multi-modal volume registration by maximization of mutual information（登记互信息最大化的综合卷）.pdfVIP

MultiMo dal Volume Registration by Maximization of Mutual Information William M Wells I I I Paul Viola Ron Kikinis swaimitedu violasalkedu kikinisbwhharvarded u Keywords MultiMo dal Image Registration Mutual squared dierences is directly prop ortional to the likeli Information Information Theory ho o d that the two images are correctly registered Unfor Categories Registration for Medical Applications tunately squared dierence is not an eective measure Abstract for the registration of dierent mo dalities Even when p erfectly registered MR and CT images taken from the A new informationtheoretic approach is presented for same individual are quite dierent In fact MR and CT nding the registration of volumetric medical images of are useful in conjunction precisely b ecause they are dif diering mo dalities Registration is achieved by adjust ferent ment of the relative p ose until the mutual information This is not to say the MR and CT images are com b etween images is maximized In our derivation of the pletely unrelated They are after all b oth informative registration pro cedure few assumptions are made ab out measures of the prop erties of human tissue Using a the nature of the imaging pro cess As a result the al large corpus of data or some physical theory it might gorithms are quite general and can foreseeably b e used b e p ossible to construct a function F that predicts with a wide variety of imaging devices CT from the corresp onding MR value at least approxi This approach works directly wi