Lung region segmentation based on multi-resolution Active Shape Model

Abstract

Objective: Segmentation of lung region in chest radiographs is a key point in lung disease diagnosis. Because the complexity of anatomical structures and the overlap of organ and tissues’ intensity ranges in chest radiographs, it has set a higher demand for the lung region segmentation than the sharp-edged medical image, and the performance of low-level segmentation methods that use local intensity criteria only are not so satisfactory in chest radiographs. In this article, we presented a new segmentation method in chest radiographs. Methodology: We investigated Active Shape Model (ASM) segmentation in chest radiograph. However, the original ASM suffers from the loss of accuracy and low speed in real time applications. In this paper, automatic point insertion, semiautomatic adjustment of initial placement and multi-resolution framework were introduced to improve the performance of original ASM. Results: 80 chest radiographs have been used to test our algorithm, and the result showed that 68 images were segmented effectively. Experiments also showed that the multi-resolution search took less CPU time required by the original method because the new method converges after less iteration. Conclusions: The improved Active Shape Model provides a fast, effective, semiautomatic and model-based method for lung region segmentation in chest radiographs. Besides chest radiographs, multi-ASM can also be used in other medical images, such as CT, MRI, ultrasound images and so on.