<title>Recognition of lung nodules from x-ray CT images considering 3D structure of objects and uncertainty of recognition</title>

Abstract

In this paper, we propose a method of recognition of lung nodules using 3D nodule and blood vessel models considering uncertainty of recognition. Region of interest(ROI) areas are extracted by our quoit filter which is a kind of Mathmatical Morphology filter. We represent nodules as sphere models, blood vessels as cylinder models and the branches of the blood vessels as the connections of the cylinder models, respectively. All of the possible models for nodules and blood vessels are generated which can occur in the ROT areas. The probabilities of the hypotheses of the ROT areas coming from the sphere models are calculated and the probabilities for the cylinder models are also calculated. The most possible sphere models and cylinder models which maximize the probabilities are searched considering uncertainty of recognition. Tf the maximum probability for the nodule model is higher, the shadow candidate is determined to be abnormal. By applying this new method to actual CT images(37 patient images), good results have been acquired.

Recognition of lung nodule shadows from chest X‐ray CT images using 3D Markov random field models

“…(1) Case of uppermost part: The radius r CB u , the length l CB u , the curve angle θ CB c , and the cross-sectional area difference are mutually independent and follow certain Gaussian distributions. The relationships between the shape and three-dimensional location of, for example, a blood vessel thick near the septum and thin near the lung wall are represented by using values dependent on the three-dimensional location of the blood vessel as middle values of the radius and length [18]. In addition, although the original shape parameters are functions of the number of branches from the blood vessel directly connected with the heart, since in the 10-mm slice CT images that are objects of this study, a shadow is projected in the body axis direction as shown in Fig.…”

“…Artificial CT images of the same size as the rectangular areas are generated by the same method as used in Ref. 18 and the sum of the mean square differences with the pixel values corresponding to the observed CT images (referred to as the SSD value below) is computed. The SSD value represents the difference between an observed CT image and an artificial image.…”

“…The likelihood is computed from the observed CT images and the artificial CT images generated from the object model set o, as in Ref. 18. The size of an artificial CT image is made to be the same as that of VOI and the model parts outside the VOI are ignored for the same reason as given in Section 4.1.2.…”

“…The candidate shadow is determined to be a nodule, if the probability of its being a nodule is high, while the candidate shadow is determined to be normal, if this probability is low. Although the authors have previously proposed a method [18] of recognizing a candidate shadow using information on surrounding shadows, this method could only generate simple shape models. The improved method proposed in this paper can handle more complicated shape models by dividing a volume of interest into a number of rectangular areas, generating models independently in these areas, and combining the models.…”

Automatic lung nodule detection in thick slice CT: a comparative study of different gating schemes in CAD

“…While commercially lung CAD products are available, most of them operate on thin slice data (≤ 2 mm thickness) to avoid partial volume effects, overshadow of vessels and vessels oriented obliquely (lookalike nodules) [1]. High resolution CT has the potential to increase diagnostic accuracy in pulmonary nodule detection compared with singledetector helical CT.…”