Zebra Medical Vision Ltd.

Nano-X Ai Ltd.

Datagen Technologies, Ltd.

Jonathan Laserson

Eli Goz

Chen Brestel

Gil Elbaz

Hadas Sheinfeld

Moti Kadosh

Amit Oved

There is provided a method, comprising feeding a medical image into a detector component trained on a first training dataset of medical images annotated with ground truth boxes depicting a visual finding, obtaining boxes each associated with a respective box score indicative of likelihood of the visual finding, converting each respective box into a respective patch, feeding patches into a patch classifier trained on a second training dataset that includes patches extracted from the ground truth box labels of the first training dataset, wherein a patch score for a patch corresponds to a box score obtained from a box corresponding to the patch, obtaining patch scores indicative of likelihood of the visual finding being depicted, and computing a dot product of the box scores and the patch scores, and providing the dot product as an image-level indication of likelihood of the visual finding being depicted in the medical image.

Two stage detector for identification of a visual finding in a medical image

A non-transitory, computer-readable medium includes instructions that causes at least one processing device to display a graphical user interface (GUI) configured to facilitate generating a synthetic dataset including a plurality of images. The GUI includes a dataset size selector to receive user input to indicate a number of images to generate and include in the synthetic dataset; a target object type selector to receive user input indicative of at least one selected target object type to feature in the synthetic dataset; one or more image parameter variability controls to receive user input indicative of at least one variation to include in the synthetic dataset relative to target object representations generated based on the at least one selected target object type; and a dataset generation control to initiate generating the synthetic dataset. The synthetic dataset is generated according to the size input, target object type input, and variability input.

Systems and methods for synthetic data generation

There is provided a system for computing a single-label neural network for detection of an indication of an acute medical condition, comprising: hardware processor(s) executing a code for: providing a multi-label training dataset including anatomical images each associated with a label indicative of visual finding type(s), or indicative of no visual finding types, training a multi-label neural network for detection of the visual finding types(s) in a target anatomical image according to the multi-label training dataset, creating a single-label training dataset including anatomical images each associated with a label indicative of the selected single visual finding type, or indicative of an absence of the single visual finding type, and training a single-label neural network for detection of the single visual finding type, by setting the trained multi-label neural network as an initial baseline of the single-label neural network, and fine-tuning and/or re-training the baseline according to the single-label training dataset.

Systems and methods for detecting an indication of a visual finding type in an anatomical image

A system for prioritizing patients for treatment, comprising: at least one hardware processor executing a code for: feeding anatomical images into a visual filter neural network for outputting a category indicative of a target body region depicted at a target sensor orientation and a rotation relative to a baseline, rejecting a sub-set of anatomical images classified into another category, rotating to the baseline images classified as rotated, identifying pixels for each image having outlier pixel intensity values denoting an injection of content, adjusting the outlier pixel intensity values to values computed as a function of non-outlier pixel intensity values, feeding each the remaining sub-set of images with adjusted outlier pixel intensity values into a classification neural network for detecting the visual finding type, generating instructions for creating a triage list for which the classification neural network detected the indication, wherein patients are selected for treatment based on the triage list.

Systems and methods for pre-processing anatomical images for feeding into a classification neural network

There is provided a method comprising: providing two anatomical images of a target individual, each captured at a unique orientation of the target individual, inputting first and second anatomical images respectively into a first and second convolutional neural network (CNN) of a classifier to respectively output first and second feature vectors, inputting a concatenation of the first and second feature vectors into a fully connected layer of the classifier, and computing an indication of distinct visual finding(s) present in the anatomical images by the fully connected layer, wherein the statistical classifier is trained on a training dataset including two anatomical images of each respective sample individual, each image captured at a respective unique orientation of the target individual, and a tag created based on an analysis that maps respective individual sentences of a text based radiology report to one of multiple indications of visual findings.

Growing community of inventors

Tel Aviv, Israel

Jonathan Laserson