|Year : 2017 | Volume
| Issue : 3 | Page : 171-175
Comparison of manual and computer-assisted measurement of cochlear nerve obtained from magnetic resonance imaging
S Jeevakala1, A Brintha Therese1, Rajeswaran Rangasami2
1 School of Electronics Engineering, VIT University, Chennai Campus, Chennai, Tamil Nadu, India
2 Department of Radiology and Imaging Sciences, Sri Ramachandra University, Chennai, Tamil Nadu, India
|Date of Web Publication||31-Aug-2017|
Department of Radiology and Imaging Sciences, Sri Ramachandra Medical College and Research Institute, Chennai - 600 116, Tamil Nadu
Source of Support: None, Conflict of Interest: None
Objectives: Cochlear implantation of sensorineural hearing loss requires accurate measurement of cochlear nerve (CN). The precise measurement of very small structures can be improved by automated segmentation and measurement. The variability and reproducibility of the computer-assisted measurement were compared with manual measurements. Materials and Methods: The 3D-constructive interference in steady state magnetic resonance imaging (MRI) images of twenty patients who were referred for MRI in the assessment of giddiness/vertigo or fitness for cochlear implant during the period from June 2013 to June 2014 were analyzed. The CN sizes were measured manually by two independent radiologists. The automatic measurements were then performed on the same images, and its correlation and agreement were calculated between automated and manual measurements. Results: The intra-observer correlation coefficients were significantly larger for cross-sectional area (CSA) of CN using automated measurements when compared to manual measurement (intra-observer r: 0.94021 vs. 0.91437). Similarly, the inter-observer correlation for CSA of CN is also higher in automated measurements (inter measurement r: 0.94786 vs. 0.92013). Conclusions: Using computer-assisted CN dimension measurement, the intra- and inter-observer correlation can be improved when compared to manual measurements. The automated measurement can assist the radiologist in eliminating the need for tedious manual tracing and thus, the time and effort for manual segmentation are also significantly reduced.
Keywords: Cochlear implantation, cochlear nerve, computer-assisted measurement
|How to cite this article:|
Jeevakala S, Therese A B, Rangasami R. Comparison of manual and computer-assisted measurement of cochlear nerve obtained from magnetic resonance imaging. Indian J Otol 2017;23:171-5
|How to cite this URL:|
Jeevakala S, Therese A B, Rangasami R. Comparison of manual and computer-assisted measurement of cochlear nerve obtained from magnetic resonance imaging. Indian J Otol [serial online] 2017 [cited 2019 Sep 21];23:171-5. Available from: http://www.indianjotol.org/text.asp?2017/23/3/171/213875
| Introduction|| |
Sensorineural hearing loss (SNHL) is considered one of the major childhood inabilities affecting 6 in 1000 children. Cochlear implantation (CI) has turned out to be progressively common in pediatric clinical practice for the treatment of congenital or profound SNHL. Measuring the size of cochlear nerve (CN) with parasagittal magnetic resonance imaging (MRI) can yield information which could be helpful for the preoperative advising of patients. An accurate assessment of CN size is essential for the successful CI. Nadol and Xu  measured the diameter of the CN in hearing impaired people based on autopsy studies. Fatterpekar et al. evaluated the length and width of the CN associated with congenital SNHL using computed tomography imaging of the temporal bone on controlled experiments. Sildiroglu et al. measured the cross-sectional area (CSA) of the CN in elderly patients with SNHL using 3D-constructive interference in steady state (CISS) MRI imaging. Russo et al. reported CN diameter of patients with profound SNHL by high-resolution MRI of the temporal bone in vivo techniques. Nakamichi et al. measured the normal size of CN and facial nerve (FN) using 3D-CISS magnetic resonance (MR) images. Komatsubara et al. explained the need of measuring the size of CN in the congenital SNHL. With the development of computerized CN size measurement methods, it is hypothesized that it might enhance the assessment of CI by significantly lessening the intra-/inter-observer variability with manual measurement. The study was intended to assess the impact of intra-/inter-observer variability of manual and automatic measurements on observer agreement in the CI.
| Materials and Methods|| |
Study design and patients
This study is to compare the CN size obtained by manual and automatic measurements of patients in the assessment of CI. The 3D-CISS images of inner ear were obtained from same MRI scanner during the period from 2013 June to 2014 June. Institutional Ethics Committee approval was obtained, and informed consent was waived.
Magnetic resonance imaging protocol
The images were obtained using 1.5 tesla MRI scanners (Siemens Avanto, Erlangen). The parameters for the MRI CISS sequence were field of view - 120 mm, slice thickness - 0.9 mm, voxel size - 193 × 256, repetition time - 2.79 ms, and echo time - 6.62 ms. Initially, axial CISS images of the inner ear were obtained. Then, oblique sagittal images were planned perpendicular to the VIII nerve for each side.
Magnetic resonance imaging evaluation by radiologist
All the MR images results were reviewed by two independent radiologists, who were blinded to patient's information. The diameter of the CN was measured at the fundus of the internal auditory canal (IAC) where the edge of CN could be identified. The parasagittal image at the middle of IAC was selected to visualize the CN, FN, vestibular nerve. The long diameter (LD) and short diameter (SD) were measured to calculate CSA using the formula (π [LD/2] [SD/2]). The measurement of LD, SD was obtained on parasagittal images with 0.01 mm electronic caliper provided by General Electric workstation.
Automated measurement methods
The main purpose of this study is to measure the size of CN using fully automated software developed using MATLAB tool. The original image is preprocessed to remove noise and other artifacts. Then, the region of interest (ROI) image containing IAC is manually extracted by the expert radiologists. The resolution of the ROI was improved by Lanczos-2 interpolation. [Figure 1]a and [Figure 1]b show the ROI before and after enhancement, respectively. The interpolated ROI was subjected to segmentation using region growing algorithm. The binary output of segmentation is shown in [Figure 1]c shows the segmented CN, FN, and vestibular cochlear nerve (VCN) in the IAC. The automated CN area of a region in a binary image can be found by multiplying the number of pixels in the region and the MRI voxel size, which is recovered from the Digital Imaging and Communications in Medicine information header. [Figure 1]d,[Figure 1]e,[Figure 1]f show the long diameter, short dimeter, and cross-sectional area of IAC and its nerves.
|Figure 1: (a) Region of interest image (b) enhanced image (c) segmented binary output, (d-f) are long diameter, short diameter, and cross-sectional area measurements, respectively, obtained using automated measurement|
Click here to view
For clinical inner ear MR imaging study, the mean LD, SD, and CSA of CN were calculated. Statistical analysis was performed using paired t-test to compare the automated and manual measurement. Pearson correlation coefficient is used to evaluate the intra-/inter-observer correlation between the automated and manual measurements of CN. Linear regressions were also used to access the intra-/inter-observer correlations. Bland-Altman plot with 95% limits of agreement (LoA) was used to access the continuous agreement between the automated and manual measurements. All the statistical analysis was performed using Microsoft Excel (Microsoft) and origin software (OriginLab). For all test, P < 0.05 was considered statistically significant.
| Results|| |
Of the twenty patients (including both right and left ears), the LD, SD, and CSA of CN were measured manually and automatically, and the comparisons of two images are shown in [Figure 2].
|Figure 2: Manual and automated measurement of cochlear nerve (a and d) manual measurement, (b and e) automated long diameter, (c and f) automated short diameter|
Click here to view
Cochlear nerve measurements
The images examined for CN had a manually measured mean of LD = 1.00 ± 0.15 mm, SD = 0.71 ± 0.11 mm, and CSA = 0.59 ± 0.14 mm 2. The automated measured mean of LD = 1.01 ± 0.17 mm, SD = 0.75 ± 0.08 mm, and CSA = 0.60 ± 0.12 mm 2. There is no significant difference between the automated and manually measured dimensions of CN (P < 0.05).
Intra-/inter-observer correlation and variations
There was good intra- and inter-observer correlation between the manual and automated measurements (r > 0.94, P < 0.01). From [Table 1], it is found that intra-/inter-observer correlation is significantly better in automated area measurement than manual measurement. In addition, [Figure 3] shows the correlation between the manual and automated measurement. As presented, the correlation is stronger in automated measurement of CSA than in manual measurement. The one-way ANOVA analysis of manual measurement of CSA shows [Table 2] the inter-observer variability of (F = 0.61636; P = 0.43571), whereas the automatic measurement was not influenced by inter-observer variability (F = 0.3017; P = 0.58498).
|Figure 3: Regression line are fitted with scatter plots (a) manual cross-sectional area measurement (b) automated cross-sectional area measurement|
Click here to view
Intra- and inter-observer agreement
The Bland–Altman plot [Figure 4] shows that the intra-observer agreement is varied from −0.13 mm 2 to 0.09 mm 2 with the bias of −0.02 mm 2, whereas the inter-observer agreement varies from −0.13 mm 2 to 0.11 mm 2 with the bias of −0.01 mm 2. The 95% LoA between manual and automatic measurement for CSA is varied from −0.08 mm 2 to 0.05 mm 2 with the bias of −0.01 mm 2. The bias value of intra-/inter-observer and manual versus automatic CSA are −0.02 mm 2, −0.01 mm 2, and −0.01 mm 2, respectively. Since bias values are closer to zero the manual and automatic measurements are in agreement. From [Figure 4] as presented it is seen more agreement between the automated measurement and manual measurement.
|Figure 4: Bland–Altman plots showing the agreement between observers (a) intra-observer agreement (manual) (b) intra-observer agreement (automatic) (c) agreement between manual and automatic measurement|
Click here to view
| Discussion|| |
The most important error in the manual measurement of the CN size is the variability between the intra- and inter-observer. Therefore a computer-assisted measurement techniques would substantially decrease reader induced errors in the assessment of CI.
In this study, the reproducibility of automatic measurement over manual estimation is obviously exhibited for the evaluation of CI. The computer-assisted algorithm for segmentation and measurement of the CN is another innovation. The automated methods need to be differentiated from the manual measurement based on the intra-/inter-observer variability and reproducibility. Nakamichi et al. measured manually the normal diameter of the CN and FN. The inter-observer reliability of automated strategy is better than manual method. Kona measured the bony canal of CN in SNHL patients and found the variation in diameter of CN in unilateral SNHL patients. Gyuang et al. reported that the mean CSA of CN tends to decrease with age and size of the CN positively correlated with the auditory performance after CI. Measuring the size of the subtle CN is usually much more challenging task due to the inhomogeneous enhancement at edges. Therefore, automated measurement could successfully adopt for the assessment of CI.
Both computer-assisted measurements (diameter and area) of the CN were less variable compared with the manual assessments. The underlying principle of the method is simple and quick. However, a careful check of the segmentation is still needed: in the instance of an over or under-segmentation of the CN. Manual corrections or new parameter of segmentation were chosen in cases of over-segmentation into neighboring structures. Under-segmentations were principally observed in extremely heterogeneous CN regions with inhomogeneous edge enhancement.
The intra- and inter-observer variability of automatic measurements were less when compared to manual measurements. The Bland–Altman plot shows significantly higher inter-observer agreement for automated versus manual measurement in CSA of CN. Based on the results of 95% LoA between manual and automated measurements varied from −0.14% to 0.08% with the bias of −0.03% for CSA measurements. The 95% limits of inter-observer agreement of CSA varied from −0.22% to 0.16% with the bias of −0.06%. The 95% limits of intra-observer agreement of CSA varied from −0.26% to 0.15% with the bias of −0.06%. Since the limits are within the clinically acceptable range (<20%), the reproducibility of automatic measurement is same as manual measurement.
Except for the motion artifacts images, the algorithm is able to segment effectively. Furthermore, the intra-observer variability of the same image was not calculated because the measurement was not expected to be varied. Although the software can provide the measurement of FN, VCN, and IAC those measurements were not included in this study. The computer-assisted measurement of area and diameter can further warrant the evaluation of pathological conditions associated with FN, VCN and IAC.
| Conclusions|| |
The result of the study demonstrates the excellent reproducibility of computer-assisted dimension measurement of CN over manual measurement. Since the intra- and inter-observer correlation is reduced in automated measurement, it helps the radiologist in the assessment of CI. The study is limited with few number of patients and can be improved by a large number of patients and its classification of pathology based on CN size.
The authors would like to thank the Sri Ramachandra Medical Center and Research Institute, Porur, India for providing the clinical real MR images.
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
| References|| |
Adunka OF, Roush PA, Teagle HF, Brown CJ, Zdanski CJ, Jewells V, et al.
Internal auditory canal morphology in children with cochlear nerve deficiency. Otol Neurotol 2006;27:793-801.
Casselman JW, Offeciers EF, De Foer B, Govaerts P, Kuhweide R, Somers T. CT and MR imaging of congential abnormalities of the inner ear and internal auditory canal. Eur J Radiol 2001;40:94-104.
Nadol JB Jr., Xu WZ. Diameter of the cochlear nerve in deaf humans: Implications for cochlear implantation. Ann Otol Rhinol Laryngol 1992;101:988-93.
Fatterpekar GM, Mukherji SK, Alley J, Lin Y, Castillo M. Hypoplasia of the bony canal for the cochlear nerve in patients with congenital sensorineural hearing loss: Initial observations. Radiology 2000;215:243-6.
Sildiroglu O, Cincik H, Sonmez G, Ozturk E, Mutlu H, Gocgeldi E, et al.
Evaluation of cochlear nerve size by magnetic resonance imaging in elderly patients with sensorineural hearing loss. Radiol Med 2010;115:483-7.
Russo EE, Manolidis S, Morriss MC. Cochlear nerve size evaluation in children with sensorineural hearing loss by high-resolution magnetic resonance imaging. Am J Otolaryngol 2006;27:166-72.
Nakamichi R, Yamazaki M, Ikeda M, Isoda H, Kawai H, Sone M, et al.
Establishing normal diameter range of the cochlear and facial nerves with 3D-CISS at 3T. Magn Reson Med Sci 2013;12:241-7.
Komatsubara S, Haruta A, Nagano Y, Kodama T. Evaluation of cochlear nerve imaging in severe congenital sensorineural hearing loss. ORL J Otorhinolaryngol Relat Spec 2007;69:198-202.
Kono T. Computed tomographic features of the bony canal of the cochlear nerve in pediatric patients with unilateral sensorineural hearing loss. Radiat Med 2008;26:115-9.
Kim BG, Chung HJ, Park JJ, Park S, Kim SH, Choi JY. Correlation of cochlear nerve size and auditory performance after cochlear implantation in postlingually deaf patients. JAMA Otolaryngol Head Neck Surg 2013;139:604-9.
[Figure 1], [Figure 2], [Figure 3], [Figure 4]
[Table 1], [Table 2]