Training system for detection and classification of artificial objects in X-ray images
AssigneeRaPID Platforms LLC
InventorsVicko Gluncic, Gady Agam, Mario Moric
Life of this patent
Estimating the term…
Abstract
The present invention involves a software based system and method which provides a series of training images with or without retained surgical items in scans that are used to train human physicians to detect such items at varying levels of difficulty.
Claims19
A server for training a user on detection of Retained Foreign Objects having radiographic markers adapted to change shape during surgery (RFOs) in images, said server comprising:
- an image database including a plurality of images of bodies, a first portion of the plurality of images having RFOs depicted, a second portion of the plurality of images lacking a depicted RFO;
- an interaction module for sending a sub-set of that plurality of images to a user device, the interaction module also adapted to receive user communication relating to the presence or absence of an RFO in a particular image;
- and
- a recording module for recording the correspondence between user communication received relating to each of the first portion and second portion of the images.
A method of using a computer to synthesize the presence of a Retained Foreign Object having radiographic markers adapted to change shape during surgery (RFO) in a body image, said method comprising the steps of:
- obtaining a first image of a body without any RFOs;
- obtaining a second image of a body having an RFO;
- extracting features of the RFO from the second image;
- and
- superimposing the extracted features with the first image to create said synthesized image.
A method of training human observers in visual detection of predetermined objects within areas of interest around a surgical field in a scan image, said method comprising the steps of:
- computing the contrast difference between areas of interest relating to the predetermined objects and their neighborhood using areas of interest in scan images;
- capturing scan images of foreign objects of interest with contrast with a known level of difficulty;
- segmenting the scan images of areas of interest to form a first binary mask;
- and
- transforming the first mask and the areas of interest in the image according to the random rotation (R) and scale (S) parameters;
- and
- smoothing the mask M1 using a Gaussian smoothing filter to produce a second binary mask so that the transition between areas of interest and other areas is smooth.
Dependencies are inferred from the claim text.