US 2021/0174504 A1Published application

Specialized computer-aided diagnosis and disease characterization with a multi-focal ensemble of convolutional neural networks

AssigneesCase Western Reserve University, Louis Stokes Cleveland Veterans Administration Medical Center

InventorsAnant Madabhushi, Nathaniel Braman, Tristan Maidment, Yijiang Chen

Life of this patent

Estimating the term…

Loading patent…

Abstract

Embodiments discussed herein facilitate determination of whether lesions are benign or malignant. One example embodiment is a method, comprising: accessing medical imaging scan(s) that are each associated with distinct angle(s) and each comprise a segmented region of interest (ROI) of that medical imaging scan comprising a lesion associated with a first region and a second region; providing the first region(s) of the medical imaging scan(s) to trained first deep learning (DL) model(s) of an ensemble and the second region(s) of the medical imaging scan(s) to trained second DL model(s) of the ensemble; and receiving, from the ensemble of DL models, an indication of whether the lesion is a benign architectural distortion (AD) or a malignant AD.

Claims26

Dependent claims

A non-transitory computer-readable medium storing computer-executable instructions that, when executed, cause a processor to perform operations, comprising:

accessing one or more medical imaging scans, wherein each medical imaging scan of the one or more medical imaging scans is associated with a distinct angle, wherein each medical imaging scan of the one or more medical imaging scans comprises a segmented region of interest (ROI) of that medical imaging scan comprising a lesion, wherein the segmented ROI of that medical imaging scan comprises one or more regions associated with that lesion;
providing a first region of the one or more regions of a first medical imaging scan of the one or more medical imaging scans to a first deep learning (DL) model of an ensemble of DL models and providing a second region of the one or more regions of a second medical imaging scan of the one or more medical imaging scans to a second DL model of the ensemble, wherein the first DL model is trained on the first region and the distinct angle of the first medical imaging scan, wherein the second DL model is trained on the second region and the distinct angle of the second medical imaging scan, and wherein at least one of the first region is different from the second region or the first medical imaging scan is different from the second medical imaging scan;
and
receiving, from the ensemble of DL models, an indication of whether the lesion is benign or malignant.

1 more element

A non-transitory computer-readable medium storing computer-executable instructions that, when executed, cause a processor to perform operations, comprising:

accessing a training set comprising, for each patient of a plurality of patients, one or more medical imaging scans of that patient, wherein each medical imaging scan of the one or more medical imaging scans is associated with a distinct angle, wherein each medical imaging scan of the one or more medical imaging scans of that patient comprises a segmented region of interest (ROI) of that medical imaging scan comprising a lesion of that patient, wherein the segmented ROI of that medical imaging scan comprises a first region of that medical imaging scan associated with the legion of that patient and a second region of that medical imaging scan associated with the lesion of that patient, and wherein the lesion of that patient is associated with a known diagnosis for that patient;
for each medical imaging scan of the training set:
randomly extracting one or more associated first samples from the first region of that medical imaging scan and one or more associated second samples from the second region of that medical imaging scan;
assigning an associated weight to each associated first sample of the one or more associated first samples and to each associated second sample of the one or more associated second samples based at least in part on the known diagnosis for that patient;
providing the one or more associated first samples of that medical imaging scan and their associated weights to an associated first Deep Learning (DL) model for the distinct view of that medical imaging scan;
providing the one or more associated second samples of that medical imaging scan and their associated weights to an associated second DL model for the distinct view of that medical imaging scan;
training the associated first DL model for the distinct view of that medical imaging scan based on the one or more associated first samples of that medical imaging scan and their associated weights;
and
training the associated second DL model for the distinct view of that medical imaging scan based on the one or more associated second samples of that medical imaging scan and their associated weights;
and
determining an associated DL model weighting for each DL model of an ensemble of DL models comprising the first DL for the distinct view of each medical imaging scan and the second DL model for the distinct view of each medical imaging scan.

8 more elements

An apparatus, comprising:

a memory configured to store one or more medical imaging scans, wherein each medical imaging scan of the one or more medical imaging scans is associated with a distinct angle, wherein each medical imaging scan of the one or more medical imaging scans comprises a segmented region of interest (ROI) of that medical imaging scan comprising a lesion, wherein the segmented ROI of that medical imaging scan comprises one or more regions associated with that lesion;
and
one or more processors configured to perform operations comprising:
providing a first region of the one or more regions of a first medical imaging scan of the one or more medical imaging scans to a first deep learning (DL) model of an ensemble of DL models and providing a second region of the one or more regions of a second medical imaging scan of the one or more medical imaging scans to a second DL model of the ensemble, wherein the first DL model is trained on the first region and the distinct angle of the first medical imaging scan, wherein the second DL model is trained on the second region and the distinct angle of the second medical imaging scan, and wherein at least one of the first region is different from the second region or the first medical imaging scan is different from the second medical imaging scan;
and
receiving, from the ensemble of DL models, an indication of whether the lesion is benign or malignant.

3 more elements

Dependencies are inferred from the claim text.

Facts

Application: 17/116,366Filed 9 Dec 2020
Priority: 9 Dec 2019
Published: 10 Jun 2021
Term ends: Not estimatedThe source records don’t support an estimate.

Classification33 codes

G16H 30/40
ICT specially adapted for the handling or processing of medical images: for processing medical images, e.g. editing
A61B 6/025
Arrangements for diagnosis sequentially in different planes; Stereoscopic radiation diagnosis: tomosynthesis
A61B 6/469
Characterised by special input means: for selecting a region of interest [ROI]
A61B 6/502
Specially adapted for specific body parts; specially adapted for specific clinical applications: for diagnosis of breast, i.e. mammography

Family

Application62/945,327Us provisional application · filed 9 Dec 2019
US 11,817,204 B2Related patent
US 2021/0174504 A1This patent

From the US filing history. Foreign family members aren’t listed.

Timeline

Citations

Open elsewhere

Building an agent or pulling many patents? Use the API or MCP server instead of scraping pages. Get an API key