Hiroko Kitaoka

Integrated system for CT-based assessment of parenchymal lung disease

High-resolution X-ray CT imaging can provide detailed structural and functional information about the respiratory system. We describe a system to facilitate quantitative regional assessment of the lung parenchyma from volumetric CT images. The system contains image segmentation algorithms to identify the key anatomic structures of the lung and image analysis algorithms to examine the lung parenchymal tissue. The system can be used to study the normal lung and to detect changes in the lung tissue due to disease processes such as emphysema.

Automated Nomenclature Labeling of the Bronchial Tree in 3D-CT Lung Images

A nomenclature labeling algorithm for the human bronchial tree down to sub-lobar segments is proposed, as a means of inter and intra subject comparisons for the evaluation of lung structure and function. The algorithm is a weighted maximum clique search of an association graph between a reference tree and an object tree. The adjacency between nodes in the association graph is defined so as to reflect the consistency between the bronchial name in the reference tree and the node connectivity in the object tree. Nodes in the association graph are weighted according to the similarity between two tree nodes in the respective trees. This algorithm is robust to various branching patterns and false branches that arise during segmentation processing. Experiments have been performed for nine airway trees extracted automatically from clinical 3D-CT data, where approximately 250 branches were contained. Of these, 95 % were accurately named.

Quantitative evaluation of spatial distribution of line structure in the lung for computer-aided diagnosis of pulmonary nodules

In the discrimination of benign and malignant pulmonary nodules, it is considered very important to evaluate the spatial distribution of the line structure around the nodule. For quantitative evaluation, a computer must be used. This paper proposes a method in which line structures, such as vessels, in the chest CT image are extracted, and their spatial distribution is quantitatively evaluated. First, the line structures in the three-dimensional CT image are extracted, and the direction of the central line is estimated from the eigenvector of the Hessian matrix. The angle between the direction vector of a point on the central line and the vector directed from that point to the nodule center is defined as the “centripetal angle.” By a statistical analysis of the distribution of the centripetal angle, the spatial distribution of the line structure is estimated. In the proposed method, only the variation of the nodule center can affect the result. Consequently, stable operation is expected with a simple manipulation. The method is applied to 20 clinical cases, and satisfactory classification results are obtained. The effects of the image reconstruction algorithm and of parameter values such as the threshold of line structure extraction on the results of the proposed method are examined, and its robustness and range of application are determined.