Comparison of image segmentation of lungs using methods: connected threshold, neighborhood connected, and threshold level set segmentation

13th South-East Asian Congress of Medical Physics 2015

物理学;医药卫生

Amanda, A.R.^1 ; Widita, R.^1

Department of Physics, Faculty of Mathematic and Natural Science, Institut Teknologi Bandung, Indonesia^1

关键词: Mean Square Error (MSE);    Original images;    Performance evaluations;    Segmentation methods;    Signal noise;    Threshold levels;    Threshold methods;   

Others  :  https://iopscience.iop.org/article/10.1088/1742-6596/694/1/012048/pdf
DOI  :  10.1088/1742-6596/694/1/012048

学科分类：卫生学

来源: IOP

【 摘 要 】

The aim of this research is to compare some image segmentation methods for lungs based on performance evaluation parameter (Mean Square Error (MSE) and Peak Signal Noise to Ratio (PSNR)). In this study, the methods compared were connected threshold, neighborhood connected, and the threshold level set segmentation on the image of the lungs. These three methods require one important parameter, i.e the threshold. The threshold interval was obtained from the histogram of the original image. The software used to segment the image here was InsightToolkit-4.7.0 (ITK). This research used 5 lung images to be analyzed. Then, the results were compared using the performance evaluation parameter determined by using MATLAB. The segmentation method is said to have a good quality if it has the smallest MSE value and the highest PSNR. The results show that four sample images match the criteria of connected threshold, while one sample refers to the threshold level set segmentation. Therefore, it can be concluded that connected threshold method is better than the other two methods for these cases.