A computer-assisted detection method is provided for detecting suspicious locations of lesions in the volumetric medical images. The method includes steps of features extraction and fusion. The first step is computing gradient feature for extraction of the layer of Partial Volume Effect (LPVE) between different tissues that related to specific organs. The LPVE will combine with the result of voxel classification to fulfill the task of tissue classification. After tissue classification, the contour of tissue boundary is determined. The gradient feature is also used to determine the direction that intensity changes. This direction that intensity changes most dramatically serves as the normal vector for voxel on the contour of the tissue boundary. The second step is to determine a local surface patch on the contour for each voxel on the contour. A local landmark system is then created on that patch and the so-called Euclidean Distance Transform Vector (EDTV) is computed based on those landmarks. The EDTV is the basic shape feature for lesion detection whose development and invasion results abnormal shape change on the tissue boundary. A vector classification algorithm for pattern recognition based on EDTVs is also provided. The voxel on the contour of tissue boundary can be grouped into areas based on similar pattern to form lesion patch and local lesion volume. That area will further be analyzed for estimation of the likelihood of lesion.