A hybrid method based on CNNs and edge-based active contour models for medical image segmentation

Abstract EN

Edge-based active contour models have been one of the most prominent and influential approaches in image segmentation.

It has been proven that they are very effective when they are applied on images with inhomogeneous intensity.

Traditionaledge-stop functions (ESFs) are usually used when edges are defined by the image gradient.

They often produce weak edgesbecause they fail to stop at the precise boundary.

In this work, a new approach integrating machine learning algorithm withedge-based model using a level set method (LSM) is proposed.

The ESF is constructed from a convolutional neural network.

Then it is applied to an edge-based active contour model.

The proposed method has been applied on medical images.

Obtainedresults have been compared to those given by k-nearest neighbors and support vector machine to confirm the effectivenessof the proposed method.