Electromagnetic imaging apparatus and process

Inventors

AL-SAFFAR, Ahmed • ABBOSH, Amin

Assignees

Emvision Medical Devices Ltd

Abstract

A computer-implemented process for electromagnetic imaging, the process including the steps of: accessing scattering data representing measurements of electromagnetic wave scattering by internal features of an object, each said measurement representing scattering of electromagnetic waves emitted by a corresponding antenna of an array of antennas disposed about an imaging domain containing at least a portion of the object, and as measured by a corresponding antenna of the array of antennas; and processing the scattering data to generate image data representing a spatial location and size of at least one internal feature of the object within the imaging domain; wherein the processing includes applying a trained message-passing graph neural network (GNN) to a graph of nodes representing spatial locations of the antennas and edges representing the measurements.

Core Innovation

The invention relates to a computer-implemented process and an electromagnetic imaging apparatus for electromagnetic imaging that uses scattering data representing electromagnetic wave scattering by internal features of an object. The scattering data comprises measurements of scattering of electromagnetic waves emitted by antennas of an array disposed about an imaging domain containing at least a portion of the object, and the processed scattering data generates image data representing a spatial location and size of at least one internal feature or a spatial distribution of at least one internal feature within the imaging domain.

The core processing applies a trained message-passing graph neural network (GNN) to a graph where nodes represent spatial locations of the antennas and edges represent the measurements. The document further explains node-level label generation in which inferred node labels encode ground-truth target overlap along lines connecting antennas, producing quantitatively defined degrees of overlap as part of the inferred node labels. The GNN message passing uses edge attribute preprocessing and attention-weighted aggregation of messages, followed by a localizer network outputting node label codes.

For image reconstruction, the inferred node labels are decoded into overlap-weighted geometric sectors to form node-wise partial images, and these partial images are summed or combined into a composite or total image. The document describes that partial images are generated with overlap-degree constraints so that contributions are included based on whether overlap degrees are greater than zero for antenna pair contributions, and that each partial image is constructed from weighted geometric sectors per transmitting antenna, corresponding to other antennas whose inferred overlap degrees determine sector weighting.

Claims Coverage

The document includes two independent claims that cover, respectively, a computer-implemented electromagnetic imaging process and an electromagnetic imaging apparatus. Each independent claim is centered on applying a trained message-passing GNN to an antenna-based graph over scattering measurements to generate image data representing spatial location and size or spatial distribution of internal features, with dependent claims specifying additional inventive refinements to message passing and image construction.

Overall claim coverage focuses on using scattering measurements from an antenna array as edges in a spatially defined antenna graph, and applying a trained message-passing GNN to infer node-level information that enables reconstruction of image data representing internal feature location/size or spatial distribution. Dependent claims further narrow message passing and image generation, including edge summarization and attention-weighted message weighting, and decoding overlap-degree-related node labels into weighted geometric sector partial images that are combined.

Stated Advantages

Documented Applications