US005999639A Ulllted States Patent [19] [11] Patent Number: Rogers et al. [45] [54] Date of Patent: Dec. 7, 1999 METHOD AND SYSTEM FOR AUTOMATED DETECTION OF CLUSTEREI) 5,627,907 5,673,332 5/1997 Gur et al. .............................. .. 382/132 9/1997 Nishikawa et al. . 382/128 MICROCALCIFICATIONS FROM DIGITAL 5,857,030 1/1999 Gaborski et al. ..................... .. 382/132 MAMMOGRAMS [75] 5,999,639 FOREIGN PATENT DOCUMENTS Inventors: Steven K. Rogers, Beavercreek; Philip WO91/07135 5/1991 WIPO .................................. .. 382/128 Amburn, Dayton; Telford S. Berkey, London; Randy P. Broussard, Huber OTHER PUBLICATIONS Heights; Martin P-_DeSimi0, Fairborn; Je?'rey Ho?'melstel', Beavercreek> an of Ohm; Edward M‘ 0ch0a> San Carman et al., “Detecting Calci?cations and Calci?cation Clusters in Digitized Mammograms,” Digital Mammogra phy ’96 (edited by Doi et al.), Elsevier Science B.V., 1996, gntomo, TiX.,ITlhorlrE1asRP. Rathbuli, pp 253_255, 1996_ Huber Heights, both of Ohio Primary Examiner—AndreW W. Johns Attorney, Agent, or Firm—Biebel & French eavercree ; 0 n . osens enge , [73] Assignee: Qualia Computing, Inc., Beavercreek, [57] ABSTRACT Ohio [21] Appl- NO? 09/ 141,802 [22] Flled' - A method and system for detecting and displaying clustered microcalci?cations in a digital mammogram, Wherein a _ single digital mammogram is ?rst automatically cropped to Aug‘ 28’ 1998 a breast area sub-image Which is then processed by means of Related US Application Data an optimized Difference of ('iaussians ?lter to ‘enhance the [60] Provisional application No. 60/057,801, Aug. 28, 1997, appearance of poFennal mlcro(_:alcl?canons 1n the Sub provisional application No_ 60/066,996’ NOV_ 28’ 1997’ and provisional application No. 60/076,760, Mar. 3, 1998. [52] [51] image. The potential microcalci?cations are thresholded, clusters are detected, features are computed for the detected US. Int. Cl? Cl. ........................ ..................................................... .. 382/132; 382/156; .. G06K 382/260; 9/00 . 382070 _ . 1 1th y s oping oca tions in the Original 382/155’ 156’ 296’ 225’ 227’ 256’ 257’ h H res 0 b't mg, 1 may a so bg e mammogram of the Suspicious detected clustered microcalci?cations are indicated. Param 260’ 270’ 272’ 128/922’ 706/13’ 924 eters for use in the detection and thresholding portions of the , system are computer-optimized by means of a genetic algo References Clted Us PATENT DOCUMENTS , , 5,463,548 y performed by global and dual-local thresholding. The loca Fleld Of Search ................................... .. [56] f p M b Y1 is pre era rithm. The results of the system are optimally combined With a radiologist’s observation of the original mammogram by combining the observations With the results, after the radi ly golfcflbeig / 10/1995 ta or et a . .................. .. / . Asada et al. ....... .. .. 364/413.02 """"""""""" " ologist has ?rst accepted or rejected individual detections 5,537,485 7/1996 Nishikawa et al. 5,625,717 4/1997 Hashimoto et al. .................. .. 382/260 reported by the System‘ 382/130 129 Claims, 28 Drawing Sheets Cropped 302 Image 340 Thresholding 350 360 Remove Detections Outside Breast ‘ Breast Mask Group into 370 1 Calculate 380 It Features 298 U.S. Patent Dec. 7,1999 Sheet 1 0f28 Get Digital 5,999,639 / 100 Mammogram l / 200 Autocrop Analysis / Region l 300 / Detect Clustered \\\ Miorocaloifications <—— Optimize Parameters l Classify Microcaloi?cations ’/ 400 l W Process Results ’/ For Display Display Deteotions Fig. 1 / 500 600 // 700 U.S. Patent Dec. 7,1999 190 Sheet 2 0f28 200 Digital 298 Autocropping —> Mammogram 5,999,639 Algorithm Breast Mask Fig 2 190 Label 29§\ 200 / Autocropping Algorithm Breast 296 Background Fig. 3 U.S. Patent Fig. 4 Dec. 7,1999 5,999,639 Sheet 3 0f 28 Start \ 202 Sub sample mammo / l 204 Create White Border / t 206 Threshold Mammo / i 208 Invert mask // 210 Dilate Mask ,/ l 212 Crop to Largest // Object To Fig. 5 U.S. Patent Dec. 7,1999 Sheet 4 0f28 5,999,639 Fig. 5 From Fig. 4 Auto Histogram Enhance 214 / 216 Select Brighter Side i 218 Initialize Search / MaskSize = 0 B = 0 To Fig. 6 U.S. Patent Dec. 7,1999 Fig 6 Sheet 5 0f28 5,999,639 g) From Fig. 5 _ 220 No Mask Too Small? Yes 4 l 222 Select Side to Search / " 224 Find Seed Pixel in Select Side / / l Region Grow // l Compute Mask Size// No Mask Too Small? 232 At Minimum Threshold? V Decrease (75 Threshold _ v @ From To To Fig. 7 Fig. 7 Fig. 7 U.S. Patent Dec. 7, 1999 From From From Fig 6 Fig. 6 Fig. 6 Sheet 6 of 28 5,999,639 @ 236 Yes No Searched Both Sides? - Threshold Entire Reinitiaiize Search image for other Side 238 240 / ‘ / i 242 Close Holes in Mask / / 244 V / Duplicate Mask ‘ 246 Erode Mask // i 248 Dilate Mask - Fig. 7 / To Fig. 8 U.S. Patent Dec. 7,1999 Sheet 7 0f28 5,999,639 From Fig- 8 Fig. 7 Compute Sizes of Old 250 and New Masks 252 Is New Mask Size < 0.5 Old Mask Size? Use Duplicate lmage Crop Mammogram to 256 Size of Largest Objeot/ V 258 Crop Adjustments // Auto Histogram Enhance / 260 Cropped Mammo / 262 Apply Loose Region Grow // To Fig. 9 U.S. Patent Dec. 7,1999 Sheet 8 0f28 5,999,639 From Fig. 9 Fig- 8 _ 264 Close Holes In Mask / i / 266 Erode Mask / ‘l _ 268 Dllate Mask / i 270 Crop Mammo to Size of Largest Object // V 272 Crop Adjustments // l Auto Histogram Enhance 274 Cropped Mammo // i 276 Apply Tight Region Grow To Fig. 10 / U.S. Patent Dec. 7,1999 Sheet 9 0f28 . 5,999,639 From Flg- 10 Fig. 9 278 Find Largest Object V Crop Mammo to Size of Largest Object / 280 ‘ 282 Cro Ad'ustments P J // 284 Invert Mask / 286 Find Largest Object// V _ 288 Invert Largest Object 7 Close Holes in Final Mask End 290 // U.S. Patent Dec. 7, 1999 Sheet 10 0f 28 Cropped lmage 5,999,639 302 310 Noise Filter l 320 Apply DOG’ Filter i 340 / Thresholding i 350 Shrink to ’/ Single Pixel 360 l Breast Mask Remove Detections Outside Breast i 370 Group into Clusters ’ i Calculate 380 Features ' Fig. 11 298 U.S. Patent Dec. 7, 1999 Sheet 11 0f 28 5,999,639 P(X.y-1) p(><-1,y) p(><,y) p(><,y+1) Fig. 12 p(><+1,y) U.S. Patent Dec. 7, 1999 Sheet 12 0f28 5,999,639 x10-3 ovBiEm .Y ,.. . muzEam -15 -10 20 xindex Fig. 14 U.S. Patent DoG Filtered Image Dec. 7, 1999 Sheet 13 0f28 p(x‘y) ‘ Compute Image Histogram > GlobalThresholdValue Globally thresholded image, g(x,y): 1, p(x,y) 2 GlobalThresholdValue O, p(x,y) < GlobalThresholdValue Fig. 15 5,999,639 Compute Global Threshold Value U.S. Patent DoG Filtered Dec. 7, 1999 Sheet 14 0f28 pow) 5,999,639 Select Group of > Pixels in N x N lmage Neighborhood of p(X,y) l Compute upper and lower thresholds: tlo : FLNN(XIY)+ klo 6NN(X’y) Locally thresholded image, l(x,y): 1, tie < p(X1y) < thi 0, otherwise Fig 16 U.S. Patent Dec. 7, 1999 Sheet 15 0f28 Fig. 17 5,999,639 U.S. Patent Dec. 7, 1999 Sheet 16 0f28 Center (N X N) Digital Mammogram 5,999,639 Compute mean and window Over _> piXe| pow) standard deviation of pixels under the window MW) and 609v) V gig Compute local threshold: T(X.y) = A + B- ulky) + 0- 6(X,y) d(X,y) ‘ Local Threshold d(Xry) > T(X,y) v Locally thresholded image, l(x,y): Fig. 18 U.S. Patent Dec. 7, 1999 (x,y) coordinates 5,999,639 Sheet 17 0f 28 Compute Distance Matrix of centroid locations D =[ d(i,i)] V Identify and count points Eliminate points with fewer within distance than uCsmin neighbors within dNN dNN of each other V Merge clusters Lists of points sharing one or associated with more common each remaining points cluster Fig. 19 U.S. Patent Dec. 7, 1999 Sheet 18 0f28 5,999,639 10 $296:1 94037625 .: 0 1O Column Index Fig. 20 U.S. Patent Dec. 7, 1999 Sheet 19 0f 28 5,999,639 Points Associated With Clusters + Compute lnterpoint Compute Covariance Distance Matrix D Matrix of Points in Cluster Compute Eigenvalues of Covariance Matrix 9\’1 ‘ Number Points in Cluster )LZ Rectangular Area _, &2 x1 _> Standard ‘ Deviation(D) V —-w Mean(D) - Median(D) + ~M Max(D) - Min(D) > Number Points in Cluster Max(D) Fig. 21