PlumX Metrics
Embed PlumX Metrics

A Novel Approach to Detect Anatomical Landmarks Using R-CNN for MEG-MRI Registration

SSRN Electronic Journal
  • 1
    Citations
  • 219
    Usage
  • 0
    Captures
  • 0
    Mentions
  • 0
    Social Media
Metric Options:   Counts1 Year3 Year

Metrics Details

  • Citations
    1
    • Citation Indexes
      1
  • Usage
    219
    • Abstract Views
      178
    • Downloads
      41

Article Description

In a general scenario, magnetoencephalography (MEG) and magnetic resonance imaging (MRI) registration is performed by manually marking the fiducial (external) markers placed at the individual’s left pre-auricle (LPA), right pre-auricle (RPA), and nasion. The drawback of using external markers can lead to inaccurate registration due to subjective error, and registration cannot be performed on MRI, which doesn’t have markers in it. In the proposed study, we introduce a novel approach that automatically detects LPA, RPA, and nasion in MR images without using any fiducial markers. Since LPA and RPA are not anatomically appealing in MR images, we detected the neighboring structure that has a well-defined anatomical relationship with LPA and RPA, such as Temporomandibular Joint (TMJ), using a region-based Convolutional neural network (R-CNN). Using the anatomical relationship between the nasion and pre-auricular point, we detected nasion using image processing techniques. Using these three anatomical landmarks, we performed MEG-MRI registration on forty-three subjects. To validate the proposed study, we performed the localization of the N20 component derived from Somatosensory Evoked Fields (SSEF) of five subjects. The handcrafted R-CNN model was successful in detecting TMJ. The result showed that registration error (in millimeters) between proposed automated registration and manual registration is 3.6028 ± 1.4037, 4.0512 ± 1.736, and 2.7118 ± 2.7942 for LPA, RPA, and nasion, respectively. The validation results obtained for five subjects showed the localization error to be less than 1 millimeter. From the results, we can say that the proposed MEG-MRI registration can replace fiducial-based MEG-MRI registration.

Bibliographic Details

Pooja Prabhu; Karunakar A. Kotegar; N. Mariyappa; Anitha H; G. K. Bhargava; Jitender Saini; Sanjib Sinha

Elsevier BV

Anatomical Landmarks; Convolutional Neural Network (CNN); Magnetoencephalography (MEG); Magnetic Resonance Imaging (MRI); Multimodality Registration; Neuroimaging

Provide Feedback

Have ideas for a new metric? Would you like to see something else here?Let us know