intensity of the lesion (S ROI ) to the standard deviation of the background noise (σ BG ) (SNR = S ROI /σ BG ) 9 . The SNR of the brainstem was calculated by the same method, as follows: SNR=S B /σ BG. Contrast was de ned as the ratio of the signal intensity of the lesion (S ROI ) to the signal intensity of the brainstem (S B ) on the b1000 map (contrast=S ROI /S B ).
The CNR was de ned as the difference between the S ROI and S B , divided by the standard deviation in the lesion ROI (σ ROI ) and the brainstem ROI (σ B ) 6-8,10 , as follows: See Formula 1 in the Supplemental Files.
The ROI of the lesion on the b1000 and ADC maps was manually drawn as 1 mm 2 at the level of the maximum diameter of the lesion, and the corresponding signal intensity and standard deviation were automatically generated on the MRI workstation. The ROI of the brainstem was de ned by selecting 10 mm 2 of the brainstem, and the signal intensity and standard deviation of each ROI were automatically generated. A circular ROI of 10 mm 2 was set in the background on the b1000 map for all patients, and the standard deviation of the ROI was automatically generated. In selecting the ROI, areas affected by susceptible artifacts or volume effects were avoided.
The parameters were measured independently and randomly by two raters, with an interval of 2 weeks. The mean value of the two measurements was selected as the nal data for further analysis. Compared with brain tissue on the DWI sequences, the diagnostic criterion for cholesteatoma was a very high signal intensity, corresponding to limited diffusion on the ADC maps (3,20) . The size of all lesions was determined on the basis of T2-weighted imaging (T2WI).

Statistical analysis
All statistical analyses and plots were performed and created using the SPSS 24.0 software package (Chicago, IL, USA), and P < 0.05 was considered statistically signi cant. The normality of all measurements from the TGSE and RESOLVE sequences was tested using the Shapiro-Wilk test. Signi cant differences in qualitative parameters between TGSE and RESOLVE DWI were determined using the Wilcoxon rank-sum test, and signi cant differences in quantitative parameters were determined using the paired t-test. The interreader correlation of the ADC as a quantitative index was evaluated using the intraclass correlation coe cient (ICC). The range of the ICC coe cient was set from 0 to 1.00, and the ICC was de ned as follows: < 0.40, poor; 0.41-0.60, moderate; 0.61-0.80, good; > 0.81, excellent 16,17 . The mean ADCs of the lesions and brainstem measured by the two observers were further calculated, and the differences between them were assessed by paired t-test.

Qualitative analysis of image quality
Comparison of the qualitative scores indicated that TGSE BLADE DWI produced less geometric distortion and ghosting artifacts (P<0.001) and higher image quality (P<0.001) than RESOLVE DWI. The average TGSE and RESOLVE scores were as follows: geometric distortion (whole), 3.97±0.15 and 3.26±0.26 (P<0.001); geometric distortion (lesion), 3.95±0.21 and 3.64±0.48 (P<0.001); ghosting artifact, 3.92±0.26 and 3.07±0.55 (P<0.001); overall image quality, Table 1 shows a comparison of the qualitative parameter scores for TGSE BLADE DWI and RESOLVE DWI, and Figure 2 indicates the distribution of the qualitative scores for TGSE BLADE DWI and RESOLVE DWI. As shown in Figure 1, axial TGSE DWI could precisely de ne the cholesteatoma lesion in the right middle ear without geometric distortion or ghosting artifacts, whereas RESOLVE DWI showed signi cant artifacts at the air-bone interface (between the mastoid, i.e., the location of the cholesteatoma lesion, and the nasal sinus).

Quantitative analysis of image quality
Comparison of the evaluated quantitative image parameters between TGSE and RESOLVE showed signi cant differences between the two groups. TGSE BLADE DWI produced a signi cantly lower SNR (P < 0.001) and higher parameter values (both contrast and CNR (P < 0.001)) than RESOLVE DWI. The results of the statistical analysis are as follows ( Table 1) In terms of measurement and evaluation of the ADC, values were measured in 40 cases, as the lesions were too small to be measured on the ADC maps in 2 cases, and excellent interreader agreement was obtained. Detailed interreader ICCs are shown in Table 2. All ADCs were measured twice by the two observers, and the average values were taken as the basis for further statistical analysis. As shown in Table 2, there was a signi cant difference in the ADC of the cholesteatoma between TGSE BLADE and RESOLVE DWI (P<0.01). The mean ADC of the cholesteatoma measured on TGSE (0.763×10 -3 s/mm 2 ) BLADE DWI was signi cantly lower than that measured on RESOLVE (0.928×10 -3 s/mm 2 ) DWI (P<0.01). Similarly, the ADC of the brainstem measured on TGSE (0.498×10 -3 s/mm 2 ) BLADE DWI was lower than that measured on RESOLVE (0.773×10 -3 s/mm 2 ) DWI (P<0.01). The box plot in Figure 3 shows the distribution of the lesion and brainstem ADCs measured on TGSE BLADE and RESOLVE DWI.
In this study, the measurement results in terms of lesion size in 42 patients showed that TGSE BLADE DWI could show small lesions more clearly than RESOLVE DWI. Compared with RESOLVE, TGSE had much better image quality at the air-bone interface (nasal sinus, middle ear, mastoid) and signi cantly reduced ghosting artifacts and distortion. Furthermore, as shown in Figure 4, axial TGSE BLADE DWI could completely and clearly show a small lesion (1.9 mm in width) located in the left tympanic cavity with less geometric distortion than RESOLVE DWI.

Discussion
DWI is increasingly applied for the evaluation of various diseases in many areas of the body. Conventional DWI (SS-EPI) is often used in head and neck diseases; however, due to interference with the T2* blurring effect and susceptibility artifacts of various tissues produced by non-movement, the image quality of conventional DWI (SS-EPI) is usually not satisfactory 26,27 . Many studies (3,21,22) have concluded that the size limit of cholesteatoma on EPI DWI is 5 mm and that smaller cholesteatoma lesions are easily missed on DW EPI. Compared with conventional SS-EPI DWI, RESOLVE has signi cantly improved image quality due to its low susceptibility-based image distortion and T2* blurring and its robust correction for motion-induced phase artifacts 23 . RESOLVE DWI is more widely used in head and neck diseases than SS-EPI DWI. However, RESOLVE still has some shortcomings that need to be resolved, for instance, image artifacts and distortion (air-bone interface) that cannot be completely eliminated and a low diagnostic rate of small lesions (<2.5 mm) 6-8, 14,24 .
To the best of our knowledge, TGSE is a new technique, the use of DWI which in head and neck diseases has not been reported in the literature. The basic imaging principles of the TGSE BLADE technique were introduced by Li et al 25 . In this study, compared to RESOLVE DWI, TGSE BLADE DWI signi cantly improved the image quality in cases of cholesteatoma by reducing susceptibility artifacts, distortion and blurring at the same scanning time (3 min 46 s). Moreover, TGSE BLADE DWI may be more valuable for the diagnosis of small cholesteatoma lesions (<2 mm).
Qualitative analysis of the image quality showed almost no geometric distortion or ghosting artifacts in the TGSE images, while in the RESOLVE images, obvious geometric distortion, mostly in the nasal cavity and mastoid, was present in 8 of 42 (20%) cases. Meanwhile, serious artifacts in the RESOLVE images were observed in 5 (12%) cases. Hu 18 et al. also demonstrated that TGSE BLADE DWI produced less geometric distortion in the brain and signal pile-up in highly susceptible areas than conventional SE-EPI. The image quality of TGSE BLADE has also been signi cantly improved. In the quantitative analysis of image quality, TGSE BLADE DWI showed higher contrast and a higher CNR than RESOLVE DWI, which are prospective results due to the lack of previous reports on TGSE BLADE DWI.
This study shows a signi cant difference in the ADC between the TGSE BLADE and RESOLVE sequences, with a signi cantly lower ADC of the cholesteatoma (P<0.01) and brainstem (P<0.01) on TGSE BLADE DWI than on RESOLVE DWI. The average ADC for the cholesteatoma on RESOLVE DWI was 0.928×10 -3 s/mm 2 , which is consistent with the cholesteatoma ADC (0.7-1.0×10 -3 s/mm 2 ) previously reported in the literature [32][33][34] . The mean ADC of the cholesteatoma and brainstem on TGSE BLADE DWI was 0.763×10 -3 s/mm 2 and 0.498×10 -3 s/mm 2 , respectively. These ndings demonstrate that the ADC obtained in our study can provide an auxiliary basis for more clinical applications of TGSE BLADE DWI in the future.
However, there are also some limitations to our study: the number of patients included in this study is relatively small, and the error caused by manual measurement cannot be eliminated, which may affect the accuracy of the true range of the ADC on TGSE BLADE DWI. Meanwhile, TGSE BLADE DWI is not without its disadvantages. The overall image SNR of TGSE is slightly lower than that of RESOLVE, mainly because placement of the gradient echo with T2* decay effects in the center of k-space reduces the image quality of TGSE, which is consistent with previous pediatric brain research reported by Ui 26 et al.

Conclusion
In conclusion, TGSE BLADE DWI is an artifact-free technique with better image quality than RESOLVE DWI in the diagnosis of cholesteatoma. TGSE BLADE DWI is also superior to RESOLVE DWI in terms of image distortion, artifacts and lesion conspicuity. In addition, TGSE BLADE DWI appears to be more effective in detecting small lesions.

Declarations
Ethics approval and consent to participate The study was approved by the Review Committee of Eye & ENT Hospital of Fudan University, and written informed consent was obtained from all patients. And for participants under 16 years old, the written informed consent was obtained from a parent or guardian.