692

Quantification of Gray/White Matter in Neonates and Adults B. c. P. Lee, 1 B. Kneeland ,1 R. J . R. Knowles,1 and P. T. Cahill1.2

Quantitation of gray I white matter is important in evaluation of cerebral blood flow, atrophy, and development of the brain. First-order statistical analysis of neonatal computed tomo­graphic (eT) images revealed that there was only a 6 Hounsfield unit (H) difference between gray and white matter compared with the observed 3 H for the standard deviation over the field of a skull water phantom . Scene segmentation methods based on first-order statistics proved unsuccessful in separating gray and white matter. A new regional clustering algorithm based on local textural properties was developed for separation of these struc­tures.

For the evaluation of cerebra l blood fl ow, atroph y, and neuro log ic development of the neonatal brain it is of great importance to be able to quantitate gray and white matter on computed tomographic (CT) scans. In adults and older infants white matter is well myelin­ated and has less photoelectric absorpt ion than gray matter [1]. In premature neonates white matter has little myelin but high water content, which accounts for lower Hounsfield numbers [2]. Although visual differences can readily be observed in gray and white matter, the quantification of the relative amounts of g ray and white matter (using first-order statistical image segmentation methods) has so far proven elusive. This is due in part to small CT density differences, the complex interdigitation of gray I white matter , the inherent noise spectrum present in CT scanners, and partial volume effects. In addition, statist ica l variations at gray I white matter borders greatly complicate the local topog raphy; thus scene segmentation of gray I white matter would require greater spatial resolution and signal-to­noise ratio than are currently available. In brief , the reason that visual differences in gray and white matter are observed is higher­order textural differences, not first-order statistica l differences.

Th e pu rpose of this study was to develop methods of statistical analys is of CT attenuation value that use these textural d ifferences to convert what is visualized into quantitative values of gray I white matter.

Materials and Methods

Cranial CT was performed in eight neonates (1 ,500 g or less) and six adults using a GE CT IT 8800. Scanning technique factors

were 480-700 mAs at 120 keV. Reconstru cted CT images were transferred to magnetic tape and analyzed on a Varian V76 com­puter with 256 kbytes memory and 9. 6 Mbytes disk space. All images were processed either as the orig inally reconstru cted 320 x 320 matrices or as zoomed 64 x 64 matrices. Displays (statis­tica lly coded images or three-dimensional relief maps) were ob­tained using a Statos elec trostatic printer I plotter.

To investigate the sensitivity of our new texture-discrimination scene-segmentation algorithms, several phantoms were developed to test textural quantification of gray I white matter from CT scans. These inc luded plexiglass, contrast material in water, and various mixtures of graduated pasta particles of 0 .1-5 mm . A neonatal skull fill ed with water was used to measure instrumental noise over the range of miliamperage values c linically used. Formalin-fi xed patho­logic sections were photog raphed and d igitized as well .

CT images of the phantoms and gray I white matter were analyzed for their first- and higher-order statistical properties. First-ord er parameters were the first, second , and third moments of local 16 x 16 matrices. For th e higher-order parameters, the magnitude and direction of the local gradient, the energy, inertia, and corre­lation were calculated for overlapping and nonoverlapping local matrices, wh ich varied from 2 x 2 to 4 x 4 in size. To determine the differences in both first- and second-order parameters , the changes in gray and white matter were examined on CT images where regions could be definitely called gray or white by visual

inspection. Since white matter predominates over gray matter, most CT images contained several independent white-matter regions of 16 x 16 pixels (normal image, 320 x 320), while normally only one such reg ion could be assigned to gray matter. In addition to these statistical parameters, the two-dimensional frequency spec­trum was calculated using a fast-Fourier transformation and the power spectrum was displayed as two-dimensional relief maps. The auto power spectrum for each gray I white region also was calculated for the skull , water, and texture phantoms.

A regional-clustering algorithm was developed to separate gray and white matter. Because white matter predominates in CT trans­verse images, the regional c lustering algorithm is initially started inside a white-matter region . The regional-c lustering algorithm then proceeds outward , bound by two constraints: the white region must exhibit simple connectivity, and the region must conform to the smooth topology imposed by anatomic struc ture. Subject to these two ph ysical constraints, a thresho ld determined by the local first-

'Department of Rad iology, New York Hospital-Cornell Medical Center , 525 East 68th Street, New York, NY 1002 1. Address reprint requests to B. C. P. Lee. ' Polytechnic Institute of New York , Brooklyn, NY 11201.

AJNR 4 :692- 695, May / June 1983 0 195-6108/ 83 / 0403-0692 $00.00 © American Roentgen Ray Society

AJNR:4 . May / June 1983 PEDIATRICS 693

Fig . 1.-A. Relief map of flood field of water-filled sku ll phantom . Note non­uniformity of field . B. Graph shows vari­ation in CT numbers (X ax is) vs. to tal number of pixels (Y ax is) of flood fie ld before and after filtration .

I.~ ; .

"1

, .. .. I f 1

~ II I II I I

II II FI L TERED l!' = \. '1 I Ilj I I I II I I ';;' I ~' I ""

II ,/ II """ RA-;JW 15"" = 3.,

I,' I ...

" ~' f" ',~., ,., j \, .... \

.·· .... 1 .... .... :::. . ..... '" I, . ; .... } ..... ~ -""',': , ,',. ~1 · I \. ( ,-I I' -1'", . I I

- 'I ~ ;" ,::. ' .. ;: II II

HISTOGRAM OF CT FLOOD FIELD CT NUMBER VARIATION

A

A B Fig. 2. -A. Plain CT of neonatal brain distinguishes between gray and

white matter. B. Zoomed magnified image of right upper quadrant. Margin between gray and white matter is no longer obvious.

order averages of the 16 x 16 matrices can be applied to local areas of the CT image. Normally, 3 x 3 local matrices were found to be optimal, for the regional-clustering algorithm was repeated one to three times until the image stabilized. The regional-c lustering algorithm considers the differences in noise spectrum or texture between the gray and white matter and operates on white reg ions, which are more uniform in texture [3].

To account for variations in equ ipment performance for d ifferent studies, each patient served as his own control. The functional parameters used in th e regional c lustering were determined for each study using unambiguous local regions of both white and gray matter.

Results

Spatial Resolution and Noise Spectrum of CT Scanner

The resolution and intrinsic instrumental noise spectrum was determined using a plexiglass GE water phantom [4]. Variations of

B

A B Fig. 3. - CT section of adu lt brain . A. Plain CT. Gray and wh ite matter are

distinct. but margins are ind istinct. B. Postcontrast CT . Gray / white matter now clearer. but textu re of wh ite matter is reduced compared with A.

about 10 Hounsfield units (H) across the fi eld were normally ob­served in the water phantom with 3-5 SO. Using a pediatric skull phantom fill ed with water, the observed standard deviation var ied from 2.8 to 3 .5 H depending on the mi lliamperage values (fig. 1). This reduction is in part due to the shaped bow-tie filter where a better image reconstruc tion is achieved due to part ial correction of beam hardening . When the reg ional-c lustering algorithm was ap­plied to the skull phantom, the standard deviation was further reduced to 1.4- 1.6 H.

Gray / White Differences in CT Density

Although gross regional differences in gray and white matter can be read ily discerned on CT slices wi th appropriate window settings . visual scene segmentat ion on zoomed sections by several d ifferent observers was unsuccessful because of th e complex interdigitations between white and gray matter (fig . 2) . Measu rements in reg ions visually identified as gray or white showed about 6 H separating the

694 PEDIATRICS AJNR:4 , May/ June 1983

first moments of the white and gray zones. Since the standard deviation of the skull phantom fields was 3 H, separation depending on first-order statistics was therefore exceed ingly difficu lt , consid­ering the partial-volume effect of the scanner and the complex topography of gray / white matter. Stud ies with intravenous contrast material showed that although the wh ite matter is distinctly en­hanced visually, the average difference in Hounsfield numbers between the gray and white matter is not inc reased (fig. 3). Th ere is also a reduction in the white-matter texture.

Therefore , quantitative scene segmentation based on first-order differences cou ld not be made either with or without intravenous contrast media. On the other hand , it is a well known fact that th e human visual system is sensitive to second- and third-order statis­tics. For these reasons, methods of image processing that use more sophisticated scene segmentation methods involving higher-order statistics are needed to quantitate the relative amounts of gray / wh ite matter.

Fig. 4 .- Texture phantom. Conica l plex iglass cylinder contains matter of similar CT numbers but different textures.

A B

Scene Segmentation Based on Textural Differences

A direct comparison of the CT slices with patholog ic sect ions proved more difficult than orig inally expected due to scaling of the images and d ifficulty in matching the CT sect ions with the anatomic sections. The pathologic slices were digitized as 128 x 128 matri­ces, and matching them to the 320 x 320 CT image's anatomic landmarks proved exceedingly difficult . Again , this is because the boundaries between the gray and white matter are very complex. Currently, a 512 x 512 digitizing system is being evaluated for matching with 320 x 320 CT images . Scaling is being performed by appropriate zoom ing of known anatomic distances determined by th e size of the ventricles and the outer border of the brain.

In order to test the ability of the regional-c lustering algorithm to quantitate gray / white matter differences, cy lindrical plexiglass phantoms were constructed so that the central conical defect could be filled with material that had Hounsfield numbers within 3- 6 H of plexiglass, but had sign ificantly different textural properties (fig . 4). (The conical shape permits an analys is of the effect iveness of both first-order and higher-order scene-segmentat ion algorithms.) Our validated nearest-neighbor edge-detection algorithm [5) was unable to separate the edges of the cone on a first-order test. The criterion for scene segmentation is the linearity of the two conical edges, which is not dependent on partial-volume effects, milliamperage values, or fi eld uniformity , as shown in figure 5. On th e other hand , our regional-clustering algorithm functioning on the differences between the texture properties of the outside plexiglass and inside con ical zone cou ld easily delimit this boundary.

In figure 6 , a zoomed (64 x 64) statist ica l map is shown of the CT numbers of the left frontal lobe of figure 3. Each symbol differs by 3H, which is consistent with the observed standard deviation of the noise level in the skull phantom. At the statist ica l con fid ence level of 1 SO, separation of gray/ white structure is not evident; however, if our reg ional-clusteri ng algorithm is applied to the same data, the typical gray / white separation observed on pathologic sect ions is c learly demonstrated.

Discussion

Because CT number variation caused by instrumental noise is very c lose to the difference between gray- and white-matter read­ings, quantitative separation of these structures based on first-order statist ics (using methods such as the nearest-neighbor edge-detec­tion algorithms) was unsuccessful , even though it is visually possible

aooe00 00000",,,,"',,,000000000000000 a0000 I 00 I 000 I 0 I 000000"0(""000'·~0 <lOO0 I 00000000000 , 00 ! 00 1000'~")00 aOOI0000000000 1 0000(0)001')0.~'j.~,~ 0000 I 000000000 , 00000100 1000';00 0001 0000000000 I 00000010000 I 0'~'~ 0000100000001 00000000 010100'000 00001 0eeeeeOOI e l OO l e 00000'J0000 0ee00 t e e eee'J t 00, 0e I 000000,~.)O."0 e000e I e0000 t 0000 t 000000 100''-'0.)0 e00000t0100000000000000 !000'J00 00000 I 0'l00 t 0000000,000001 <).)0').3 000'l00000'l'l'l'l00'l0'lI e0(;0 i 0eO'l')0 000000Q10000000(~01 Q4-)OBc)001 0(.)000 0000000e000)0 -)0e00·1 00')000 I 100.,'(0)0 00<3000000000e t 00 t <3000<3le00'J000 eee00000000 t 0 I 001 00100,'01000(:")') 000000 0000'[ 000000000 '[ ( 100'-)00('('0 00000000)0100101010 t 001 0 t 00000''-'000 0000000000 ! 00e0 1000 t t00000000e 0e000e00000000~1000v:.'~"1t)0t!\0001;'Ot~ 000000000000000 0000ee000000 000 00000000000000 10001000000000ee 0000000e 000000 100 t <3e0000000000 0000000000000<30tOO I 00000e00000 0000000t0001 000000le000000'J00 e 00000000100e0 t 0000000.)0<:'000000 00000000010000 t 00000000000000e 0000000000000e0000000e00000000

c

Fig. 5.- Conical plex iglass phantom fi lled with water (A) and wi th material of similar CT number but with textu re (B). It is not possible to separate contents from plexig lass wall. C, Nearest-neigh­bor edge-detection algorithm applied to texture phantom. Note lack of c lear edge of vessel wall .

AJNR:4, May / June 1983

,' r: _ Z1:Z:t.,.. 1'+/' .. .,.n 1:.t'# ... • ..... , .. .. , .. -""''''''f..,//+, ... a -<lZ •• ......--<.l:-LI""lsn ::z/ ... , . Z1:::tZ, .a:x/'Z- . . .. , ,.. ...... 1'44 . ....... r_ ...... ... r / - -- - :="lr "' -2:- T:::t..SLSIIISSS' I _ n:: ... .... Z T .. ,.. -.!/ Zo-Z//-I: · ff'l+N - ru -N-N-H· ' - r)l z:J'O. · -rnCLt.."'SLSSSIIIII .. ' '''+'1' to/N/ ZT/ ZL-- " tl'Z", .. ,.. ...... N_ .n -N" " +' :- ... - I , r::= I - ')Z- t.L,t.'S'" " "'Z " ",Z"" +/ / 4 ,·ZOz;t ::r++/ .- · 1 ,, " ' - -N'f'i .. .. , N .. 1' -:·I :;:- I : -O - ;:'1~I.L'I"II'

•• / " Co' ''. ''' .. +.-.... Z ;.+Z%N · , r-' .. n · · · r~ ... . .. ... . ::O-- r: :----= , H tS LL llIllS . / / 1:2 /"' ... . ..-+ .... , - , ..... - . • • - N - H- , . ·t4 ·· .. • .. I : r r "':'::::::::: " " 1: - ' 'LLLIIIII

. ...... , .... _ .. , ... ... , .... ZX/ o-N. ) • • t ...... ' • • • • r .. (J'v • • • • ~ ..... ::::. It . ::: - ':::- .~ 'II'I

. , .- .. " ... It , " 'C +- o , ... . ·t / / . N'l · r' ....... • • ",,,j_ . . t h I . N • • " : . ' I II , . .. I I II ",- - 1101 _". 1:" I ... , ( 1: / - + / ......... . + + . + / .... , .. , f U NS - N - " " . r .... - • - " - !'fr "1' f l ' - I I " • :J....l..l..L II • • . ' NB+ ':t/;N+ · .. ' .. ·~/ · · ) · 1 . N · · ~ · 8 - ·N · ·N .. ·, ·N.· t . - .- :C!LOL ILI'I N .. , +!'t'Ul: ,,·r't:+N+ "".N+N'I+- 1 • • - - , 1'1 - _N _ - ~ " ..... 1'1 .... ~ • • , I : I - I ' --')LILI'I'I'I

~ ...... " ":'«"/ ' . 1'1 / " • • I' .. I't'l . N- _. ' ~II'~ . B· - " • I , " . .. N_ 1---=0:-- :": - ' - J\.... I !'N+"fJ:l:r1: ~ ' / " "N" " +N"C" •• · ""' · ·N'I · N·· ' r ·~~ " ' ~f'rl i . ' - I 1-:= - :- '~_L . r!" • • , '( ', / 1'1 .. , / . 1'1 .. .. " ... ( .... I I ··' I ' I N ) -. " - . t , - ' ; 1 r , I. r 11 <::: - 1 :-L -" 'I'IL . .... z n .L/'V "" ' ~h- +. " " f'I'I'!- II.· ) . " , , ' Z ' N; ", r 2 - : -- 11 u ' - 'IZ fl - -L'I 'Ii) , ; .. '- 1/ '1'1+ , ...... / .. !'f'V .. • • · rl ·I'I-N )81'1 ' -M'I/N. , .. " I r , Z::: ·I If."'Z- - I --t)::t>)o.l'l ' . / / ; 'Z:J " " ZS,': " I'I. + 'S' .1'1 . ' B) - ' _ . - . • .... , r t'f ' 2:- ' . :-- t . :-Z-::Ij-,,: , : •• J'l • ,-':J: / ' z ·r . + • • ; ! ..... ... " ..... I't'l · S' 1 ) I •• ·rl · / I " I I ; 1 1-- ' : - - .11-2, " ::: • • .u.. .... ~ l-: ,; . ' U'f'I "' I ' N.~ I ·· · · IBBN''' · · " :I t"" "tl ,r - L-t , ' 1 , 1;//· ··· . , ... :;. ,. ... ... ,_I, t ......... ·I"t'f"i · N+'M ...... " I't'l _ IBB I · • • M'I .. . " r-l : OZ - 1-I2 :r2:: ' N" , ' = '-A.. N; ..... ·' .. + ' '' r .. N • • +N'I::r '' .. I'I / .. ,..I+!')88 1 B'-l ·· ~N • • NI I: I '' . " '22: 11 '" . ,1 , --- : . ... 1'1-1 + / .r ... ... ; " N . rf'!" "' ; +I'fI"N'I . I • • 1'1 . 1'1 ' 8I'I'1 If l " " • • I • r,., I , '" N', ... , ::OOt. :: . N tN" f '" + / .. . J . +f'ot-t+tf'f- +NI~ ' I 1 I I H I ' · N • • • 'D ' • ' ~I' •••• t! • • • • , r - -- r Z • • , N'l r Z J·+ ; T+N - I't"!-N .. . .. . .. !'. 1t 1- 8 11 • • 1'1 - . ) I ' 1f'1 ' .1'1 _ •• · 111 · .. . 111 - 11 ::" ' -n . N . ... ;;' / ' '''N · . ... ... .. 1'1 I N'I". I .. ... . I _n · · rl · I ) 1 , • • I • • r-t'i 1 • '1 t~· ~ :::::::· , ' Z r : .N ... ... ::z:r ...... ; . ... ...... . I I · EI'I+N"I~ J e ..... 1 1 8I'~ . , I D- -NB - · -1'1 - ." . .... , ' " I r: - 1' _ ,

•• _ .. ; . / 1" .. / • .I' +f'I I'N. INB ) " N· · · I' , fI . ·e I ' (j'~I' • • 1_ - ff . I'I ' " 1'1 ' , :: " I I ::' Ir- I .. .. ... .. ... .. / r % . ... N . .... . H'f'I . I'I . o • • •. • ~ • • - B.I ; )' · .II . ,II, •• 11 1 / '1 , 1 1' 1 11 " :

lito" , 1+2 / 1:'«: ' //" ·.N· / ... ··N ' ' 11 11'1 • • ",. , SS · ' / h ' ·n ·· rt I r:: " . • : ' , • .: . N .. ,, 9: J·/ C-- / ... + . /'++ ... ...... // ·· ... B · · I . ·· S· ,) · , ) · I •• /'1/ r: Z. '. 1" ' -/ 1 - : ... , /,, " rz.22 "", ; .. ... // ' N+ . · NBB I 1" . 1'1 I ··· 1 • • • • • • ,~ ... f , : - . ... I , • •• N·: lI"' · *. I ..... ' _: " ... N ... . ..... . , " " ... . N I . SB I * . I BI _. I • • S" •• · . ' r lr •• • z:J'I . I ' : ' " 1/ /1 . .. . ... , ;:: , ," ,J!'I-I . .... ' ..... / .. / N+ ... .. .. . , "'B8B · · e· I · IN "~" I , li t· t t:// I· I " - I I, •• • f +. r: ......... N · . / .. c ::r .. N / +N III · I · · ·. B) · · )·· ·· , I . 1. - 1 1 / . · ' 'II , ". " " , , ~

. , ... . . ............. /) . I'4"f'IZ./I N" ·· · .. · ·r f'l · · SI I 1 •• 1. 1'1 .. .. - ... / .. . , . , • • - .1'1. , ' 1. I I 1./ . rf'! l,,·//,,1: '.N"'N l:<"'I'I / +f'I'I'! -' , · · 8l'I · · B ·· ' 88l'1·N, r " '~ · · N · · ' t l l :::::: 1 ... +N . <::" . N ... N+N ...... + · ... •• ...... N l I 18S " , B8 · N8S I · SB - N· · '1'1 . '1'1 / ' : .. , r t :::: . . ... / .... ' ;::.rX.N. N . ........ • ... / ... + ..... N· · ·N+N!3N I1 ·· S . I S 1 ... • • • • N/ · • • ,'1 1 1 . ' -. / +.+ ... ' / " ...... N ) ..... N . N%/ . .. N+N · • • N · ' 1'1 - 8 9 · ~I·B · ~IN.tl · · . N/; I , tIl': : ., .. , + ... +. " t: ....... ~H ·· ... ···N. / 't.- · ... • • • ,. 1 •• · Bt-f'I ) IN"N'!-'C'f"I"" rr, . Z·': ' .. . . , • • _: .N ... / ... . I't'! . ... .. ............. . , . ge I ' .1'1 ' • B 1~1I'I8B * '~ _ I'lo , .... f": " - I'": T2:::::2 ." o/ ' .. -: ., . ... . .. . ) .+N .. , ...... / ' . ""d 1t;8 1/ 1o ' · 3 11 1.··· ' .. - .:: " : ',' : , . 11 . .. . 1111 ' 1 " .. ... ... . - B ' . 1'1 • • ' __ 1'1_ -~h ' 3 ' '" ' . ~hN. ,. q .. r" , I '_ I , · ;:OZ: ' - "· 1, • • , • • • 1.1 / ' ... - ' N . r""' I N"' ~"'- · 19 18S··· " ," I r~ ... I't~·-· 1'1· 1', - - : . 1.:: . . .. , . - I : r .. .... ~,.f'i ... Nl ·· 8N. ,.,...·- ·· · · S I • • 8D " .. ' I II , ... .... - I' Ir : •• ' , •. , : , .. . _ N ....... . . 'B . · B· ... // • • N·N • • " "l'hBSBNf.ilo I • __ • •• "IIIN • •• • o. ' : ' . • • . , r r " I "'N. , . .. N8 ,)I ." N--Nlh8 1' S "· ' B · 3N!· ..... . . .. "· II ...... .. . . N 1'. : , ., ' ...... " :'f"f. .. ' +i'I .... ... NB BjJ"f'I · N .. · N I 8 I B l -S8· - - . ... N ... ,<!'hN' -: , · N, ' -- ' :: , •.

'~;;;;! ~: .~~ :~~:~::: : ! :~ ; ~:~;;..~: :~~: ~ ;:.::*:;,:: :~.:~: ~ '. : ... ~::: .. :~ ::I:~ ._r ...... --.... .«:12/ + ••• 11) 11 1'f'I ) · Ihl" 3BB- ) ' rl . N_ . ... - 1'1 •• _ . oN · " Z .. ' l· _· . 'q. ·1/1 ... NNZO.' ... ... t::t:N ... N/ •• 9 S •• • +N' • •• S8S • • , . . . ... , ···N·H · · .. _ ··· . N···~"

... .. 'j lo: :r / ".+N. N'+'N •• • le·N· · , .NB ' · " r ~lf'f'I" " • I . ... , .1'1 I •• -+ · f'~/'INI'I · • 'IN _ ' I ... , _: / %, " " .I' I"t'f"i, I · l B J ...... 1'1 . I B I I ··N j N ... !I ) • • • • n. _. I I " . 1'1- ' "~ .I"I!;--

.r! .. . ... rL - "~::. · "'~~~~~ l •• ~~~~:, I I .r l~ ._ • • 1 .1"18~ ~~ ' ::~~~:r::r.lf.'~ · 11_11

A

PEDIATRICS 695

3 1 ~ lJlaI 2 1a131 ~1 ) 1 ~ 1 2 1j I 3121 41714 12 1 i I i 12 1 i 10 iii 'il h HoJ iii 9 101 0 \,1 91 0 9 11.11 11 21 1121 5 17 1?l 7 14 lJI 2 1) I <l t 5 t5 t 7 t?1 7 17 171717 1 717 17 \ 717 4 H 1312 12 121) 15 \ ) 1212 13 14 131 5 171?1 4 12 121 3 1J1I iii iii S 9 9 999987 8 IiI hHI 121J 15 15 151 4 1)1l 121 2 1) HI 5 t 71717I?I?17 1?1?17 1?1?17 5 1)12Ic I 2121) 141 2\1iH2 14\41 2 15 1? 1?151)1 2 1)1)! 1 iii 81 " 8 9 8 8 8 8 8 8 8 9 9I IJ I 2 15 151414 141)1J1J1 4 15HI41 717 17 1?1 7!?17 1717!?17 17 6 1 1 11 1 11 2 121212 111 \1 1 1 lJ ll l)14 1?15Hl ) 1I111 111 9 iii 9 9 8 8 7:- 8 8 B 9 9 9 10 1)17 17 14 141 5 14 141 71 7 t? 15 15 1711t?1717 1717t7t71?171? 7 I l l oJ l ? 1312 10 12 I \1II I IJ I ) I I I)1 41 )1 2 1) 11 iii B 9 B iii 8 7 B 76 77 777109111 4 17 15 151 5 141515171 717 151 4 14 15171717 1717 1717171717 8 111 111 1312 121) 1) 11 1 11 212 12 13 14 12 12 11107 7 B iii 8 ~ 8 ., 7 7 7 7 7 a 8 gI0111 ) 14 141 ) 131 ) 1415 15 15 14 13121313 15 t 717I7 t ?1717 17 17 17 9 111 11 \1 12 12 121) 1) 12 11 111 2 12 13141 ) 1 1111 0 7 6 8 9 8 8 8 8 7 7 7 8 ? ? 7 1 ~ J(" 1 12 12 111 2 111 2 141 3 12 12 1)12 12 12 141 5 15 15 17 171717 17 17 17

loJ 1 11111 I \lh'1 1 12 1) 12 111 0 1111121 4 11111110 9 8 9 9 8 7 8 ';' 8 9 9 7 7 68 9h)\e I0 10 1011111 212 12 12 111 2 12 13 1J 1) 1) 1) 1213 IS I 7 17 171 7 17 II 1 1 1 .~ l l 989 12 12 12 111 0 10 111 2 1111 9 999910 9 7 7 6 7 89 888779999 9 81 01 1 1 1 1 1101 0 12lJ I 213 14 1313 1) 12I <1 I S 1717t? 17 17 I Z h"ll ll 9 8 9 121) 12 1 11 0 9 10 111 2 10 8 8 'iI 8 'iI10 9 7 6 6 6 7 8 8 7 8 8 7? 9 8 113 9 9 9 111 01 1 91 11 2 12 1) 13 I <1 I)1 2 I <1 IS I 7 1711111?17 17 I ) 1\11,'1\11 \1 8 1,) 1)13 12 121 11 0 9 1 1 1 1 9 7 iii 9 8 8 998 6 I;> 7 8" 7 9 8 7 9 10 1;) 1 1 10 9 9 9 111 0 10 10 12 13 I H <lI S I3I <1 17 17 171717 1' 1717 17 I~ ; J 11 11111 2 131J 1 2 1 2 1 11 ~ 9 10 1151 8 5110 9 8 9 8 7 7 7 8 8 7 ?? 7 8 8 101 1 1iH I I I I I 9 10 9 12 1 1 10 12 1<1I S IS IS I<l I <l 17 17 171? i? i?17 1? 17 IS 9 Q! II Z I 3 I J I ) 1l1l1 2 111.a 9 10 10 9 51 10111 0 51 9 8 8 8 " 8 8 7 7 7 7 7 10 1010 9 111 2 1110 9 11 IZIZll l ) I<l IS I<l 15 1 <1 I S 1S 1SIS1?17 17 17i?i? 16 9 9 1( 1 ) 1J 1 3 1 2 1 ~ 1111 1 110 9 9101 0 10 1 1111 13 9 8 7 8 7 7 6 S e 7 , 6 7 9 12 10 9 11t11 2 101 1 1 11313I<11<1 I <1I -4 I -41 4 15 I S I<1 14 14 15i?I ? l'I I 7i? I "' 91"IIt) I·H 3 12 I \1 10 IZ12 1010 9 9 101 0 11 1 1 988 7 88 6 S <I <1 6 77 ?l0 11 10 9 1 112 12. I I I3 I <1I<1I S IS IS 1<1 I <1 I <1I<1 i?1 7 IS IJI <l 17 17i? 17 17 17 I S 1~III Z I 3 1J1)13 1 0 11 l c I 2 111 1 999101111 8 7?? 7 , 7 6 S S 6 8 8 8 9 11 10 10 11 12.111 2 13I<1 I ) I ) IS I 7 141 4 1] 1<l I7I7I S 15 17 17 I ' ,7 17 17 17 19 L II..J.!))I3HI3I~ 111 2 1 ZIZ 1 I Ll 9 91 11211 10 8 7 6 6 5 6 7':; 6 7 8 8 8 9 111 ( 11 12 13 1) 1) 15I<11 3 13 IS I 7 15 1) 121 4 I S I5 15 15 17 15 I<1I ) IS I 7 17 z,' 1 .1 I ~ UI3IJHlah" I OI2 111 1 1 c I 0 9 101111 10 9 8 7 6 5 5 5 S S 6 ? 8 8 8 9 1111 1 11 21J1) 13 I <1 i? 17 I<1I <1 1S I <l 12 1213 I <1 13 I <1 131] 12 12 111 ) I ' 17 ~ I 1l1 .~HH I ~ \J 1 2 t." 111 0 9 9 9 9 8 9 1010 9 8 8 7 7 6 6 5 <I <1 6851 8 1 IdI I10 11t11 3 I J I <lI S1 ? IS I4 1212 1) 12 12 13 I J I 21 11\ 12 111111 13i?1 7 22 h ' \I \J I~I)1212 11111 \) 8 7 8 8 8 81 0 10 51 9 51 8 8 7 7 S ) <1 67 88 ? 81 0 101 ( 11 13 1)15I <1 IS))12 111 2 131 ) 1) I <1I ) 12 11101011 11 13 1) 1717 2J l. q . 'III ] 1111 12 I Z1 1 1 1 9 778 951 1011 109910 iii 8 8 6 S S S 7 7 7 7 8 8 8 51 11 I1 I)13 1 3 1 2 11 11I cI21] I<l tll ] I ~11 11 9 10 12 13 1 <1 17 17 17 2~ '1 ",,,1 111 2 12 12 1\\ 111110 777 9910 10109 iii 9 8 8 7 6 5 6 6 6 8 7 7 7 8 7 8 8 9 11 11111010 12 12 12 12 1211 1 1 1011 9 11 111 )1 <l I Si?1?1? :~ • ' ·1.1\.' 1 .~ 1 ) 11 1 11 \l 1110 8 6 7 8 8 8 9 51 51 8 9 9 8 7 S 5 S 6 5 8 8 ? 8 8 ' 7 7 7 51 51 8 9 9 911 11 113 9 9 9 1e 10 1 1 1 11 3 14 141 517171S

.o... :j <1 1 11 11 3 12 1\1111110 8 7 8 8 8 ?? 51 9 9 g 8 9 8 6 S S 5 6 7 9 7 7 7 6 6 6 Ii 7 8 8 8 8 8 9 10 9 9 9 10111 2 13 1<1 15 15 1<1 I<1IH<lI <I "' • n I11 12 1': 1111111 ,) 8 7 899 8 7 6 7 88 g 51 9 8 6 S ' 766 8 7665 6 ? 6 6 7 7 8 7 8 8 9 8 8 9111 2 14\<11 5 15 15 1 <1 141 ) 12 12

.~ = ' ::::: : : ~ : ~ :~ : 1 : ~ 1 ~ ~ :; ~ ~ g; ~ ~ ~ ~: : ~ ~ ~ ~; : ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ : ~ : ~ 1 : 1 ~:~:~: ~ : ~ : ~ : :: ~: : :j:;: ~ : ~ : ~ ·.l hl\''I\H I II 13 12 1311 1010 9 9 10 9 8 7 8 7 7 7 7 6 7 7'77?., 6 5 6 5 5 5 6 7 7 7 7 7 9I e I0 111 2 Il' 3 11111 2 12 1) 1) 1)131J12 1)1 ]

I ::';:.~:~:3 : ~ :~ : ~: ~ : ~ l g ::l; I ; : : : : I: g:~I: ~ ~ ~ ~ ~ ; ~ ~ ; ~ ~ ~ ~ ~ ~ ~ ~ ~ ; ~I: ~ ~ I~ I ~: ~ :~: ~ :I : ~ : ~ : ~ : ~ : ~ : ::~:~ : ~ : ~ : ~ : ~ : ~ 1 " ll l l l r~ 1 3 1)14 I Si3 1 2 1 0 1 0 8 910111111101 0 8 7 6 5 <I S 5 6 7 6 S 6 6 6 6 6 6 6 6 7 9 9 10 1 11 2 I <1 I ) 12 1 11 2 1) 12 10 1d I0 111 2 12 12 11 1213

~ l " lL III I I ) I<lI~I)1 3 111 0 9 81 0 11111 2 1 1111 0 8 ? 7 55 S S 6 6 6 <I 6 6 6 6 S 6 6 5 6 9 9 10101 1 i312 10 1C1l cI 3 12 1 1 10 1111I3t<1I 2 101 1I c 1 . ' 1! 11 11 Ic l ·I\ ~l c I2 1\1 0 8 9101111 121 2 119 99 8 7? 6 6 6 6 6 S <I 5 6 5 S 6 S S 6 910 910111 1 12 111t1 0 1Z12 1011 1( 1 11 2 1) 12 10111 2

, 1.'1 1 1" 1 "h1 1 ~12 1 1 111 1 9 8 7 911 10 12 12 9 910 9 8 7 6 ? 6 6 7 7 6 <I 5 6 6 6 6 6 6 7 10 1 11 0 101 1 1212 11111111111 0 VUHtlH!1 llt21 2. 111 .1 1 1 • l " ll } b QII 12 111 0 10 10 8 7 9 91011119999 77 7 66 7776 55556? 6 6 7 10101 0 10 111 0 1 1 11 1 11 0 10 9 8 9 9 1( 1 1 12 12 12 13 1)

'10.1 10 :- 9 121)11 111l1 '" 8 7 8 9 10 11 10 10 9 9 8 8 8 " 6 S 6 6 6 7 7 S <I <I 6? S 4 6? 8 8 9 10 9 9 10 9 9 B 9991( 11 1212 ) ))3 1<1 1<1 I 'J J cl\l'ilH l r. h ' 10 9 9 8 8 10 10 1010 9 11 98 99 86 5 6 S 5 766 <I 4 6 7 S ) 5 S 6 6 89788889991( 12 13131213 121) 1<1 , "11 ' 'I' ' l\l l l ~ 1l1 1 I('l 9 9 8 8 9 10 9 8 W 1 11 0 9 9 9 8 7 6 7 7 8 6 6 S 6 5 6 S <I S 5 5 4 S 8 8 7 7 8 8 8 8 1010 11 1213131213 13 12 1<1

0\ 10': 1.'11 1. 1) 12 1':: 1\) 10 9:- 8 9 9 8 7 9 10111 0 10 8 8 S:' to 7 7 6 S 6 6 5 5 3 ~ 6? 6 5 6 8 B 7 7 8 8 8 9 11111 ?, 12131) ) ) 1) 1313 1<1 • I \.1 11 I J l l h1 l o) 9 9 8 7 a 8 7 7 8 1011}0 9 9 9 8 6 S 50 S S <I 5 7 6 -4 <I S 6 7 5 S 6 7 7 8 8 7 9 8 9 111 2 1 ~121313 1 3 1<1I<1I<1I<1

, I : !' 11 1. 1·1I4111Ih"l10h) 9 8 8 18 8 8 10 111 01 01 01\1 9 D ~ ~ 5 S 5 6 6 S S S 6 8 to <I <I 5 7 7 8 9 9 9 10 10 12 13 1) 1<11<11<11<1 1<11 5 1 <1 1<1 I·, I I l.' I. · I·I\~ I Z I \l1 1 1111 99 :> 7 8 8 9 910 10\111 10 1 11 ,) 7 S S 5 6 6 7 6 ~ 5 6 777 <I <1 67810 1(101 2 10 11 1) 1)13 1<11 5 1<1) <1 \<11 5 1<11 2 I' 1 .1 1 11 ' 1'11 2 1) 12 12 12 111 09 8 ':; 6 7 8 7 8 9 10 10 10 1\) 10 7 6 5 S <I 6 6 6 S 6 7 7 7 6 S 6 a 8 9 11311 12 12 1 1 1113121315 15151) 1<11<1 13 12 I, I.··' ~ bl~ I Z I IL 2 1 2 11 987 ?? 7 7 7 to 8 }0 9 9 10 9? S S <I 5 5 6 6 S 5 5 5 7 6 6 71010 9 101(11 1 111 111)12 IZ I <l IS I <1 13121J I<lI ) 1- ·1 .· ll l' J II1 2 12 12 12 1 11 ,) 9 8 8 B 7 6 6 6 5 8 10 9 999 7 6655 566 5 ~ <I S' 7 7 9 11111010 10 9 10 1( 11 12 1 11 11 2 12 12 11 12 1) 12 1 <1 4':; 1. · I III I. · l c I J I ) 121 11\1 9 9 8 7 8 7 <I S <I S 7 9 9 9 108 8? 6 6 5 6 8 8 6 <I <I <1 68 8 10 1 1 1 11 0 10 10 10 101 0 10 999 B 9 10 12 12 12 131) .\·l 1 11111 1r. 13 131)\312 10 9 8 77 88 S) <I <I 7 7 7 8 8 98.:; 5 5 S 6 8 6 6 S ~ S 6 8 8 9 10 9 8 9 111 2 12 11 10 9 10 9 8 8 10 111 2 12 1312

.': '11 11 1': 1) 1) 1<114 12 11 10 9 " 8 9 7 S <I S ? 7'? 8 9 9 9 5 5 5 '" 5 13? 5 6 7 6 7 8 9 9 8 8 8 9 1 1 13 1 <1 12 10 101 111 1(10 1 1 1) 1) 1) 12 11 : 1 1 11 " I ,'1 I IZ I J I 4 13 IZ I IIILI 9' 8 9 8 7 6 5 7:' 8 8 8 9 B 5 6 6 D 6 6 S 6 6 7 8 8 8 8 8 7 B 8 9 101)1) 111l1 0 1 11 2 1 1101 1 1Z 121210 9 ">.- • '; ' 101] 1 1 ~ IS I ) 1I1l 1 2 1 2 1,) 9 9 10 9 7 6 5 S 6 7 8 7 8 7 7 6 ? S S <I S 6 6 6 8 9 8 8 8 9 8 9 9 10 111 1 10 10 10 111 2 10 l u lll l 11 988 .1 ... : 9 111" l c \3 1 1111 3 13 111 0 10 10 9 7 5 <4 56? 8 7 8 8 ;> 7 S S 3 <I <1 55 679889899889999 9 10 1 1 1Z 1110 10 10 9 8 8 •• \ • ~ ' j 'J l ': IS IS I 3 1 111 IZ IZ II 8 9 1;) 9 6 S 5 5 5 7 8 8 8 8 6 7 S ) 5 e 6 6 S 6 7 8 810 8 8 8 8 9 7 8 7 6 8 910 11 111 0 10 9 9 9 9 9

I ~ ' '1'~1 lISP I ) 1 1 1(1 1111 9 8 8 9 7 6 S S 5 6 6 8 7 ") ? .:; S S -4 6 8 9 8 to 7 7 7 8 8 8 6 7 9 9 8 7 S 6 ., 8 9 10 9 9 9 8 9 10 10 10 • I 'I <1 "' 1 Jl c IS\J IZ I I 1,)1,) 9 8 7 8 7 S 5 S 5 S 7 9 7 7 ., 7 5 4 S 6 7 9 8 to 7 9 8 7 7 7 6 to 7 B 7 7 6 5 6 8 9 9 8 8 8 8 9 9 9 9 .' • '1 h~ 1 41 :- 1 :-IJ I 2 10 9 7987 8 7 5 'So S 6 6 6 7 7 :- 8 8 6 6 6 5 D 6 ? 6 .:; 9 8 a 8 8 6 5 5 6 7 7 7 7 8 8 9 8 7 8 8 9 1010 910 '., 1·>"1 '''' 1 1 1:- 1:' 1 ) 12 116 898777 SS6? 66 :- 66 6 8:'e to 5 ~ 5 06:- 7 7 7 8 8 6 3 3 6 8 109 B B 98 7 8 799 1 11 1 12 11

J 1 , h' I J I "" I '..I1 I )IJ II101\l 8 :-:- 6 S 6 ? '" 6 7 7 5 -4 S S';'; 6? S S 6 6 :- 7 7 7 8 7 S <I S 7 8 910 9 8 9 9 7 8 9 8 9 111 11313

B

1 dll' .1 ' 1C, 1 112 1J1J1 110 9 :- 7 6 S S 6 5 6 6 6 6 S 4 ~ S';:- 6 . 6 S S S S D 5 6 6 D 6'779109 10 1110 9910989 12 13 12 ".'1 "1 11·1\ ' 15 1 111111 11 1 <;I 8"' '' 6 S 6 S S S 6 6 a 6 S 4 S"; '" ~ 50 S S S S a 7 S S 6:- 8 :' 9 8 101 Z 111 11 211 10 111 2 1110101 1111 2 '1 1 1 1 1 1\ 1 1 " l o;; lll0h~h1 1I1 G S 7 7 6 6 6 :- 7 6 6 S 6 6 5!> '" - ~ '" 5';'; a 6 7 7 :' 6 6 Ii B 8 8101 11 2 1010 12 12 10 1 11 11 11 1 11l1l Z12

1. '11 1 110 '1 1\ ' 1411 \ ll .~ 1 <! 1 2 1 211 9 ? '; 5 D 8 8:- S . 6 7 6 -I4 !Eo :-:- U ,;:- :- U '..o!> ', ? 7 7 a., 8 9 9 1011 99 10 1099 9 10 10 111 2 1Z 12 I:' I'- I I>' ,, ·.1;1 ' 1t'II ,nl 1 r. "'- I P I ' lt 5

Fig. 6.-A, Raw data of zoomed view of right upper quadrant of f ig. 3A. number). B, Regional c lustering algori thm applied to this data. Gray and white matter now are c learly demarcated. Note wide variation of CT numbers (each symbol represents different CT

to distinguish gray and wh ite matter. Other methods (e.g. , th e global maximum-likelihood method) are difficult to use because of the small size of the spatia l regions that are c learl y white or gray matter.

The development in this study of th e regional-clustering algorithm (based on the measured textural properties of gray and wh ite matter) is a first step in the use of higher-order statisti cs for scene segmentation . Its potential has already been validated by a specially constructed textural ph antom . Current research in image scal ing , image orientation , and alignment will allow direct correlation with pathologic sections .

Finally , neonatal CT scans were used in this study as a standard for the evaluation of developing brain , wh ile adult CT scans were used as the standard for evaluat ing abnormalities of myelinated mature brain . The results of our analysis showed no significant differences in the gray / wh ite matter differentiation , even though the degree of myelination and water content of the white matter differed greatl y between the two groups.

REFERENCES 1. Brooks RA , Di Chiro G, Keller MR . Explanation of cerebral

white-gray contrast in computed tomography. J Comput Assis t Tomogr 1980;4 : 489-492

2. Dobbing J, Sands J. Quantitati ve g rowth and development of human brain . Arch Dis Child 1979;48: 757 - 769

3. Cahill PT, Kn owles RJR , Kneeland B, Lee BCP. Segmentation of white / gray matter by a tex ture based c lustering algorithm . In: Proceedings of the 6th international conference on pa ttern recognition. Silver Spring , MD: IEEE Computer Society, 1982 : 633- 636

4 . Hanson KM . Detectabi li ty in computer tomographic images. Med Phys 1979;6 : 441 -445

5 . Cahi ll PT , Knowles RJR, Tsen 0 , Pouapinya R. Evaluation of edge detection algorithms for nuc lear medic ine via roc and shape analysis with adaptive thresholding. Proc Inst Electrica l Electronic Eng 1981 ;28: 64-68