Edit this page

NA-MIC Project Weeks

Back to Projects List

Refinement of the method used to determine surgical class based on the shape of the carotis syphon

Key Investigators

Project Description

This is an ongoing project from last year.

Stroke is a leading cause of death worldwide of which ischaemic stroke is the more common. Mechanical thrombectomy involves inserting a catheter into the cerebral vasculature to remove blood clot. Catheter devices with different parameters are available to perform the procedure of which the correct one must be selected beforehand to avoid blockage. Clinical experience suggests that large lumen aspiration catheters were most commonly stuck at the anterior curvature of the carotid syphon.

We categorised 53 studies into four groups. Previously, we extracted nine features based on vessel geometry for classification purposes.

Objective

  1. Objective A. Our main objective is to refine the extracted attribute values in order to enhance the classification results.
  2. Objective B. Vessel segmentation is also part of the process that is performed manually currently. We are trying to make it automatic.

Approach and Plan

  1. Using Weka, figure out what features and classification method provide the best result.
  2. We plan to gather CT and ground truth data for MONAI Auto3dSeg segmentation training.

Progress and Next Steps

  1. Absolute distance values were normalized by computing the ratio of distance relative to the distance along the centreline of the vessel.
  2. Starting from the first cross section, difference values were computed from absolute vessel cross section area values.
  3. Attributes were inspected using Weka functions.

Results: Even the new attributes are not really a good descriptor of best choice available as ground truth.

Next steps: Ground truth should be considered in a different way. Instead of opting for one single choice, a percentage value could be assigned to each intrumentation. From the recorded surgical log data, we know which devices were tested and which were unsuccessful. These unsuccessful instrument applications could also be used to train the classification method.

Illustrations

Background and References

No response