|
3. Quantitative Evaluation Method for Lung Tumor with Fractal Analysis of X-ray CT Images
Mikio Yamamoto1, Mieko Kurano1, Ryohei Ogawa1,2, Hideya Fukuoka3 Hironao Kawashima2, Hironobu Ohmatsu4, Noriyuki Moriyama5
(1NIRS; 2Keio Univ.; 3Tokyo Univ. of Agriculture; 4National Cancer Center Hospital East; 5National Cancer Center Hospital)
Keywords: lung tumor, fractal dimension, helical X-ray CT
It is important to provide a qualitative diagnosis of malignancy or benignancy in diagnosing a lesion in the lung area. Introducing image processing into the qualitative diagnosis have contributed to diagnosis for early treatment. With development of a newer X-ray CT scanner, high quality diagnosis can be expected using image processing. In this study, we used the morphological features of the intratumoral division structure and the property of the tumor limb (boundary between a tumor and the circumference of a normal lung). The quantification of the quality of the
tumor in the lung area was tried using its outside (exterior) and inside (interior) features.
We used chest X-ray CT images from the National Cancer Center Hospital. Images were photographed using a helical X-ray CT scanner. Only the lung piece was reconstructed with 512x512 pixels in an image. The slice (1mm interval) image was obtained under the conditions that the X-ray beam was 2mm wide and the CT value was in the region of -4096HU and +3046. We
looked at 15 cases (six benign examples, nine malign zed. 2) It is reasonable to suspect lung cancer on spiculation, corona radiata and a notch with deep lobation. 3) A benign tumor (hamartoma) takes on the shape of a well-defined circular limb. In order to distinguish the benign tumor from a malignant tumor, we tried quantification in the marginal region using the degree of gray-change and the shape factor. We confirmed that the gray-change was clear and the shape factor was high for a benign image, but the gray-change was complicatedly irregular and circular shape was low for a malignant image. As shown in Fig.3, the degree of gray-change can be extracted by the total number of gray level gradient directions.
For quantification in an interior tumor, differential diagnosis with the benign growth lesion is difficult for poorly differentiated adenocarcinoma and squamous cell carcinoma since they show the solid growth shadow. Dispersion and uniformity of the gray level in the intratumoral division are also noticed, making interpretation difficult. We then noted the value of each pixel of image datum, since a locationally approached pixel has a relation to other pixels. As treatment, which quantified the aspect of this arrangement, we tried quantification using a fractal dimension and entropy analysis. The fractal dimension analysis may possibly show complexity of the surface shape in a noninteger dimension, and the entropy analysis shows uniformity of the image.
We chose the central part of five slices from each case, and used a total of 75 slices. The number of the gray level gradient directions, the fractal dimension, and the entropy were placed in the three-dimensional space from the tumor division in each slice. The secondary discriminant function (three-dimensional space is divided by cutting it into a secondary curve) was used as discriminant of the benign and malignancy groups. As a result of these analyses, we were able to distinguish benign and malignant tumors at probabilities of 93.3% and 97.7%, respectively. The morphological analysis using these indexes seems to be effective for the qualitative diagnosis of the tumor. In the future, we will examine a practical application by adapting many cases with a minute analysis for an individual index.
