Intensity Confusion Matters: An Intensity-Distance Guided Loss for Bronchus Segmentation

TL;DR

A novel Intensity-Distance Guided loss function is introduced, which assigns adaptive weights to different image voxels for mining hard samples that cause the intensity confusion, and is verified that tackling the intensity confusion issue helps to significantly improve bronchus segmentation.

Abstract

Automatic segmentation of the bronchial tree from CT imaging is important, as it provides structural information for disease diagnosis. Despite the merits of previous automatic bronchus segmentation methods, they have paied less attention to the issue we term as \textit{Intensity Confusion}, wherein the intensity values of certain background voxels approach those of the foreground voxels within bronchi. Conversely, the intensity values of some foreground voxels are nearly identical to those of background voxels. This proximity in intensity values introduces significant challenges to neural network methodologies. To address the issue, we introduce a novel Intensity-Distance Guided loss function, which assigns adaptive weights to different image voxels for mining hard samples that cause the intensity confusion. The proposed loss estimates the voxel-level hardness of samples, on the basis of the following intensity and distance priors. We regard a voxel as a hard sample if it is in: (1) the background and has an intensity value close to the bronchus region; (2) the bronchus region and is of higher intensity than most voxels inside the bronchus; (3) the background region and at a short distance from the bronchus. Extensive experiments not only show the superiority of our method compared with the state-of-the-art methods, but also verify that tackling the intensity confusion issue helps to significantly improve bronchus segmentation. Project page: https://github.com/lhaof/ICM.

Figures (4)

• Figure 1: Intensity distribution and priors. (a) displays the intensity distribution of misclassified voxels. (b) and (c) illustrate two intensity priors for estimating sample hardness in and out of bronchus, respectively. The harder and the easier regions of airway are surrounded by red, orange, yellow boundaries.
• Figure 2: Overview of the proposed intensity-distance guided loss. The bronchus, hard and easy regions are surrounded by red, orange and yellow boundaries, respectively. The upper part and lower part generates the intensity-based weight map and distance-based weight map, respectively.
• Figure 3: The original label and our re-labeled result. The bronchus region is in red.
• Figure 4: Qualitative analysis with incorrect segmentations highlighted in yellow, shows that our model outperforms others trained with UNet, including cl-Dice, Focal, and RD losses. The Focal loss model tends to overestimate the bronchus size by mistaking background for foreground, suggesting that merely weighting the foreground is insufficient.