An Exploratory Study to Detect Ménière's Disease in Conventional MRI Scans Using Radiomics

Abstract

Objective: The purpose of this exploratory study was to investigate whether a quantitative image analysis of the labyrinth in conventional magnetic resonance imaging (MRI) scans using a radiomics approach showed differences between patients with Ménière's disease (MD) and the control group.

Materials and methods: In this retrospective study, MRI scans of the affected labyrinths of 24 patients with MD were compared to the MRI scans of labyrinths of 29 patients with an idiopathic asymmetrical sensorineural hearing loss. The 1.5- and 3-T MRI scans had been previously made in a clinical setting between 2008 and 2015. 3D Slicer 4.4 was used to extract several substructures of the labyrinth. A quantitative analysis of the normalized radiomic image features was performed in Mathematica 10. The image features of the two groups were statistically compared.

Results: For numerous image features, there was a statistically significant difference (p-value <0.05) between the MD group and the control group. The statistically significant differences in image features were localized in all the substructures of the labyrinth: 43 in the anterior semicircular canal, 10 in the vestibule, 22 in the cochlea, 12 in the posterior semicircular canal, 24 in the horizontal semicircular canal, 11 in the common crus, and 44 in the volume containing the reuniting duct. Furthermore, some figures contain vertical or horizontal bands (three or more statistically significant image features in the same image feature). Several bands were seen: 9 bands in the anterior semicircular canal, 1 band in the vestibule, 3 bands in the cochlea, 0 bands in the posterior semicircular canal, 5 bands in the horizontal semicircular canal, 3 bands in the common crus, and 10 bands in the volume containing the reuniting duct.

Conclusion: In this exploratory study, several differences were found in image features between the MD group and the control group by using a quantitative radiomics approach on high resolution T2-weighted MRI scans of the labyrinth. Further research should be aimed at validating these results and translating them in a potential clinical diagnostic method to detect MD in MRI scans.

Keywords: 3D models; MRI; Ménière’s disease; imaging; labyrinth; quantitative image analysis; radiomics; vertigo.

Figure 1

Models of the labyrinth and its substructures. (A) Labyrinth, (B) anterior semicircular canal, (C) vestibule, (D) cochlea, (E) posterior semicircular canal, (F) horizontal semicircular canal, (G) common crus, and (H) volume containing the reuniting duct.

Figure 2

Results of the statistical analysis of the radiomic image features. Every p-value <0.05 is shown in color. Primary radiomic features are shown on the x-axis; these represent the different image processing filters. The secondary radiomic features are shown on the y-axis; these are the statistical values calculated from the primary radiomic features. A list of the primary radiomic features can be found in Table 2 and a list of the secondary radiomic features can be found in Table 3. A vertical band means that there are several significant image features in one primary radiomic feature, and a horizontal band means that there are several significant image features in one secondary radiomic feature. The figures represent one of the substructures of the labyrinth: (A) anterior semicircular canal, (B) vestibule, (C) cochlea, (D) posterior semicircular canal, (E) horizontal semicircular canal, (F) common crus, (G) volume containing the reuniting duct, and (H) p-value legend.

Figure 2

Results of the statistical analysis of the radiomic image features. Every p-value <0.05 is shown in color. Primary radiomic features are shown on the x-axis; these represent the different image processing filters. The secondary radiomic features are shown on the y-axis; these are the statistical values calculated from the primary radiomic features. A list of the primary radiomic features can be found in Table 2 and a list of the secondary radiomic features can be found in Table 3. A vertical band means that there are several significant image features in one primary radiomic feature, and a horizontal band means that there are several significant image features in one secondary radiomic feature. The figures represent one of the substructures of the labyrinth: (A) anterior semicircular canal, (B) vestibule, (C) cochlea, (D) posterior semicircular canal, (E) horizontal semicircular canal, (F) common crus, (G) volume containing the reuniting duct, and (H) p-value legend.

Figure 3

Possible relationships between image features of patients with Ménière’s disease, patients with other balance disorders, and patients with a healthy labyrinth. In this study, the yellow and green groups are investigated.