We had presented that by using a special combination of variables, the accuracy in distinguishing between benign and malignant tumors was increased.
Variables such as mean SUV and SUVmax of PET, area of CT and its mean, and the texture characteristics from GLCM were applied with GA and SVM classifier to produce the most appropriate factors for differentiation.
Our current study implemented a total of 26 variables to apply with GA and SVM. The co-occurrence matrix had four distance vectors which would generate four times the numbers of variables for operation. In order to prevent over-training due to the small number of patients, we needed to operate the variables in four distance vectors separately. The number of variables which is much than the total number of image would decrease the accuracy rate.
The future direction is that researchers can investigate whether the adding of more screening variables can produce a better combination of variables. Besides texture feature, other features such as shape features, statistical features and characteristics of the relevant position can also be taken into account. These will make computer aided diagnosis (CAD) easily to distinguish the nodules which could not be diagnosed with unaided eyes.
For further implementation, the combination of variables can be applied in medical imaging diagnosis. Physicians can manually select any suspicious nodules from PET/CT images and let the computer compute the screened variables immediately to determine whether the particular solitary nodule is benign or malignant. The accuracy of diagnosis can thus be improved and it will
bring physicians benefits in PET/CT imaging cancer diagnosis.
45
References
方惠珊，(2007)，肺腺癌細胞之電腦輔助圖形辨識，碩士論文，國立雲林科 技大學資訊管理系
行政院衛生署 Department of Health, Executive Yuan, R.O.C. (TAIWAN)，101 年死因統計結果分析
(2013/06/06)，http://www.doh.gov.tw/CHT2006/DisplayStatisticFile.aspx
?d=88577&s=1
陳玥汝，(2012)，使用透視法於肺部PET/CT DICOM 影像之三維立體顯示，
碩士論文，國立高雄大學資訊工程所碩士班
曾林維，(2006)，Texture analysis of computed tomography images in acute middle cerebral artery ischemic stroke (利用紋理分析於電腦斷層影像偵 測急性缺血性中大腦動脈中風)，碩士學位論文，中原大學醫學工程學 系
黃俊忠，(2006)，The mass detection in mammography using feature images (利 用X光乳房攝影產生之紋理特徵影像在腫瘤偵測上之研究)，碩士論文，
國立中央大學資訊工程研究所
楊勝智，(2006)，Computer-aided system for diagnosis of masses in breast medical images (乳房醫學影像之腫瘤電腦輔助診斷系統)，博士論文，
國立成功大學電機工程學系
鄭 煥 勳 ， (2006) ， Feature analysis for lung nodule in dynamic contrast enhancement CT image (肺臟腫瘤於動態顯影CT影像之特徵分析)，碩士 論文，中原大學醫學工程學系
賴鄭婷，(2008)，Integrating the PSONN and Boltzmann function for feature
selection and classification of Lymph node in ultrasound images(結合粒子 群體最佳化神經網路與波茲曼函數應用於超音波淋巴結影像之最佳化 特徵選取及分類)，碩士論文，國立雲林科技大學資訊工程研究所碩士 班
Albregtsen, F., (2008), Statistical texture measures computed from gray level coocurrence matrices, Image Processing Laboratory Department of Informatics University of Oslo.
Clausi, D.A., (2002), An analysis of co-occurrence texture statistics as a function of grey level quantization, Can. J. Remote Sensing, 28, 45-62.
da Silva Sousaa, J.R.F., Silvaa, A.C., de Paivaa, A.C., Nunesb, R.A., (2010), Methodology for automatic detection of lung nodules in computerized tomography images. Computer Methods and Programs in Biomedicine, 98, 1-14.
Hall, M.N., Jagannathan, J.P., Ramaiya, N.H., Shinagare, A.B. and Van den Abbeele, A.D., (2010), Imaging of Extraosseous Myeloma: CT, PET/CT, and MRI Features, American Journal of Roentgenology, 195, 1057-1065.
Han, D., Bayouth, J., Song, Q., Taurani, A., Sonka, M., Buatti, J. and Wu, X., (2011), Globally optimal tumor segmentation in PET-CT images: a graph-based co-segmentation method. Information Processing in Medical Imaging, 22, 245-256.
Haralick, R.M., (1979), Statistical and structural approaches to texture, Proceedings of the IEEE, 67, 786-804.
Haralick, R.M., Shanmugam, K. and Dinstein, I., (1973) Textural features for
47
image classification, IEEE Transactions on Systems, Man and Cybernetics, SMC-3, 610-621.
Hu, S. and Reinhardt, J.M., (2001), Automatic lung segmentation for accurate quantitation of volumetric X-Ray CT images. IEEE Transactions on Medical Imaging, 20, 490-498.
Khalaf, M., Abdel-Nabi H., Baker, J., Shao, Y., Lamonica D. and Gona J., (2008), Relation between nodule size and 18F-FDG-PET SUV for malignant and benign pulmonary nodules, Journal of Hematology &
Oncology.
Li, C., Xu, C., Gui, C. and Fox, M.D., (2005), Level set evolution without re-initialization: a new variational formulation, IEEE Computer Society Conference on Computer Vision and Pattern Recognition.
Marom, E.M., Erasmus, J.J. and Patz Jr, E.F., (2000), Lung cancer and positron emission tomography with fluorodeoxyglucose, Lung Cancer, 28, 187–202.
Masa-ah, P., Tuntawiroon, M., Soongsathitanon, S., (2010), A novel scheme for Standardized Uptake Value (SUV) calculation in PET scans.
International Journal of Mathematical Models and Methods in Applied Sciences, 4, 291-299.
McNitt-Gray, M.F., Hart, E.M., Wyckoff, N., Sayre, J.W., Goldin, J.G., Aberle, D.R., (1999), A pattern classification approach to characterizing solitary pulmonary nodules imaged on high resolution CT: preliminary results, The International Journal of Medical Physics Research and Practice, 26,
880-888
McNitt-Gray, M.F., Wyckoff, N., Sayre, J.W., Goldin, J.G., Aberle, D.R., (1999), The effects of co-occurrence matrix based texture parameters on the classification of solitary pulmonary nodules imaged on computed tomography, Computerized Medical Imaging and Graphics, 23, 339-348.
Mitchell, M., (1999), An Introduction to Genetic Algorithm, Cambridge, Massachusetts: Massachusetts Institute of Technology.
Moreno-Torres, J.G., Sáez, J.A. and Francisco Herrera, (2012), Study on the impact of partition-induced dataset shift on k-fold cross-validation, IEEE Transactions on Neural Networks and Learning Systems, 23, 1304-1312.
Potesil, V., Huang, X. and Zhou, X.S., (2007), Automated tumour delineation using joint PET/CT information, Medical Imaging 2007: Computer-Aided Diagnosis, 6514.
Rangayyan, R.M., Guliato, D., de Carvalho,J.D, and Santiago, S.A., (2006) Feature extraction from the turning angle function for the classification of contours of breast tumors. IEEE Special Topic Symposium on Information Technology in Biomedicine, Iaonnina, Greece, 4 pages CDROM.
Soh, L.K. and Tsatsoulis, C., (1999), Texture analysis of sar sea ice imagery using gray level co-occurrence matrices, IEEE Transactions on Geoscience and Remote Sensing, 37, 780-795.
Zsoter, N., Bandi, P., Szabo, G., Toth, Z., Bundschuh, R.A., Dinges, J. and Papp, L., (2012), PET-CT based automated lung nodule detection, 34th Annual International Conference of the IEEE EMBS.
49
Appendix A
The table below shows the description of malignant nodules, starting from the left to right: the original CT image, image after preprocessing and the corresponding PET image of CT image.
a hypermetabolic cavitated nodule in right upper lung
size: 2.2 cm, SUV: 4.8
lung mass in the right upper lung size: 3.3 cm, SUV: 6.6
pulmonary nodule in the right upper lung
size: 2.0 cm, SUV: 4.8
a lung nodule in the left upper lung size: 2.3 × 1.5 × 1.7 cm, SUV: 6.8
a hypermetabolic nodule in the right upper lung
3.7 × 2.2 × 2.1 cm, SUV: 6.0
a lung nodule in the left upper lung size: 2.2 cm, SUV: 5.3
a lung mass in the left upper lung size: 4.3 × 3.5 × 2.8 cm, SUV: 11.5
a hypermetabolic tumor in the right upper lung
size: 2.7 × 2.4 × 2.7 cm, SUV: 8.6 small pulmonary nodule in the right middle lung,
size: 0.4 cm
a pulmonary nodule in the right lower lung
size: 1.2 cm, SUV: 4.3
a pulmonary mass in the left lower lung
size: 3.5 cm, SUV: 5.9
a pulmonary nodule in the left upper lung
size: 2.5 cm, SUV: 4.7
lung nodule in the RUL size: 1.7 cm, SUV: 3.8
a lung mass in the right upper lung size: 3.5 × 2.4 × 2.3 cm, SUV: 5.0
51
a hypermetabolic nodule in the left upper lung
size: 2.4 cm, SUV: 2.9
a hypermetabolic mass in the right upper lung
4.0 × 3.1 × 2.7 cm in size, SUV: 2.2 a pulmonary nodule in the right upper lung
size: 1.7 cm, SUV: 3.1
a lung nodule in the left upper lung size: 1.5 cm, SUV: 6.1
a lung nodule in the left upper lung size: 1.6 cm, SUV: 3.7
a FDG-avid lung mass in the right upper lung
size: 3.4 × 2.3 × 2.4 cm, SUV: 11.7 a small pulmonary nodule in the right lower lung, adjacent pulmonary tissue (0.8cm)
a pulmonary lesion in the right upper lung
size: 1.2 cm, SUV: 1.4
a pulmonary nodule in the RLL size:
1.3 cm, SUV: 4.1
two pulmonary nodules one in the LLL
size: 0.9 cm, SUV: 1.4
a hypermetabolic nodule in the right upper lung
size: 4.3 × 3.3 × 3.3 cm, SUV: 7.6
a lung tumor in the left upper lung size: 3.5 × 2.8 × 2.4 cm, SUV: 4.4
a hypermetabolic nodule in the left upper lung
size: 2.0 × 1.6 × 1.9 cm, SUV: 6.2
lung mass in RML
size: 4.3 × 3.1 × 2.7 cm, SUV: 10.3
a lung tumor in the right upper lung size: 3.5 × 2.7 × 2.5 cm, SUV: 10.4
53
Appendix B
The table below shows the description of benign nodules, starting from the left to right: the original CT image, image after preprocessing and the
corresponding PET image of CT image.
sella mass, with volume 2.64 ml, length 2.0 cm, SUV (initial) 8.28
a hypermetabolic nodule in the left lower lung, 2.9 × 2.1 × 1.7 cm in size, SUVmax: 3.8
a nodule, 1.0 cm, SUV 1.2, in left lower lung
a lung mass in the right lower lung size: 4.7 × 3.0 × 2.5 cm, SUV: 4.3
in LUL mass [4.8cm], with volume 6.44 ml, SUV 0.88
a patchy lesion in the right upper lung size: 1.1 cm, SUV: 0.8
ground glass lesion in the right middle lung
left upper lung is absent due to previous operation
a hypermetabolic mass in the right upper lung
size: 4.1cm, SUV; 9.0
a hypermetabolic nodule in the right upper lung, 2.4 cm in size, SUV: 4.2
in the 1.0 cm LLL lung mass,with volume 0.39 ml, SUV 0.44
seen on CT images, with
indistinguishable FDG uptake from adjacent pulmonary tissue
one is in the RML (size: 0.7 cm) and another is in the RUL (size: 0.4 cm)
A lobulated nodule in right lower lung size: 3 cm, SUV: 2.4
55
Multiple enlarged lymph nodes, up to 1 cm SUV up to 4.8, in right lower interlobar, both hilar para-tracheal
a lung nodule in the right upper lung size: 1.5 cm, SUV: 3.6
a pulmonary nodule in the left upper lung
size: 1.5 cm, SUV: 6.2
a tumor in left upper lung size: 1.8 cm, SUV 6.8
a furhter increase of FDG, SUV 8.2, is noted on delayed scan.
a patch of mild FDG uptake, SUV 1.4, in right middle lung.
in RUL nodule, with volume 1.16 ml, length 1.3 cm, SUV 1.75
in left upper lung lesion size: 2 cm, SUV 2.3,
A patchy infiltration with mild FDG uptake, SUV 2.2, in superior segment of left lower lung.
pulmonary lesions in the right lower lung
size: 2.0 cm, SUV: 1.1
pulmonary lesions in left lower lung size: 0.8 cm, SUV: 1.3
a lung mass in the right lower lung size : 6.2 × 4.1 × 3.2 cm, SUV: 3.7
nodule in right lower lung size : 1.8 cm, SUV 2.5
a small RLL lung nodule size : 0.9 cm, SUV: 1.7
relatively decreased cerebral cortical uptake of FDG is observed.
a pulmonary nodule in the left upper lung
size: 1.1 cm, SUV: 1.0
57
a lung nodule in the right lower lung size: 1.5 cm, SUV: 1.0
pulmonary nodule in the right upper lung
size: 1.5 cm, SUV: 2.6
a lung mass in the right middle lung size: 3.6 × 2.6 × 3.1 cm, SUV: 4.9.
a nodule in left lower lung size: 2.2 cm, SUV 4.5
a hypermetabolic nodule in the right main bronchus
size: 3 × 4 × 2.4cm, SUV 10.3
in RUL lung nodule, with volume 2.23 ml, SUV: 4.08