Systems and methods for automated analysis of medical images
Inventors
TRAN, Dang-Dinh-Ang • SEAH, Jarrel • Huang, David • Vuong, David • HOLT, Xavier • Nothrop, Marc Justin • Austin, Benjamin • Lee, Aaron • AMOROSO, Marco
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-11646119-B2
Publication Date
2023-05-09
Expiration Date
Abstract
This disclosure relates to detecting visual findings in anatomical images. Methods comprise inputting anatomical images into a neural network to output a feature vector and computing an indication of visual findings being present in the images by a dense layer of the neural network that takes as input the feature vector and outputs an indication of whether each of the visual findings is present in the anatomical images. The neural network is trained on a training dataset including anatomical images, and labels associated with the anatomical images and each of the visual findings. The visual findings may be organised as a hierarchical ontology tree. The neural network may be trained by evaluating the performance of neural networks in detecting the visual findings and a negation pair class which comprises anatomical images where a first visual finding is identified in the absence of a second visual finding.
Core Innovation
The invention relates to detecting a plurality of visual findings in one or more anatomical images of a subject. One or more anatomical images are input into a first convolutional neural network component of a primary neural network to output a feature vector, and a dense layer takes the feature vector as input and outputs an indication of whether each of the plurality of visual findings is present in at least one of the one or more anatomical images.
The primary neural network is trained on a training dataset that includes, for each of a plurality of subjects, one or more anatomical images and a plurality of labels associated with the one or more anatomical images and each of the respective visual findings. In training and/or evaluation, performance of a plurality of neural networks in detecting the plurality of visual findings is evaluated while accounting for correlation between one or more pairs of the plurality of visual findings.
The plurality of visual findings is organized as a hierarchical ontology tree, and training comprises evaluating performance at different levels of the hierarchy of the ontology tree. The training and/or evaluation can also account for at least one negation pair class, comprising anatomical images where a first one of the plurality of visual findings is identified in the absence of a second one of the plurality of visual findings, with label mapping that covers internal nodes and terminal leaves.
Claims Coverage
The provided set includes six independent method claims for detecting and for training a neural network to detect a plurality of visual findings in one or more anatomical images. The inventive coverage centers on a primary CNN-based network with a dense layer producing per-finding presence indications, with training and evaluation that is correlation-aware, negation-pair-aware, and hierarchical-ontology-tree level-aware.
Primary cnn feature vector with dense per-finding presence output
Inputting one or more anatomical images into a first convolutional neural network component of a primary neural network to output a feature vector, and computing an indication of whether each of the plurality of visual findings is present in at least one of the one or more anatomical images by a dense layer of the primary neural network that takes as input the feature vector.
Correlation-aware performance evaluation using plurality of neural networks
Training the primary neural network by evaluating performance of a plurality of neural networks in detecting the plurality of visual findings, wherein the performance evaluation accounts for correlation between one or more pairs of the plurality of visual findings.
Hierarchical ontology tree training across hierarchy levels
Organising the plurality of visual findings as a hierarchical ontology tree, and evaluating performance of the primary neural network at different levels of the hierarchy of the ontology tree.
Negation pair class evaluation capturing presence-in-absence
Training the primary neural network by evaluating the performance of a plurality of neural networks in detecting the plurality of visual findings and at least one negation pair class which comprises anatomical images where a first one of the plurality of visual findings is identified in the absence of a second one of the plurality of visual findings.
Training with data-store retrieved labeled dataset
Retrieving, from a data store, a training dataset including, for each of a plurality of subjects, one or more anatomical images and a plurality of labels associated with the one or more anatomical images and each of the respective visual findings, and training the primary neural network using the training dataset.
Across the independent claims, the invention covers detection using a CNN feature vector followed by a dense layer producing per-finding presence indications, with training and/or evaluation that incorporates correlation accounting and may also incorporate negation pair classes and hierarchical ontology tree level-based evaluation.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Not explicitly described in patent.
Interested in licensing this patent?