Abstract: A computer-implemented method for determining a pathology in 3D image data is describe wherein the method may comprise:receiving at least a first 3D image of a body part, the 3D image comprising voxels associated with a predetermined image volume; a first 3D convolutional neural network determining a position of a volume of interest (VOI) in the image volume of the first 3D image, the VOI being associated with a pathology of the body part, the VOI defining a sub-volume of the image volume; determining first VOI voxels by selecting voxels of the first 3D image that have a position within the VOI as determined by the first 3D convolution neural network and providing the first VOI voxels to the input of a second 3D convolutional neural network; the second 3D convolutional ON neural network, determining a target label value on the basis of at least the first VOI voxels, the target label value being indicative of the presence or absence of a pathology in the VOI; and, generating a medical report by associating the