System and computer-implemented method for segmenting an image

A computer-implemented method for segmenting an input image, the method comprising:

• receiving an input image into a machine learning system for performing segmentation of the image based on a set of model parameters, the machine learning system having been trained using a previous set of training images provided with labelled segmentations, wherein the input image is received without a labelled segmentation;
• performing a segmentation of the input image into multiple segments by solving jointly for:
• (i) a segmentation of the input image;
• and (ii) an updated set of model parameters for the machine learning system, whereby the machine learning system is tuned by the input image;
• receiving, from a user into the machine learning system, at least one indication, wherein each indication corresponds to a particular segment from the multiple segments, and marks one or more locations of the input image as belonging to that particular segment;
• and
• performing a further, refined, segmentation of the input image into multiple segments using the at least one indication by solving jointly for:
• (i) a segmentation of the input image;
• and (ii) an updated set of model parameters for the machine learning system.

A computer-implemented method for segmenting a multi-modal input image using a multi-layered machine learning system, the method comprising:

• receiving an input image into the machine learning system comprising multiple imaging modalities (M-space) and generating multiple modality-conditioned features (F-space);
• processing the input image by performing a first transform within one or more layers of the machine learning system across F-space and a second transform within one or more further layers of the machine learning system across M-space to perform the segmentation, wherein the machine learning system comprises:
• a back-end which processes the input image by performing said first and second transforms, and which merges the output from said first and second transforms across M-space;
• and
• a front-end which receives the back-end merged output and generates therefrom a segmentation of the input image based on a monomodal-like analysis.