automatic cochlea multi-modal images segmentation · 2018-04-03 · automatic cochlea multi-modal...

Automatic Cochlea Multi-modal Images Segmentation

Al-Dhamari, CI2018

Automatic Cochlea Multi-modal Images Segmentation Using Adaptive Stochastic Gradient Descent

Ibraheem Al-Dhamari, Sabine Bauer, Roland Jacob and Dietrich Paulus

Active Vision Group

Institute of Medical Technology and Information Processing

March 2018

1

Automatic Cochlea Multi-modal Images Segmentation

Al-Dhamari, CI2018

Disclosure

2

There is no relevant conflict of interest related to this presentation.

Automatic Cochlea Multi-modal Images Segmentation

Al-Dhamari, CI2018

Introduction: Problem Definition

- A practical technique to get cochlea segmentation from medical images is needed.

- Doctors need cochlea measurement before the surgery to select the best implant for a patient.

3

Image source: Blausen.com staff (2014). "Medical gallery of Blausen Medical 2014". WikiJournal of Medicine

- Medical image analysis based on well segmented cochlea may provide these measurements.

Automatic Cochlea Multi-modal Images Segmentation

Al-Dhamari, CI2018

Introduction: Cochlea Segmentation

4

Input image Segmentation

Automatic Cochlea Multi-modal Images Segmentation

Al-Dhamari, CI2018

Introduction: Cochlea Segmentation Challenges

5

-  Low resolution images, the best spacing we have is [0.125, 0.125, 0.50] mm for CBCT.

-  Cochlea is a small organ with a complicated structure. Cochlea scalas are not visible in the clinical images.

-  Different types of images CT, MRI and CBCT.

Automatic Cochlea Multi-modal Images Segmentation

Al-Dhamari, CI2018

Methods: Adaptive Stochastic Gradient Descent

6

-  An optimizer for medical image registration proposed by Klein et al [1].

[1] Klein et al: Adaptive stochastic gradient descent optimisation for image registration. International Journal of Computer Vision, 81(3):227-239,(2009).

-  We used it in our method ACIR [2] to solve the problem of cochlea multi-modal image registration and fusion.

[2] Ibraheem Al-Dhamari et al, ACIR: automatic cochlea image registration. Proceedings of SPIE Medical Imaging 2017, Image Processing. 2017;10133(10):47–67.

-  The proposed method is to use ACIR in an model-atlas-based segmentation to solve the problem of cochlea segmentation.

CBCT-CT MR-CT CBCTb-CTBCTa

Automatic Cochlea Multi-modal Images Segmentation

Al-Dhamari, CI2018

Methods: Adaptive Stochastic Gradient Descent

7

-  ASGD is implemented in elastix, a standard medical image registration tool,[3] which is available as a public open-source.

[3] http://elastix.isi.uu.nl

-  ACIR and this work are implemented as 3D Slicer, a standard medical image processing tool, [4] plugins which is also available as a public open-source.

[4] https://www.slicer.org

CBCT-CT MR-CT CBCTb-CTBCTa

Automatic Cochlea Multi-modal Images Segmentation

Al-Dhamari, CI2018

Methods: Cochlea Atlas-based Segmentation

8

Automatic Cochlea Multi-modal Images Segmentation

Al-Dhamari, CI2018

Methods: Cochlea Model

9 [5] Gerber et al, A multiscale imaging and modelling dataset of the human inner ear, Scientific Data 4, 170132 (2017).

-  A high resolution mCT cochlea image from a public dataset [5] is prepared and used.

-  The mCT segmentation is registered to a clinical cochlea image. -  The cochlea clinical image and its segmentation are used as a model.

Automatic Cochlea Multi-modal Images Segmentation

Al-Dhamari, CI2018

Methods: Cochlea Model

10

-  A high resolution mCT cochlea image from a public dataset [5] is prepared and used.

-  The mCT segmentation is registered to a clinical cochlea image. -  The cochlea clinical image and its segmentation are used as a model.

[5] Gerber et al, A multiscale imaging and modelling dataset of the human inner ear, Scientific Data 4, 170132 (2017).

Automatic Cochlea Multi-modal Images Segmentation

Al-Dhamari, CI2018

Sample results

11

CBCT MRI CT

Automatic Cochlea Multi-modal Images Segmentation

Al-Dhamari, CI2018

Conclusion

- A model-atlas-based method for automatic cochlea multi-modal image segmentation is proposed.

12

- The proposed method segments a cochlea clinical image and generates a 3d model in less than 3 seconds.

- Using higher resolution histology model may provide better segmentation.

Automatic Cochlea Multi-modal Images Segmentation

Al-Dhamari, CI2018 13

Thanks