Methods of assessing lung disease in chest x-rays
Inventors
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12039724-B2
Publication Date
2024-07-16
Expiration Date
Abstract
The present invention provides systems and methods for analyzing chronic pulmonary diseases, such as lung cancer, using machine learning (ML) systems that detect lung nodules in chest x-rays. The methods and systems of the invention allows for the detection of lung abnormalities in chest x-ray images using at least two neural networks.
Core Innovation
The invention provides a system for detecting lung abnormalities by processing a chest x-ray image with an image pre-processing module that resizes the chest x-ray image to produce a first image at a down-sampled or up-sampled resolution and segments the image into at least one subsection representing an organ of a body. A collection of neural networks includes a first neural network that analyzes the first image and a second neural network that analyzes the subsection, and each neural network independently makes an inference as to the presence of an abnormality.
An ensemble classifier reports the presence of an abnormality at a location in the lung using the collection of neural network inferences as inputs. For lung nodules, the system outputs first and second sets of scores indicating probabilities of nodules at locations in the lung from the first neural network and the second neural network, and the ensemble classifier combines the first set of scores and the second set of scores to report the presence of a nodule at the location in the lung.
For location-focused detection, the invention performs object detection on the chest x-ray file, creates a bounding box for an object detected in the file, and selects the subsection for the second image from within the box. In the diagnostic method, an image file of a chest x-ray from a patient is resized into a first image depicting the entire x-ray at a reduced resolution and a subsection is placed into a second image at an original resolution, and the first and second images are analyzed in parallel by respective first and second neural networks to output scores indicating a probability of a nodule.
The invention addresses the need to detect lung abnormalities, including lung nodules and nodule locations, from chest x-rays while managing image quality and positioning accuracy. The system is operable to analyze the first image for quality and positioning accuracy and reject an image from processing, and, when rejected, provide a real-time notification instructing a technician to acquire another image.
Claims Coverage
The partial content includes three independent claims directed to detecting lung abnormalities or nodules and a diagnostic method. Across the independent claims, the main inventive features are organized around resizing and organ-related subsection selection, parallel/global and local neural network inference, an ensemble classifier that reports abnormality or nodule presence at a lung location, object-detection-based subsection selection, and quality gating with real-time technician notification.
Ensemble detection using down-sampled whole image and organ subsections
An image pre-processing module resizes a chest x-ray image to produce a first image at a down-sampled or up-sampled resolution and segments the image into at least one subsection representing an organ of a body; a collection of neural networks includes a first neural network analyzing the first image and a second neural network analyzing the subsection, each independently making an inference as to the presence of an abnormality; and an ensemble classifier reports the presence of an abnormality at a location in the lung using the collection of neural network inferences as inputs.
Quality and positioning accuracy rejection with real-time technician notification
The system is operable to analyze the first image for quality and positioning accuracy and reject an image from processing and, when rejected, provide a real-time notification instructing a technician to acquire another image.
Two-network scoring for nodule probabilities on down-sampled and original-resolution views
An image pre-processing module resizes a chest x-ray file to produce a first image at a down-sampled resolution and places a subsection of the chest x-ray file into a second image at an original resolution; a first neural network analyzes the first image to output a first set of scores indicating probabilities of nodules at locations in the lung; a second neural network analyzes the second image to output a second set of scores of probabilities of a nodule at a location in the lung; and an ensemble classifier reports the presence of the nodule in the location in the lung using the first set of scores and the second set of scores as inputs.
Object detection bounding box subsection selection for nodule scoring
The second neural network performs an object detection operation on the chest x-ray file, creates a bounding box for an object detected in the file, and selects the subsection for the second image from within the box.
Parallel analysis with AUC-constrained learned associations for diagnostic detection
A diagnostic method provides an image file of a chest x-ray from a patient to a machine learning system that resizes the image into a first image depicting the entire x-ray at a reduced resolution and places a subsection into a second image at an original resolution; analyzes the first and second images in parallel by respective first and second neural networks to output scores indicating a probability of a nodule; wherein the machine learning system has been trained to learn associations between features in chest x-rays and known pathology results with an area under the curve (AUC) of true positives over false positives for learned feature associations between 0.7 and 1; and operates the machine learning system to detect lung nodules.
Across the independent claims, the core claim coverage centers on detecting lung abnormalities or nodules from chest x-rays using resizing and organ-related subsection creation, parallel global and local neural network inference, and an ensemble classifier that reports presence at lung locations. The independent claims further cover quality and positioning rejection with real-time technician notification, object-detection bounding boxes to select the subsection for scoring, and a diagnostic method that includes training performance characterized by an AUC range between 0.7 and 1 for true positives over false positives.
Stated Advantages
Enables detection of lung abnormalities and reporting presence at a location in the lung.
Provides location reporting for lung nodules using combined neural-network inferences and scores.
Improves processing reliability by rejecting images based on quality and positioning accuracy and providing real-time notification to acquire another image.
Documented Applications
Detecting lung abnormalities, including lung nodules, from patient chest x-ray images.
Detecting lung nodules in a diagnostic method using a machine learning system trained with known pathology results.
Interested in licensing this patent?