Preliminary study on the automated detection of breast tumors using the characteristic features from unenhanced MR images

(1)

Preliminary study on the automated detection of breast tumors using the characteristic features from unenhanced MR images

Hayato Adachi*

^a

, Atsushi Teramoto

^a

, Satomi Miyajo

^b

, Osamu Yamamuro

^b

, Kumiko Ohmi

^b

, Masami Nishio

^c

, Hiroshi Fujita

^d

a

Graduate School of Health Sciences, Fujita Health University, 1-98 Dengakugakubo, Kutsukake-cho, Toyoake-city, Aichi 470-1192, Japan;

b

East Nagoya Imaging Diagnosis Center,

3-4-26 Jiyugaoka, Chikusa-ku, Nagoya 464-0044, Japan;

c

Nagoya Radiological Diagnosis Center, 1-162 Hokke Nakagawa-ku, Nagoya 454-0933, Japan;

d

Department of Intelligent Image Information, Division of Regeneration and Advanced Medical Sciences, Graduate School of Medicine, Gifu University,

1-1 Yanagido, Gifu 501-1194, Japan;

ABSTRACT

Breast cancer incidence tends to rise globally and the mortality rate for breast cancer is increasing in Japan.

There are various screening modalities for breast cancer, and MRI examinations with high detection rate are used for high-risk groups, which are genetically prone to develop breast cancer. In the breast MRI examination, unenhanced T1 and T2 weighted images shows no significant difference in signal value between tumor and normal tissue. Therefore, tumors are identified with use of contrast enhanced kinetic curve obtained by dynamic scan using contrast agent. Some computer aided diagnosis methods using dynamic contrast enhanced MR images also have been proposed. However, contrast agent produces the allergic reaction in rare case; it should not be used for screening examinees. Here, MRI provides the anatomical and functional information by using various sequences without contrast agents. According to the reports, this information can discriminate between tumor and normal tissue. In this study, we analyzed unenhanced MR images by using plural sequences and developed an automated method for the detection of tumors. First, we extracted the breast region from the T1-weighted image semi-automatically. Next, using the threshold determined by considering the signal intensities of tumor and normal tissue, a thresholding method was applied for diffusion-weighted image to extract the first candidate regions. After labeling processing, the breast region removes outside candidates from Initial candidates. Then false positives are reduced by the rule- based classifier. Finally, we examined the remaining candidates as possible tumor regions. We applied the proposed method to 54 cases of MR images and evaluated its usefulness. As a result, the detection sensitivity was 71.9% and the abnormal regions were clearly detected. These results indicate that the proposed method may be useful for tumor detection in unenhanced breast MR images.

Keywords: MRI, Computer-aided detection (CAD), Breast cancer

1. INTRODUCTION

Breast cancer incidence tends to rise globally and the mortality rate for breast cancer is increasing in Japan. There are various screening modalities for breast cancer, and MRI examinations with high detection rate are used for high-risk groups, which are genetically prone to develop breast cancer. In the breast MRI examination, unenhanced T1 and T2 weighted images shows no significant difference in signal value between tumor and normal tissue. Therefore, tumors are identified with use of contrast enhanced kinetic curve obtained by dynamic scan using contrast agent. Some computer

* ha.adachi.712 @gmail.com TEL: +81-562-93-9415; FAX: +81-562-93-4595

Medical Imaging 2015: Computer-Aided Diagnosis, edited by Lubomir M. Hadjiiski, Georgia D. Tourassi, Proc. of SPIE Vol. 9414, 94142A · © 2015 SPIE · CCC code: 1605-7422/15/$18 · doi: 10.1117/12.2081683

Proc. of SPIE Vol. 9414 94142A-1

(2)

aided diagnos agent produc anatomical an information c automated de

2.1 Image da Dataset consi spread diagno weighted ima weighted ima Among 54 ab 2.2 Method o Figure .1 sho semi-automat regions. The by the rule-ba

Fig.1 An ove detected usin positives are

sis methods u es the allergic nd functional can discrimin etection schem

ataset isted of 54 Jap osis. The MR age, contrast e age was 0.62 bnormal cases

overview ows the flow c

tically. Next, breast region ased classifier

erview of the ng diffusion-w

reduced by th

using dynamic c reaction in r information b nate between t me of breast tu

panese women R images were enhanced T1-w

× 0.62 × 0.9 , there were 7

chart in our pr a threshold p removes outs r.

proposed me weighed imag he rule-based c

c contrast enh rare case; it sh by using vario tumor and no umors by the a

2.

n with breast e scanned usi weighted ima 9 mm³ and th 74 tumors dete

roposed metho rocessing was side candidate

ethod. Breast r ge. Above re classifier.

hanced MR im hould not be u ous sequences

ormal tissue [ analysis of var

METHOD

MR images fr ing SIEMENS age and diffus hat of the diffu

ected by radio

od. First, the b s applied for es from Initial

region is extr esults are com

mages also hav used for screen without contr [2-4]. So the rious unenhan

DS

rom 2009 to 2 S unit (signaH

ion-weighted fusion–weight

logists.

breast region diffusion-wei l candidates. T

racted from T mbined to rem

ve been propo ning examine rast agents. A purpose of th nced MR imag

2012, examine HDxt 3T) and image. The s ed image was

is extracted fr ghted image t Then false po

T1-weighted im move the out

osed [1]. How ees. Here, MR ccording to th his study is t ges.

ed for the purp d their sequen spatial resolut

s 1.37 × 1.37

rom the T1-w to detect the f sitives (FPs) a

mage. Initial c tside candida

wever, contras RI provides the

he reports, this o develop the

pose of cancer nces were T1

ion of the T1 7 × 2.95 mm³

weighted image first candidate are eliminated

candidates are ates, and false st

e s e

r -

3-.

e e d

e e

(3)

2.3 Extractio Breast region tracking in ea points of the Above-menti the trunk bod removing the nine points fo 2.4 Initial ca After correcti images to det 2.5 False pos After labeling determined b Therefore, FP using the vo candidates ha

Fig. 2 The res weighted ima breast region

Fig

on of breast r n is extracted ach slice (Fig.

trunk body a oned processi dy region is g e trunk body f or generation o andidates dete ing the resolu tect the initial sitives reduct g processing, by the T1-we Ps inside breas

lume of cand aving the volu

sults of extrac age (a). The tru

(d).

g.3 Red points

region using T from T1-wei .2.b). Next the and the sternum

ing is applied generated by

from the who of the trunk bo ection using d ution for diffu

candidate reg tion

initial candid eighted image st regions are didate region.

ume smaller th

cting breast reg unk body (c) a

s in T1-weight

T1-weighted ighted images e slice of the t m point on th for the breast interpolating ole body (Fig.

ody are set ma diffusion-wei usion-weighted gions.

dates outsides e. However, t identified by The number han determined

gion using T1 also is extract

ted image sho images s (Fig.2.a). Fi

top of nipple he slice are ga t bottom end a among three .2.d). If the ex anually.

ghted images d image, the t

s the breast re there are app the rule-based r of pixel in d value were e

-weighted im ted. Removing

w three acqui

irst, the whol is selected. A ained; a quadr and upper end functions (Fi xtraction of b

s

thresholding m

egion are elim roximate 20 d classifier. Th

candidate reg eliminated.

mages. The who g the trunk bo

red points to g

e body region As shown in Fi

ratic function d slice. Subseq ig.2.c). The b breast region f

method is app

minated using FPs per case he rule-based gion was calc

ole body (b) i dy from the w

generate the tr

ns are obtaine ig.3, the bilate is made from quently, a curv

reast region i fails, the abov

plied for diffu

the extracted e in the elimi classifier dist culated as the

s extracted fro whole body ma

runk body reg

ed by contour eral outermos m these points ved surface as is obtained by ve- mentioned

usion-weighted

d breast region inated results tinguished FPs

e volume; the

om T1- akes the

gion.

r st s.

s y d

d

n s.

s e

(4)

Fig.4 The exp (a) is binariz positives are

The breast re evaluated by and the minim considering th determined co Because of th detection sen

Figure 5 show tumor was no buried the sig

In this study, combined T1 image and the candidates ar were 9.9 per breast MR im

This work wa

perimental res ed to make in eliminated by

egion extractio 3-fold cross- mum volume he signal inten onsidering the he low intens sitivity was 7 ws the results ot detected. B gnal intensity.

we have pro 1-weighted an e initial candi re reduced ad case. These r mages.

as granted in p

sults of detecti nitial candida y the rule-base

on and detect -validation. Th

for FP reduct nsities of tum e volume of tu sity of skin, th 1.9% and the s that the tum Because the tu

oposed a semi nd diffusion-w

dates from dif dictively with esults indicate

part supported

ing initial can ates (b).Then o

ed classifier (d

3. E

tion sensitivity hrough the va tion - were ad mor and norma umor and norm

he extraction number of FP mor was detec umor was sm

4. C

-automated sc weighted imag ffusion-weigh h the rule-bas

e that the prop

ACKN

d by a Grant-in

ndidates using outside candi d).

EXPERIME

y were evalua alidation, two djusted. The th al tissue. As a mal tissue.

of the breast Ps per case wa cted correctly mall, we guess

CONCLUSI

cheme for det ges. Proposed hted image. Af

ed classifier.

posed method

NOWLEDG

n-Aid for scie

diffusion-wei dates are rem

ENTS

ated manually parameters - hreshold value

same manner region was f as 9.9 per case . On the othe sed the rough

IONS

tecting breast d method dete fter the FP red

As a result, d d may be usefu

GMENT

ntific Researc

ighted images moved by brea

y. In addition, - the threshold

e for initial de r, minimum vo failed in 2 cas e.

er hand, Fig.6 h resolution of

tumor in une ects the breas

duction using detection sens ful for the tum

ch on Innovati

s. Diffusion-w ast region (c).

, the proposed d value for in etection was d olume for FP ses. In cross-v

6 shows the re f diffusion-we

enhanced MR t region from breast region sitivity was 7 mor detection i

ive Areas, ME

weighted image Finally, false

d method was nitial detection determined by reduction was validation, the

esults that the eighted image

R images using m T1-weighted , the remained 1.9% and FPs in unenhanced

EXT, Japan e e

s n y s e

e e

g d d s d

(5)

(a) Con Fig.5 Tumors each images.

(a) Cont Fig.6 Tumor n location of tu

ntrast enhance s detected by (a) shows the

trast enhanced not detected b umor identified

ed T1WI proposed me e exact locatio

d T1WI by proposed m

d by radiologi

(b) Diff thod. Each lin on of tumor ide

(b) Diffusion method. Red ci

ist.

fusion-weighte ne shows a di entified by rad

-weighted ima ircles show th

ed image ifferent case.

diologist.

age (c he tumor locat

(c) T Red circles s

c) Tumor cand tions in each im

Tumor candid how the tumo

didate mages. (a) sho

date

or locations in

ows the exact n

(6)

References

[1] Agliozzo, S., De Luca, M., Bracco, C., et al., “Computer-aided diagnosis for dynamic contrast-enhanced breast MRI of mass-like lesions using a multiparametric model combining a selection of morphological, kinetic, and spatiotemporal features.” Medical physics, 39(4), 1704-1715(2012).

[2] Yabuuchi, H., Matsuo, Y., Sunami, S., et al., “Detection of non-palpable breast cancer in asymptomatic women by using unenhanced diffusion-weighted and T2-weighted MR imaging: comparison with mammography and dynamic contrast-enhanced MR imaging,” European Radiology, 21(1), 11-17( 2011).

[3] Kuroki-Suzuki, S., Kuroki, Y., Nasu, K., et al., “Detecting breast cancer with non-contrast MR imaging:

combining diffusion-weighted and STIR imaging,” Magnetic Resonance in Medical Sciences, 6(1), 21- 27( 2007).

[4] Ito, R., Ikeda, Y., Otake, K., et al., “Study on Use of MR-mammography in Health Check-ups,” Ningen Dock International, 26(1), 56-61(2011), in Japanese.