07_特集6_再校_0516.indd

Size: px

Start display at page:

Download "07_特集6_再校_0516.indd"

もえりあると
5 years ago
Views:

1 158 MEDICAL IMAGING TECHNOLOGY Vol. 35 No. 3 May 2017 * * PET/CT PET/CT PET/CT CT PET CT 10 PET/CT FROC PET/CT Med Imag Tech 35 3 : , magnetic resonance imaging MRI computed tomography CT positron emission tomography/computed tomography PET/CT CT MRI PET PET PET/CT teramoto@fujita-hu.ac.jp PET/CT CT PET PET/CT PET CT computer aided diagnosis CAD CAD 2 2 CAD CAD CAD 3

2 Med Imag Tech Vol. 35 No. 3 May CAD 4 PET/CT PET CAD 5, 6 PET CT CAD 7, 8 CT 9, 10 PET/CT CAD PET/CT c 2 a b z c xy 3 2. PET/CT 1 CT PET B cube z Z B1 Z B2 xy Y B c 2 CT CT L L A L B 11 3 z Th bone CT 3

4 y x 4 x sx ex y 4 sx y 4 y Y B Y B xy 5 z Z B1 Z B2 xy 5 B cube T Th

3 160 Med Imag Tech Vol. 35 No. 3 May xy a 5 Z B Z B2 L B 4 xy xy 12, 13 Th bone CT y x 4 x sx ex y 4 sx y 4 y Y B Y B xy 5 z Z B1 Z B2 xy 5 B cube T Th breast T A B cube A T L B final 2 1 SUV PET PET standardized uptake value SUV 14 SUV = 2 [Bq/g] [Bq]/ [g] 1 PET 15 2

zx f zy f zz ) 2 2 3 λ 1 λ 2 λ 3 Z dot Z dot (λ 1, λ 2, λ 3 )={ λ 3 2 / λ 1, λ 1 < 0, λ 2 < 0, λ 3 < 0 0,

4 Med Imag Tech Vol. 35 No. 3 May a PET b 7 SUV c 8 SUV a PET b SUV c SUV a b 2 2 f σ H 2 H = ( f xx f xy f xz f yx f yy f yz f zx f zy f zz ) λ 1 λ 2 λ 3 Z dot Z dot (λ 1, λ 2, λ 3 )={ λ 3 2 / λ 1, λ 1 < 0, λ 2 < 0, λ 3 < 0 0, otherwise 3 PET Z dot 6 4 PET SUV I (x, y, z)=i(x, y, z)+e mass (Z dot (x, y, z)) 4 4 I PET SUV I SUV PET SUV Z dot E mass SUV E mass 7 M 1 M 2 E 1 E 2 PET SUV 8 8 b c SUV 3 SUV PET Th d 9 SUV 4 CT PET 9 c d 5

162 Med Imag Tech Vol. 35 No. 3 May 2017 V FP x y z L B (L B L A ) 0.3 L B x y z Z B1 3. 9 a PET b c d 1 10 10 11 31 42 3 True Point Biograph 40 CT 0.97 0.97 2.00 mm 3 PET 4.07 4.07 2.00 mm 3 10 25.

5 162 Med Imag Tech Vol. 35 No. 3 May 2017 V FP x y z L B (L B L A ) 0.3 L B x y z Z B a PET b c d True Point Biograph 40 CT mm 3 PET mm ±9.07 mm SUV 4.44±3.07 Th bone = 150 [H.U.] Th breast = 500 [H.U.] V FP 19 pixel SUV 20 mm SUVmax 1.5 σ H = 1.5 M 1 = 10 7 M 2 = E 1 = 1.0 E 2 = mm 20 mm SUV Visual studio Th d FROC 3 FROC c

6 Med Imag Tech Vol. 35 No. 3 May a PET/CT b 12 a PET/CT b 13 a PET/CT b c 10 FROC a b c c 11 SUVmax

7 164 Med Imag Tech Vol. 35 No. 3 May mm SUV V FP 5. PET CT , p : , CAD 23: 19-26, Yoshikawa R, Teramoto A, Matsubara T, et al.: Automated detection of architectural distortion using improved adaptive Gabor filter. In Fujita H, Hara T, Muramatsu C, eds., Breast Imaging, Lecture Notes in Computer Science vol The 12th International Workshop on Breast Imaging, Springer, Cham, 2014, pp Guan H, Kubota T, Huang X, et al.: Automatic hot spot detection and segmentation in whole body FDG- PET images. In proceedings of the 13th IEEE International Conference in Image Processing, Atlanta, 2006, pp Ballangan C, Wang X, Fulham S, et al.: Lung tumor segmentation in PET images using graph cuts. Comput Methods Programs Biomed 109: , Teramoto A, Fujita H, Takahashi K, et al.: Hybrid method for the detection of puimonary nodules using positron emission tomography/computed tomography. Int J CARS 9: 59-69, Teramoto A, Fujita H, Takahashi K, et al.: Automated detection of pulmonary nodules in PET/CT images: Ensemble false-positives reduction using a convolutional neural network technique. Med Phys 43: , : CT Med Imag Tech 26: , : 3 X CT Med Imag Tech 31: 62-71, Teramoto A, Fujita H: Fast lung nodule detection in chest CT images using cylindrical nodule-enhancement filter. Int J CARS 8: , CT 22: , CT Med Imag Tech 30: 43-52, Keys JW: SUV: Standard uptake or silly useless value? J Nucl Med 36: , Li Q, Sone S, Doi K, et al.: Selective enhancement filters for nodules, vessels, and airway walls in two- and three-dimensional CT scans. Med Phys 30: , 2003

8 Med Imag Tech Vol. 35 No. 3 May Preliminary Study on an Automated Extraction of Breast Region and Automated Detection of Breast Tumors and Axillary Metastasis Using PET/CT Images Natsuki MINOURA 1, Atsushi TERAMOTO 1, Katsuaki TAKAHASHI 2, Osamu YAMAMURO 2, Masami NISHIO 3, Tsuneo TAMAKI 2, Hiroshi FUJITA 4 1 Graduate School of Health Sciences, Fujita Health University 2 East Nagoya Imaging Diagnosis Center 3 Nagoya Radiological Diagnosis Center 4 Graduate School of Medicine, Gifu University Positron emission tomography PET and X-ray computed tomography CT are used for the localization and analysis of breast cancer and axillary metastasis. In this study, we develop a method for the automated detection of breast tumors and axillary metastasis in PET/CT images. Our scheme extracts the breast region, which includes axilla, from CT images and then detects high-uptake regions inside the breast region from PET images. First, a bounding box is calculated for the breast and the axilla using bone and lung information obtained from CT images. Second, high-uptake regions are detected in PET images using massive structure enhancement and thresholding. The areas outside the breast regions are excluded from initial candidate regions. False positives FPs are eliminated using the location and the shape of initial candidate regions before obtaining final candidate regions. In our experiments, we evaluated tumor detection ability of the proposed method. Breast regions were identified and extracted correctly in all cases. Sensitivity of tumor detection was 0.76 with a number of FPs/case of 3.9. These results indicate that the proposed method may be useful for breast tumor and axillary metastasis detection using PET/CT images. Key words: PET/CT, Breast cancer, Automated detection, Region extraction, Hessian matrix Med Imag Tech 35 3 : , 2017

9 166 Med Imag Tech Vol. 35 No. 3 May PET CAD X X CT A PET PET CT CAD 2017 IEEE SPIE * * *

SICE東北支部研究集会資料（2017年）

SICE東北支部研究集会資料（2017年） 307 (2017.2.27) 307-8 Deep Convolutional Neural Network X Detecting Masses in Mammograms Based on Transfer Learning of A Deep Convolutional Neural Network Shintaro Suzuki, Xiaoyong Zhang, Noriyasu Homma,