Methods and systems for detecting acquisition errors in medical images
Inventors
Putha, Preetham • Tadepalli, Manoj • Reddy, Bhargava • Raj, Tarun • Jagirdar, Ammar • Rao, Pooja • Warier, Prashant
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-11508065-B2
Publication Date
2022-11-22
Expiration Date
Abstract
This disclosure generally pertains to methods and systems for automatically detecting acquisition errors in a medical image using machine learning. Certain embodiments relate to methods for the development of deep learning algorithms that perform machine recognition of specific features and conditions in imaging and other medical data. Another embodiment provides systems for detecting acquisition errors in an X-ray image, the system comprising a non-transitory computer-readable medium storing a preprocessing quality control module that, when executed by at least one electronic processor, is configured to generate associated classifications identifying characteristics of the medical image.
Core Innovation
A preprocessing quality control module for detecting acquisition errors in an X-ray image generates associated classifications identifying characteristics of the medical image. The classifications correspond to anterior-posterior/posterior-anterior view, erect/supine, clipped anatomy, under/over exposure, patient rotation, and inadequate inspiration.
The module is developed with machine learning and uses training X-ray images that are manually labeled with the respective characteristics and projection views, posture, exposure status, or inspiration-related anatomy conditions. For clipped anatomy, the required anatomy is clipped out in each training X-ray image using a deep learning based computer vision module.
For patient rotation, the patient rotation classifier comprises a U-net based segmentation model trained to give heatmaps of clavicles and spinous processes. For inspiration, the inspiration classifier comprises two U-net based segmentation models to segment the ribs and to segment the diaphragm respectively.
An ensemble model combines the anterior-posterior/posterior-anterior classifier, the erect/supine classifier, the clipped anatomy classifier, the under/over exposure classifier, the patient rotation classifier, and the inspiration classifier to generate a final algorithm. The output is used to determine whether the X-ray image is diagnostically acceptable or diagnostically unacceptable.
Claims Coverage
The document contains two independent claims (system and method) that cover a preprocessing quality control module for detecting acquisition errors by generating multiple machine-learning classifications for distinct X-ray acquisition characteristics, then combining them using an ensemble model to generate a final algorithm. Across the independent claims, the inventive features are organized around six machine-learning classifiers and an ensemble model that produces a final prediction regarding diagnostic acceptability.
Preprocessing quality control module that generates acquisition-error classifications
A preprocessing quality control module configured to generate associated classifications identifying characteristics of the medical image for anterior-posterior/posterior-anterior, erect/supine, clipped anatomy, under/over exposure, patient rotation, and inspiration.
Ensemble model combining six acquisition-error classifiers into a final algorithm
An ensemble model configured to combine the anterior-posterior/posterior-anterior classifier, the erect/supine classifier, the clipped anatomy classifier, the under/over exposure classifier, the patient rotation classifier, and the inspiration classifier to generate a final algorithm.
Anterior-posterior/posterior-anterior classifier with manually labeled ap/pa training images
An anterior-posterior/posterior-anterior classifier developed by performing machine learning using a first set of training X-ray images manually labeled with an anterior-posterior (AP) projection view or a posterior-anterior (PA) projection view.
Erect/supine classifier with manually labeled erect/supine training images
An erect/supine classifier developed by performing machine learning using a second set of training X-ray images manually labeled with erect or supine.
Clipped anatomy classifier using deep learning based computer vision to clip required anatomy in training images
A clipped anatomy classifier developed by performing machine learning using a third set of training X-ray images wherein a required anatomy is clipped out in each of the third set of training X-ray images using a deep learning based computer vision module.
Under/over exposure classifier with manually labeled under/over training images
An under/over exposure classifier developed by performing machine learning using a fourth set of training X-ray image manually labeled with under or over exposure.
Patient rotation classifier using U-net based segmentation heatmaps
A patient rotation classifier developed by performing machine learning using a fifth set of training X-ray images wherein the patient rotation classifier comprises a U-net based segmentation model trained to give heatmaps of clavicles and spinous processes.
Inspiration classifier using two U-net based segmentation models for ribs and diaphragm
An inspiration classifier developed by performing machine learning using a sixth set of training X-ray images wherein the inspiration classifier comprises two U-net based segmentation models to segment the ribs and to segment the diaphragm respectively.
Diagnostic acceptability prediction output from plurality of classifications via preprocessing quality control module
Determining a plurality of classifications of the X-ray image via a preprocessing quality control module developed with machine learning and outputting a prediction on whether the X-ray image is diagnostically acceptable or diagnostically unacceptable.
Across the independent claims, the claim coverage centers on a preprocessing quality control module that produces multiple acquisition-error related classifications and combines them by an ensemble model to generate a final algorithm. The method claims further require outputting a diagnostic acceptability prediction based on the combined results.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Not explicitly described in patent.
Interested in licensing this patent?