A content-aware segmentation approach for medical images

Abstract EN

Image segmentation is required as a preprocessing step in many image-related applications.

Existing segmentation techniques can be classified into three categories: boundary-based, region-based and pixel-based direct classification.

All of these techniques segment the image according to low-level imagery features without being aware of the semantics of the segmented regions.

This paper introduces a novel approach for content aware image segmentation suitable for medical images.

Lung computed tomography (CT) is considered as a case study.

The approach utilizes semantic features rather than low-level imagery features.

The basic idea is to train the segmentation engine to recognize the normal appearance and structure of lungs so that it is able to mark any anomalous tissues occurring in a given scan.

The results are satisfactory in terms of delineation accuracy and more refinements are suggested for improvements.