Lung air estimation using non-invasive techniques can be used for assessment of the lung function and effective diagnosis of lung disease (e.g. chronic obstructive pulmonary disease). A novel technique is proposed to automatically estimate the lung air volume and its variations throughout respiration cycle using 4D thoracic CT images. The technique is based on an advanced thresholding method which benefits from parenchyma tissue mechanical properties in addition to the lung image data over the respiration cycle to determine the segmentation threshold values. This led to an optimization framework which was developed to find optimal values of the segmentation threshold. In general, image intensity of each voxel in a lung CT image is the intensity resultant of tissue and air within the voxel. This gives rise to the known partial volume effect. An important feature of the proposed technique is its treatment of this effect to improve the volume estimation accuracy. For this purpose, voxels’ Air Volume Portion Coefficients (AVPCs) are calculated then multiplied by corresponding voxel size and voxel number which is determined using the optimization algorithm. The technique was applied to CT scans of ex vivo porcine lung and to lung images of COPD patients. Results indicate that the air volume estimated using the proposed method is significantly more accurate when AVPCs are used in the air volume estimation as errors of only ∼ 1% and less than 6.7% were obtained in the ex vivo lung and patient studies, respectively. Moreover, the proposed method demonstrates a robust performance for low resolution CT images where accuracy of other methods is known to decrease. The performance characteristics of the proposed method, including being completely image based, fully automatic combined with its demonstrated high accuracy makes it a strong candidate to complement/replace pulmonary function diagnostic tests that involve experimental measurement of function parameters. Moreover, being fully computational without involving any manual steps indicates its adaptability to be utilized as a core component within an effective expert system framework for lung disease diagnosis.
Moghadas-Dastjerdi, H., Ahmadzadeh, M., & Samani, A. (2017). Towards computer based lung disease diagnosis using accurate lung air segmentation of CT images in exhalation and inhalation phases. Expert Systems with Applications, 71, 396–403. https://doi.org/10.1016/j.eswa.2016.11.013