Segmentation Performance Comparison Considering Regional Characteristics in Chest X-ray Using Deep Learning

Abstract

Chest radiography is one of the most widely used diagnostic methods in hospitals, but it is difficult to read clearly because several human organ tissues and bones overlap. Therefore, various image processing and rib segmentation methods have been proposed to focus on the desired target. However, it is challenging to segment ribs elaborately using deep learning because they cannot reflect the characteristics of each region. Identifying which region has specific characteristics vulnerable to deep learning is an essential indicator of developing segmentation methods in medical imaging. Therefore, it is necessary to compare the deep learning performance differences based on regional characteristics. This study compares the differences in deep learning performance based on the rib region to verify whether deep learning reflects the characteristics of each part and to demonstrate why this regional performance difference has occurred. We utilized 195 normal chest X-ray datasets with data augmentation for learning and 5-fold cross-validation. To compare segmentation performance, the rib image was divided vertically and horizontally based on the spine, clavicle, heart, and lower organs, which are characteristic indicators of the baseline chest X-ray. Resultingly, we found that the deep learning model showed a 6–7% difference in the segmentation performance depending on the regional characteristics of the rib. We verified that the performance differences in each region cannot be ignored. This study will enable a more precise segmentation of the ribs and the development of practical deep learning algorithms.

Automatic Extraction of Muscle Parameters with Attention UNet in Ultrasonography

“…The Attention UNet, an improved variant of the well-established UNet [ 24 ], has been incorporated to delineate the superficial and deep aponeuroses. The original version of UNet has shown outstanding performance in biomedical problems [ 25 , 26 ], mostly due to its ability to learn from a small amount of data. The Attention UNet variant is also boosted with attention gates to highlight better salient features passed through the skip connections.…”

Motion Artifact Detection Based on Regional–Temporal Graph Attention Network from Head Computed Tomography Images

“…X-ray images of the cervical spine have been automatically segmented using the shape-aware deep segmentation network UNet-S. 'S' refers to the use of the updated shape-aware loss function [36]. Researchers have also verified that it is critical to take rib region information into account and confirmed the necessity of region-specific features in the segmentation task [37], which may benefit the classification task. It has also been demonstrated that the incorporation of prior anatomical knowledge may effectively improve the performance.…”

Cardiac Magnetic Resonance Image Segmentation Method Based on Multi-Scale Feature Fusion and Sequence Relationship Learning

“…However, cardiac MRI images are 3D sequential images, in which a set of images usually contains dozens or even hundreds of slices, and the size and shape of the left atrium in each slice are different, so segmenting a set of images involves a large and time-consuming workload. In addition, MRI images contain multiple cardiac tissue structures, and the contrast between different tissue structures is low, which means that the segmenter must possess a high level of expertise [ 4 ]. The above features increase the difficulty of cardiac MRI image segmentation.…”