Вы находитесь на странице: 1из 4

36 1 Jpn. J. Med. Phys. Vol. 36 No.

1: 3538 (2016)

*1 2
1

2

Applications of Machine Learning for Radiation Therapy


Hidetaka ARIMURA*1, Takahiro NAKAMOTO2
1
Division of Medical Quantum Science, Department of Health Sciences, Faculty of Medical Sciences, Kyushu University
2
Division of Medical Quantum Science, Department of Health Sciences, Graduate School of Medical Sciences, Kyushu University

Radiation therapy has been highly advanced as image guided radiation therapy (IGRT) by making advantage of image engineer-
ing technologies. Recently, novel frameworks based on image engineering technologies as well as machine learning technologies
have been studied for sophisticating the radiation therapy. In this review paper, the author introduces several researches of appli-
cations of machine learning for radiation therapy. For examples, a method to determine the threshold values for standardized up-
take value (SUV) for estimation of gross tumor volume (GTV) in positron emission tomography (PET) images, an approach to esti-
mate the multileaf collimator (MLC) position errors between treatment plans and radiation delivery time, and prediction
frameworks for esophageal stenosis and radiation pneumonitis risk after radiation therapy are described. Finally, the author intro-
duces seven issues that one should consider when applying machine learning models to radiation therapy.

Keywords: image guided radiation therapy, machine learning, outcome prediction

f()
1.
xunknown yf(xunknown)

(image guided radiation therapy: IGRT)
1, 2CBCT (cone-beam com- GTV (gross tumor volume)
puted tomography) X PET (positron emission tomography) SUV
(standardized uptake value)
(multileaf collimator: MLC)



3
12
1

xlearn y
yf(xlearn)
2.GTV
f()
Kerhet (non-small-cell lung can-
cer: NSCLC) GTV
PET SUV70 SUVmax 70%
6GTV PTV (planning
target volume)
1 Kerhet

* 8128582 311
Division of Medical Quantum Science, Department of Health Sciences, Faculty of Medical Sciences, Kyushu University, 311
Maida, Higashi-ku, Fukuoka 8128582, Japan
E-mail: arimurah@med.kyushu-u.ac.jp

35
36 1

SUV70 istic) ANN area under ROC curve


SUV 0.1, (AUC) 0.85 mean lung
0.15, 0.2 6 doseV20V25V30Lyman NTCP (normal tissue compli-
SUV70 cation probability)Kutcher NTCP
ROC AUC 0.5
6 SUV70 ANN
SVM Lee (Bayesian network:
(support vector machine) BN) 9BN
Kerhet Jaccard index (JI) GTV
6

GTV

0.82 JI BN
SUV GTV 10Lee

DVH
Lee
ROC AUC 0.83
3.MLC

Carlson VMAT (volumetric
5.
modulated arc therapy) MLC 7
DICOM-RT (digital imaging Atsumi
and communications in medicine for radiation therapy)
MLC dynamic
MLC log (DynaLog) 11 T14stage I-III

(simple linear regres- (CR: complete remission)
sion model), (multiple linear regres- 109
sion model), (model based on 3
the random forest algorithm), Cubist (cubist mod-
el based on the cubist algorithm)
ANN
T stage
(p0.031, p0.0001, p
DVH (dose 0.0011)
volume histogram) ANN 2
DVH
DVH 0.864
Carlson

4.

Su
8
(articial neural network: ANN)
DVH VD(D10, 20,
30, 40, 50, 60, 70, 80)
(n26) 1 (n 2
116) 1 ROC (receiver operating character-

36
Jpn. J. Med. Phys. Vol. 36 No. 1 (2016)

(p0.001)T stage


6.
1 Troost EG, Thorwarth D, Oyen WJ: Imaging-based treat-
ment adaptation in radiation oncology. J. Nucl. Med. 56:
19221929, 2015
Kang 2 Baumann M, Krause M, Overgaard J, et al.: Radiation
12 oncology in the era of precision medicine. Nat. Rev. Can-
cer 16: 234249, 2016

3 Lin T, Cervio LI, Tang X, et al.: Fluoroscopic tumor


tracking for image-guided lung cancer radiotherapy.
(1)DVH Phys. Med. Biol. 54: 981992, 2009
4 El Naqa I1, Bradley JD, Lindsay PE, et al.: Predicting ra-

diotherapy outcomes using statistical learning tech-
niques. Phys. Med. Biol. 54: S9S30, 2009
5 Naqa IE, Deasy JO, Mu Y, et al.: Datamining approaches
(2) for modeling tumor control probability. Acta Oncol. 49:
13631373, 2010
DVH 6 Kerhet A, Small C, Quon H, et al.: Application of machine
learning methodology for PET-based denition of lung
(3) cancer. Curr. Oncol. 17: 41, 2010
7 Carlson JN, Park JM, Park SY, et al.: A machine learning
approach to the accurate prediction of multi-leaf collima-
(4) tor positional errors. Phys. Med. Biol. 61: 25142531,
2016
8 Su M, Miften M, Whiddon C, et al.: An articial neural
network for predicting the incidence of radiation pneumo-
nitis. Med. Phys. 32: 318325, 2005
9 Lee S, Ybarra N, Jeyaseelan K, et al.: Bayesian network
ensemble as a multivariate strategy to predict radiation
(5) pneumonitis risk. Med. Phys. 42: 24212430, 2015
10

41: 483492, 2003
11 Atsumi K, Shioyama Y, Arimura H, et al.: Esophageal
stenosis associated with tumor regression in radiation
therapy for esophageal cancer: frequency and prediction.
Int. J. Radiat. Oncol. Biol. Phys. 82: 19731980, 2012
(6) 12 Kang J, Schwartz R, Flickinger J, et al.: Machine learn-
ing approaches for predicting radiation therapy outcomes:
(7) A clinician s perspective. Int. J. Radiat. Oncol. Biol. Phys.
93: 11271135, 2015
ANNSVM
13 Bishop C: Pattern recognition and Machine Learning.
2006, Springer.
14 Suzuki K, Zhang J, Xu J: Massive-training articial neu-
7. ral network coupled with Laplacian-eigenfunction-based
dimensionality reduction for computer-aided detection of
polyps in CT colonography. IEEE Trans. Med. Imaging
13 MTANN (massive-training 29: 19071917, 2010
15 Anthimopoulos M, Christodoulidis S, Ebner L, et al.:
articial neural network)14 15
Lung pattern classication for interstitial lung diseases
using a deep convolutional neural network. IEEE Trans.
Med. Imaging 2016 Feb 29.Epub ahead of print

37
36 1

38

Вам также может понравиться