IP Library › Granted Patent US 12,450,731
Granted Patent B2
US 12,450,731 · App. 17/583,954 · Granted Oct 21, 2025

Automated spine health assessment using neural networks

Inventors: Chamith Sudesh Rajapakse (Cherry Hill, NJ); Abhinav Suri (San Antonio, TX)
Assignee: THE TRUSTEES OF THE UNIVERSITY OF PENNSYLVANIA
G06T7/0012A61B6/505A61B6/5217G06T7/11G16H30/40G16H50/30G06T2207/20084G06T2207/30012
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,450,731
App. No.
17/583,954
Granted
Oct 21, 2025
Kind
B2
Abstract

In some examples, a method includes receiving a medical image of at least a portion of a spine in a patient. The method includes supplying the medical image to a neural network trained using training spine images and, for each training spine image, one or more spine measurement annotations. The method includes detecting, using the neural network, five or more vertebral landmarks for each of a plurality of vertebral bodies depicted in the medical image. The method includes outputting, for at least a first vertebral body, one or more deformity measurements based on the vertebral landmarks for the first vertebral body.

Claims (44)

1. A method for automatic assessment of spine health, the method comprising:

receiving a medical image of at least a portion of a spine in a patient;

supplying the medical image to a deep learning network trained using a plurality of training spine images and, for each training spine image, one or more spine measurement annotations, wherein supplying the medical image to the deep learning network includes supplying the medical image to a feature generation network that performs convolutions and produces, as outputs, learnable features;

providing the outputs from the feature generation network to a region recognition network that produces, as output, an objectness logits map showing a probability of an approximate region containing an object;

combining the features and the objectness logits map to produce a combined output from the region recognition network and providing the combined output to a landmark network;

detecting, using the landmark network, five or more vertebral landmarks for each of a plurality of vertebral bodies depicted in the medical image; and

outputting, for at least a first vertebral body, one or more deformity measurements based on the vertebral landmarks for the first vertebral body.

2. The method of claim 1 , comprising determining an anterior height, middle height, and posterior height of each vertebral body using the vertebral landmarks for the vertebral body.

3. The method of claim 2 , comprising performing a deformity analysis of each vertebral body using the anterior height, middle height, and posterior height of the vertebral body.

4. The method of claim 1 , comprising determining a measure of lumbar lordosis using the vertebral landmarks of the vertebral bodies.

5. The method of claim 1 , comprising producing individual segmentations for each vertebral body.

6. The method of claim 5 , comprising generating, by the region recognition network, a final bounding box for each vertebral body, outputting the final bounding box to a segmentation network, and wherein producing the individual segmentations comprises generating, by the segmentation network, a segmentation mask within each final bounding box.

7. The method of claim 1 , wherein the medical image is one of: a magnetic resonance imaging (MRI) image, a computed tomography (CT) image, and an X-ray image.

8. A system for non-invasively predicting patient-specific bone resilience or toughness, the system comprising:

at least one processor; and

a spine health assessor implemented on the at least one processor and configured to perform operations comprising:

receiving a medical image of at least a portion of a spine in a patient;

supplying the medical image to a deep learning network trained using a plurality of training spine images and, for each training spine image, one or more spine measurement annotations, wherein supplying the medical image to the deep learning network includes supplying the medical image to a feature generation network that performs convolutions and produces, as outputs, learnable features;

providing the outputs from the feature generation network to a region recognition network that produces, as output, an objectness logits map showing a probability of an approximate region containing an object;

combining the features and the objectness logits map to produce a combined output from the region recognition network and providing the combined output to a landmark network;

detecting, using the landmark network, five or more vertebral landmarks for each of a plurality of vertebral bodies depicted in the medical image; and

outputting, for at least a first vertebral body, one or more deformity measurements based on the vertebral landmarks for the first vertebral body.

9. The system of claim 8 , the operations comprising determining an anterior height, middle height, and posterior height of each vertebral body using the vertebral landmarks for the vertebral body.

10. The system of claim 9 , the operations comprising performing a deformity analysis of each vertebral body using the anterior height, middle height, and posterior height of the vertebral body.

11. The system of claim 8 , the operations comprising determining a measure of lumbar lordosis using the vertebral landmarks of the vertebral bodies.

12. The system of claim 8 , the operations comprising producing individual segmentations for each vertebral body.

13. The system of claim 12 , the operations comprising generating, by the region recognition network, a final bounding box for each vertebral body, outputting the final bounding box to a segmentation network, and wherein producing the individual segmentations comprises generating, by the segmentation network, and a segmentation mask within each final bounding box.

14. The system of claim 8 , wherein the medical image is one of: a magnetic resonance imaging (MRI) image, a computed tomography (CT) image, and an X-ray image.

15. A non-transitory computer readable medium storing executable instructions that when executed by at least one processor of a computer control the computer to perform operations comprising:

receiving a medical image of at least a portion of a spine in a patient;

supplying the medical image to a deep learning network trained using a plurality of training spine images and, for each training spine image, one or more spine measurement annotations, wherein supplying the medical image to the deep learning network includes supplying the medical image to a feature generation network that performs convolutions and produces, as outputs, learnable features;

providing the outputs from the feature generation network to a region recognition network that produces, as output, an objectness logits map showing a probability of an approximate region containing an object;

combining the features and the objectness logits map to produce a combined output from the region recognition network and providing the combined output to a landmark network;

detecting, using the landmark network, five or more vertebral landmarks for each of a plurality of vertebral bodies depicted in the medical image; and

outputting, for at least a first vertebral body, one or more deformity measurements based on the vertebral landmarks for the first vertebral body.

16. A method for automatic assessment of spine health, the method comprising:

receiving at least one medical image of at least a portion of a spine in a patient;

supplying the at least one medical image to a deep learning network trained

using a plurality of training spine images and, for each training spine image,

one or more spine measurement annotations, wherein supplying the medical image to the deep learning network includes supplying the medical image to a feature generation network that performs convolutions and produces, as outputs, learnable features;

providing the outputs from the feature generation network to a region recognition network that produces, as output, an objectness logits map showing a probability of an approximate region containing an object;

combining the features and the objectness logits map to produce a combined output from the region recognition network and providing the combined output to a landmark network;

producing, using the landmark network, individual segmentation for each of a plurality of vertebral bodies or intervertebral disks depicted in the at least one medical image; and

outputting, for at least one vertebral body or intervertebral disk, one or more anatomical measurements based on the vertebral landmarks for the at least one vertebral body.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Feb 7, 2022
From: RAJAPAKSE, CHAMITH SUDESH; SURI, ABHINAV
To: THE TRUSTEES OF THE UNIVERSITY OF PENNSYLVANIA
Reel/Frame 058912/0272 →
Continuity (2)
Provisional Application 63141280 · Jan 25, 2021
Related Publication 20220237779A1 · Jul 28, 2022
References Cited (34)
US 8891848B2 · Blumfield et al. · 2014 [cited by applicant]
US 8965083B2 · Ben Ayed et al. · 2015 [cited by applicant]
US 10366491B2 · Yang et al. · 2019 [cited by applicant]
US 20070242869A1 · Luo et al. · 2007 [cited by applicant]
US 20080044074A1 · Jerebko et al. · 2008 [cited by applicant]
US 20190370957A1 · Manickam · 2019 [cited by examiner]
US 20210097678A1 · Dutta · 2021 [cited by examiner]
Sekuboyina et al., “VerSe: A Vertebrae labelling and segmentation benchmark for multi-detector CT images,” Medical Image Analysis, vol. 73, pp. 1-32 (2021). [cited by applicant]
Weaver et al., “Prevalanece and Cost of Subsequent Fractures Among U.S. Patients with an Incident Fracture,” J Manag Care Spec Pharm, vol. 23, No. 4, pp. 1-13 (2017). [cited by applicant]
Rajapakse et al., “Vertebral deformities and fractures are associated with MRI and pQCT measures obtained at the distal tibia and radius of postmenopausal women,” Osteoporosis Int, vol. 25, pp. 973-982 (2014). [cited by applicant]
Wong et al., “Vertebral compression fractures: a review of current management and multimodal therapy,” Journal of Multidisciplinary Healthcare, vol. 6, pp. 205-214 (2013). [cited by applicant]
Cunha-Henriques et al., “Postmenopausal Women with Osteoporosis and Musculoskeletal Status: A Comparative Cross-Sectional Study,” J Clin Med Res, vol. 3, No. 4, pp. 168-176 (2011). [cited by applicant]
Bartalena et al., “Prevalence of thoracolumbar vertebral fractures on multidetector CT Underreporting by radiologists,” European Journal of Radiology, vol. 69, pp. 555-559 (2009). [cited by applicant]
Salaffi et al., “The Burden of Prevalent Fractures on Health-Related Quality of Life in Postmenopausal Women with Osteoporosis: The IMOF Study,” The Journal of Rheumatology, pp. 1-10 (2007). [cited by applicant]
“Bone Health and Osteoporosis: A Report of the Surgeon General,” Department of Health and Human Services, pp. 1-437 (2004). [cited by applicant]
Lenchik et al., “Diagnosis of Osteoporotic Vertebral Fractures: Importance of Recognition and Description by Radiologists,” AJR, vol. 183, pp. 1-10 (Oct. 2004). [cited by applicant]
Silverman et al., “The Relationship of Health-Related Quality of Life to Prevalent and Incident Vertebral Fractures in Postmenopausal Women with Osteoporosis,” Arthritis & Rheumatism, vol. 44, No. 11, pp. 2611-2619 (Nov… [cited by applicant]
Genant et al., “Vertebral Fractures Assessment Using a Semiquantitative Technique,” Journal of Bone and Mineral Research, vol. 8, No. 9, pp. 1137-1148 (1993). [cited by applicant]
Chae et al., “Decentralized convolutional neural network for evaluating spinal deformity with spinopelvic parameters,” Computer Methods and Programs in Biomedicine, vol. 197, pp. 1-12 (2020). [cited by applicant]
Kim et al., “Web-Based Spine Segmentation Using Deep Learning in Computed Tomography Images,” Healthc Inform Res., vol. 26, No. 1, pp. 1-7 (2020). [cited by applicant]
Cho et al., “Automated Measurement of Lumbar Lordosis on Radiographs Using Machine Learning and Computer Vision,” Global Spine Journal, vol. 10, No. 5, pp. 1-8 (2020). [cited by applicant]
Huang et al., “Spine Explorer: a deep learning based fully automated program for efficient and reliable quantifications of the vertebrae and discs on sagittal lumbar spine MR images,” The Spine Journal, vol. 20, pp. 1-1… [cited by applicant]
Masad et al., “Automated measurements of lumbar lordosis in T2-MR images using decision tree classifier and morphological image processing,” Engineering Science and Technology, an International Journal, vol. 22, pp. 1-8… [cited by applicant]
Vania et al., “Automatic spine segmentation from CT images using Convolutional Neural Network via redundant generation of class labels,” Journal of Computational Design and Engineering, vol. 6, pp. 1-9 (2019). [cited by applicant]
Paszke et al., “PyTorch: An Imperative Style, High-Performance Deep Learning Library,” 33rd Conference on Neural Information Processing Systems (NeurIPS 2019), pp. 1-12 (2019). [cited by applicant]
Han et al., “Spine-GAN: Semantic segmentation of multiple spinal structures,” Medical Image Analysis, vol. 50, pp. 1-13 (2018). [cited by applicant]
Lu et al., “Deep Spine: Automated Lumbar Vertebral Segmentation, Disc-Level Designation, and Spinal Stenosis Grading Using Deep Learning,” Proceedings of Machine Learning Research, vol. 85, pp. 1-16 (2018). [cited by applicant]
Ebrahimi et al., “Vertebral corners detection on sagittal X-rays based on shape modeling, random forest classifiers and dedicated visual features,” Computer Methods in Biomechanics and Biomedical Engineering: Imaging & … [cited by applicant]
Chen et al., “Automatic Localization and Identification of Vertebrae in Spine CT via a Joint Learning Model with Deep Neural Networks,” MICCAI 2015, Part I, LNCS 9349, pp. 1-8 (2015). [cited by applicant]
Schindelin et al., “Fiji: an open-source platform for biological-image analysis,” Nature Methods, vol. 9, No. 7, pp. 1-7 (Jul. 2012). [cited by applicant]
Van der Walt et al., “The NumPy Array: A Structure for Efficient Numerical Computation,” Computing in Science & Engineering, pp. 1-9 (Mar./Apr. 2011). [cited by applicant]
Pedregosa et al., “Scikit-learn: Machine Learning in Python,” Journal of Medicine Learning Research, vol. 12, pp. 1-6 (2011). [cited by applicant]
Lin, “Identification of Spinal Deformity Classification with Total Curvature Analysis and Artificial Neural Network,” IEEE Transactions on Biomedical Engineering, vol. 55, No. 1, pp. 376-382 (Jan. 2008). [cited by applicant]
Yushkevich et al., “User-guided 3D active contour segmentation of anatomical structures: Significantly improved efficiency and reliability,” Neurolmage, vol. 31, pp. 1-13 (2006). [cited by applicant]