Methods for interactive annotation of medical images in a client-server architecture

A method for predictive two-dimensional (2D) segmentation of medical image frames, comprising:

• determining a first two-dimensional (2D) medical image frame from a set of three-dimensional (3D) medical scan data;
• determining a first user-defined 2D segmentation that annotates a region of interest (ROI) in the first 2D medical image frame based on a first user interaction;
• determining a second 2D medical image frame adjacent to the first 2D medical image frame in the set of 3D medical scan data;
• providing, as input to a set of trained annotation extrapolation neural networks (AENNs), the first 2D medical image frame, the second 2D medical image frame, and the first user-defined 2D segmentation;
• and
• receiving, as output from the trained AENNs, a first predicted 2D segmentation indicating a predicted ROI in the second 2D medical image frame.

A computing system, comprising:

• at least one memory configured to store a set of trained annotation extrapolation neural networks (AENNs);
• and
• at least one processor configured to execute stored instructions to perform actions comprising:
• providing, as input to the set of trained AENNs, a first two-dimensional (2D) medical image frame rendered from a first slice of three-dimensional (3D) medical scan data, a second 2D medical image frame rendered from a second slice adjacent to the first slice in the 3D medical scan data, and a first user-defined 2D segmentation annotating a region of interest (ROI) in the first 2D medical image frame;
• receiving, as output from the set of trained AENNs, a first predicted 2D segmentation indicating a predicted ROI in the second 2D medical image frame;
• providing, as input to the set of trained AENNs, the second 2D medical image frame, a third 2D medical image frame rendered from a third slice adjacent to the second slice in the 3D medical scan data, and the first predicted 2D segmentation;
• and
• receiving, as output from the set of trained AENNs, a second predicted 2D segmentation indicating a predicted ROI in the third 2D medical image frame.