A computer-implemented method of detecting an object in a three-dimensional medical image 302 comprises determining 320 the values of a plurality of features at each voxel in at least a portion of the medical image. Each feature characterises a respective property of the medical image at a particular voxel. The likelihood probability distribution of each feature is calculated 330 based on the values of the features and prior medical knowledge, wherein the prior medical knowledge comprises one or more parameters derived from training data. A probability map is generated 340 by using Bayes' law to combine the likelihood probability distributions, and the probability map is analysed 350 to detect an object. The feature may be an appearance feature, shape feature or anatomical feature. The appearance feature may comprise image intensity information or a wavelet feature. The shape feature may be a second order shape feature calculated from eigenvalues of a hessian matrix.
展开▼
