The Research Of Lung Parenchyma Segmentation Algorithm Based On Superpixel

The automatic segmentation of lung parenchyma based on CT images is a key step in the computer-aided diagnosis of pulmonary diseases.The quality of segmentation of lung parenchyma directly determines the correctness of subsequent lung disease diagnosis.Because the lung CT images have the characteristics of fuzzy and intensity inhomogeneities,the traditional lung parenchyma segmentation algorithms that simply use the image intensity information are not ideal for segmentation results.Based on the super-pixel segmentation theory,in this paper,two robust lung parenchymal segmentation algorithms are proposed considering the characteristics of local regions of the image comprehensively.Following is the main work and achievements:1.We study the super-pixel segmentation theories and compare the performance of the three mainstream superpixel segmentation algorithms.The results show that the SLICO algorithm has good controllability,the resulting superpixels are uniform in size and have high tightness and boundary adherence.2.Combined with the SLICO superpixel segmentation algorithm,a segmentation framework of lung parenchyma based on superpixel clustering is proposed in this paper.Meanwhile,a texture and intensity characteristics extraction method suitable for irregular superpixels is proposed.The performance of the proposed algorithm is validated on 100 CT images of 10 patients which are obtained from the open lung dataset,i.e.kaggle.The experimental results demonstrate that the mean Dice similarity coefficient of the proposed algorithm is 96.65%,the mean over-segmentation rate is 0.51%and the mean under-segmentation rate is 2.23%respectively.Thus it can be seen that the proposed method has superior segmentation performance.3.An adaptive local contrast enhancement algorithm based on superpixel is proposed to solve the problem that the lung CT images are fuzzy and the intensities of the images are inhomogeneous.Firstly,the SLICO algorithm is used to pre-segment the CT image into multiple superpixels.Secondly,the grayscale information of the superpixel is counted.Finally,the fuzzy region is automatically positioned and the contrast of the region is improved adaptively according to the statistical results.The result of the experiment shows this algorithm can effectively improve the contrast of the blurred region while preserving the edge information.4.A novel robust lung parenchyma segmentation method is proposed by analysing the deficiency of lung parenchyma segmentation algorithm based on threshold segmentation.Firstly,the image is pre-divided into multiple superpixels by using linear iterative clustering based on the grayscale information of the image.Then,the contrast enhancement is performed by using the adaptive local contrast enhancement algorithm.Finally,refinement segmentation is performed by employing threshold and morphological operations to extract the adhesion area and lung parenchyma of the lungs accurately.The performance of the proposed algorithm is validated on 300 CT images of 30 patients which are obtained from the open lung dataset,i.e.kaggle.The experimental results demonstrate that the mean Dice coefficient of the proposed algorithm is 98.65%,the mean over-segmentation is 0.21%and the mean under-segmentation is 1.33%respectively.The overall segmentation performance is significantly improved compared with the classical threshold operation and morphological methods.