n Interactve Bone-Scntgraph Dagnoss B a Characterstc-Pont-Based Fuzz Inference Sste V SC 893E04007 89 8 90 7 3
n Interactve Bone-Scntgraph Dagnoss B a Characterstc-Pont-Based Fuzz Inference Sste 89 E-al: qtn@al.chna.edu.t sste nputs. The prototpe sste s pleented n JV. The eperents th doctors shoed the effectveness of the proposed nteractve dagnoss n ( adng doctors b autoatcall arng nute abnoral locatons. ( : Interactve dagnoss, bone scntgraph, 0 characterstc pont, fuzz nference sste (dvde-and-conquer (CP [-3] CP (bac-propagaton [4-6] JV [7] [8-9] ( [0-] Bone scntgraph s a senstve and (CP nonnvasve ethod to dagnose bone tuors. lthough artfcal neural netors ( have been appled to autoatc dagnoss of soe edcal (bac-propagaton (dvde-and-conquer agng, there s no autoatc sste to ad doctors n 0 the checng of bone-scntgraph agng. Ths [0-] JV paper proposes an nteractve bone-scntgraph dagnoss b a characterstc-pont-based fuzz nference sste (. In our prevous research, has been shon that an cople sstes CP ( can be suffcentl descrbed b th no ore than 0 fuzz rules. The frugalt of usng fuzz rules results fro the dvde-and-conquer ablt of algorth n fndng CPs. lso, due to the CPs, hch are the locatons of nput fuzz sets, the bac-propagaton learnng of the other paraeters s. ore effcent snce the centers of nput fuzz sets do not have to be learned. fter dscussons th Tc-99 doctors, setr and brghtness are chosen as to
56 04 (pel 6 t c r T S ( t, t.. C( ( C(, (centrod C( ( s( T d ( = s(, S( + + = {( s( T, s( s( a, b, a =,, +, b =,, + }. Step : Centrod ethod ( Qucsort d( 0 55 L C ( =, = 0,..., 03 d( L = { t t. s t. s, < }, 0 55 Step : Fne-tune t. s t t 0, t,, t t C( n C( + n a T T d( + a, b d( a, b, b= f,, + f 0 T T n,, f T f 5 s ( brg( Segentaton lgorth Let I ( = 0, = 0,,..., S ( t c = S ( tc at ( = S ( Choose t, so that I ( > 0, = 0,...,. ( If no such t ests, then stop. tc s( brg( Let t c = t and S chec = S tep = { t c } 0 3 If S chec = Φ, then let S ( t c = S ( d( d(c( tep ( S and go to step. s( = S Reove the frst eleent t of S chec and add t nto S tep. 4 The locaton of t s (,. Chec all the eght eleents at (,,, ( +, +, around (,. If an eleent t has the propert that t. s tc. s r, t. s T and I ( = 0, then add t nto the set S chec as the last eleent of the set and let I ( =. 5 Go to step 3. brg( S 4 r 3 (9 log 0 C( = arg + S ( S d( n C( brg( = 0 log0(9+ S S ( S ( t c S S log 0 (9 S S s( +
, L,, R,, CP, CP., L,, R,, (bac-propagaton ( ( [0][]. E = = (. α ( t CP ( ( (, = ( [( CP CP t = η = (, = CP ( =, CP ] + α ( t If s, and and n s If s, and and n s, n, then s B, n, then s B n t ( B n E + R,, ( t = η [ (,, L-R = ( µ, (, ep(, f L,, ( = (, ep(, f R,, (, L,, R,,, <,, µ B 0, f t < 0 [0][] µ ( = ep( B ( ( = µ (... (, µ, n n = = = 4 ( t = η = [( ( ( = ] + ( E + t = η [ ( ( = ( L,, (, ( (, 3 L,,, 3 R,, ] + α ] + α 0 <η < 0 < + ( t step, f t 0 + ( t = L,, R,, ( t ( t <α.
fuzz logc approach, IEEE Transactons on edcal Iagng, vol. 7, no. 4, pp. 489-497, 998. CP CP [4] H. L Y. Wang, K. J. R. Lu, S. C. B. Lo, and. T. Freedan, Coputerzed radographc ass detecton part II: decson support b featured database vsualzaton and odular neural netors, IEEE Transactons on edcal Iagng, vol. 0, no. 4, pp. 30-33, 00. 3. [5] S. Yu and L. Guan, CD sste for the autoatc detecton of clustered JV crocalcfcatons n dgtzed aogra fls, [] IEEE Transactons on edcal Iagng, vol. 9, no., pp. 5-6, 000. gaa caera (Seens, [6]. G. Penedo,. J. Carrera,. osquera, and D. E-ca, US Technetu-99-DP Cabello, Coputer-aded dagnoss: a (ethlene dphosphonate. neural-netor-based approach to lung nodule. detecton, IEEE Transactons on edcal Iagng, vol. 7, no. 6, pp. 87-880, 998. [7] K. Van Laere, K. Caser, D. Uttendaele, W. ondelaers, C. De Sadeleer,. Sons, and R. Derc, Technetu-99-DP scntgraph and long-ter follo-up of treated prar algnant bone tuors, J. ucl. ed., vol. 39, no. 9, pp. CP 563-569, 998. CP (0, 0 [8] F. Pons, L. lvares, P. Pers,. Guanabens, S. CP Vdal-Scart,. onegal, J. Pava,.. Ballesta, J. CP unosgez, and R. Herranz, Quanttatve evaluaton of bone scntgraph n the assessent of Paget s dsease actvt, ucl. ed. Coun., vol. 0, no. 6, pp. 55-58, 999. [9] S. H. K, S. K. Chung, Y. W. Bah, Y. H. Par, S. Y. Lee, and H. S. Sohn, Whole-bod and pnhole bone scntgraphc anfestatons of Reter s sndroe: dstrbuton patterns and earl and JV characterstc sgns, Eur, J. ucl. ed., vol, 6, [] no., pp. 63-70, 999. ( [0] T. K. Yn, Fuzz o delng and control: a characterstc-pont approach, Ph.D. thess, Boedcal Engneerng pplcatons, Purdue Unv., Indana, U. S.., 996. Bass, and Councaton (EI CP [] T. K. Yn and C. S. G. Lee, CP characterstc-pont-based fuzz nference sste, n 996 san Fuzz Sst. Sp., 996, pp. 533-538. [] http://s.chna.edu.t/qtn/research. [] G. Zahlann, B. Kochner, I. Ug D. Schuhann, B. Lesenfeld,. Wegner,. Oberaer, and. ertz, Hbrd fuzz age processng for stuaton assessent, IEEE Eng. In edcne and Bolog, pp. 76-83, Januar/Februar, 000. [] J. R. ansfeld,. G. Soa, J. R. Paette, B. bdulrauf,. F. Stranc, and H. H. antsch, Tssue vablt b ultspectral near nfrared agng: a fuzz c-eans clusterng analss, IEEE Transactons on edcal Iagng, vol. 7, no. 6, pp. 0-08, 998. [3] W. Par, E.. Hoffan, and. Sona, Segentaton of ntrathoracc ara trees: a 5
Scntgraph Iage Local-au Based Segentaton Step : Iage Processng Start Read Iage Dagnoss Setr and Brghtness dd oral Ponts dd bnoral Ponts Learnng Vsual ddng Tranng Step : Presentaton of Saples Interactve Fuzz Rules and Dagnoss Doctor Tranng Saples lgorth Constructon of a Prototpe of Bac-Propagaton Learnng of c,v (,b ( ( c,v (,b ( ( (3,b (3 (3 Partton CP nodes Output value Suaton node node 6