Near-optimal Keypoint Sampling for Fast Pathological Lung Segmentation

TL;DR

This paper introduces a fast and reliable lung segmentation method from CT scans that combines region-based segmentation with local descriptor classification on an optimized sampling grid, improving accuracy and speed.

Contribution

It presents a novel two-stage segmentation approach that integrates fuzzy connectedness and local texture descriptors with near-optimal keypoint sampling for pathological lung delineation.

Findings

Method is fast and robust in segmenting pathological lungs.

Improves accuracy over existing standards.

Potential to enhance clinical routine tasks.

Abstract

Accurate delineation of pathological lungs from computed tomography (CT) images remains mostly unsolved because available methods fail to provide a reliable generic solution due to high variability of abnormality appearance. Local descriptor-based classification methods have shown to work well in annotating pathologies; however, these methods are usually computationally intensive which restricts their widespread use in real-time or near-real-time clinical applications. In this paper, we present a novel approach for fast, accurate, reliable segmentation of pathological lungs from CT scans by combining region-based segmentation method with local descriptor classification that is performed on an optimized sampling grid. Our method works in two stages; during stage one, we adapted the fuzzy connectedness (FC) image segmentation algorithm to perform initial lung parenchyma extraction. In the second stage, texture-based local descriptors are utilized to segment abnormal imaging patterns using a near optimal keypoint analysis by employing centroid of supervoxel as grid points. The quantitative results show that our pathological lung segmentation method is fast, robust, and improves on current standards and has potential to enhance the performance of routine clinical tasks.