Deep variational auto-encoders for unsupervised glomerular classification

Brendon Lutnick, Rabi Yacoub, Kuang-Yu Jen, John E. Tomaszewski, Sanjay Jain, Pinaki Sarder

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

The adoption of deep learning techniques in medical applications has thus far been limited by the availability of the large labeled datasets required to robustly train neural networks, as well as difficulty interpreting these networks. However, recent techniques for unsupervised training of neural networks promise to address these issues, leveraging only structure to model input data. We propose the use of a variational autoencoder (VAE) which utilizes data from an animal model to augment the training set and non-linear dimensionality reduction to map this data to human sets. This architecture utilizes variational inference, performed on latent parameters, to statistically model the probability distribution of training data in a latent feature space. We show the feasibility of VAEs, using images of mouse and human renal glomeruli from various pathological stages of diabetic nephropathy (DN), to model the progression of structural changes which occur in DN. When plotted in a 2-dimentional latent space, human and mouse glomeruli, show separation with some overlap, suggesting that the data is continuous, and can be statistically correlated. When DN stage is plotted in this latent space, trends in disease pathology are visualized.

Original languageEnglish (US)
Title of host publicationMedical Imaging 2018
Subtitle of host publicationDigital Pathology
PublisherSPIE
Volume10581
ISBN (Electronic)9781510616516
DOIs
StatePublished - Jan 1 2018
EventMedical Imaging 2018: Digital Pathology - Houston, United States
Duration: Feb 11 2018Feb 12 2018

Other

OtherMedical Imaging 2018: Digital Pathology
CountryUnited States
CityHouston
Period2/11/182/12/18

Fingerprint

Diabetic Nephropathies
coders
glomerulus
education
mice
animal models
Structural Models
pathology
Neural networks
inference
progressions
learning
availability
Teaching
Medical applications
Pathology
Animal Models
Learning
Probability distributions
Kidney

Keywords

  • Cross species modeling
  • Knowledge transfer
  • Unsupervised modeling
  • Variational autoencoder

ASJC Scopus subject areas

  • Electronic, Optical and Magnetic Materials
  • Biomaterials
  • Atomic and Molecular Physics, and Optics
  • Radiology Nuclear Medicine and imaging

Cite this

Lutnick, B., Yacoub, R., Jen, K-Y., Tomaszewski, J. E., Jain, S., & Sarder, P. (2018). Deep variational auto-encoders for unsupervised glomerular classification. In Medical Imaging 2018: Digital Pathology (Vol. 10581). [105810C] SPIE. https://doi.org/10.1117/12.2295456

Deep variational auto-encoders for unsupervised glomerular classification. / Lutnick, Brendon; Yacoub, Rabi; Jen, Kuang-Yu; Tomaszewski, John E.; Jain, Sanjay; Sarder, Pinaki.

Medical Imaging 2018: Digital Pathology. Vol. 10581 SPIE, 2018. 105810C.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Lutnick, B, Yacoub, R, Jen, K-Y, Tomaszewski, JE, Jain, S & Sarder, P 2018, Deep variational auto-encoders for unsupervised glomerular classification. in Medical Imaging 2018: Digital Pathology. vol. 10581, 105810C, SPIE, Medical Imaging 2018: Digital Pathology, Houston, United States, 2/11/18. https://doi.org/10.1117/12.2295456
Lutnick B, Yacoub R, Jen K-Y, Tomaszewski JE, Jain S, Sarder P. Deep variational auto-encoders for unsupervised glomerular classification. In Medical Imaging 2018: Digital Pathology. Vol. 10581. SPIE. 2018. 105810C https://doi.org/10.1117/12.2295456
Lutnick, Brendon ; Yacoub, Rabi ; Jen, Kuang-Yu ; Tomaszewski, John E. ; Jain, Sanjay ; Sarder, Pinaki. / Deep variational auto-encoders for unsupervised glomerular classification. Medical Imaging 2018: Digital Pathology. Vol. 10581 SPIE, 2018.
@inproceedings{a05d7c80ef6440fc8ff77d65060e9334,
title = "Deep variational auto-encoders for unsupervised glomerular classification",
abstract = "The adoption of deep learning techniques in medical applications has thus far been limited by the availability of the large labeled datasets required to robustly train neural networks, as well as difficulty interpreting these networks. However, recent techniques for unsupervised training of neural networks promise to address these issues, leveraging only structure to model input data. We propose the use of a variational autoencoder (VAE) which utilizes data from an animal model to augment the training set and non-linear dimensionality reduction to map this data to human sets. This architecture utilizes variational inference, performed on latent parameters, to statistically model the probability distribution of training data in a latent feature space. We show the feasibility of VAEs, using images of mouse and human renal glomeruli from various pathological stages of diabetic nephropathy (DN), to model the progression of structural changes which occur in DN. When plotted in a 2-dimentional latent space, human and mouse glomeruli, show separation with some overlap, suggesting that the data is continuous, and can be statistically correlated. When DN stage is plotted in this latent space, trends in disease pathology are visualized.",
keywords = "Cross species modeling, Knowledge transfer, Unsupervised modeling, Variational autoencoder",
author = "Brendon Lutnick and Rabi Yacoub and Kuang-Yu Jen and Tomaszewski, {John E.} and Sanjay Jain and Pinaki Sarder",
year = "2018",
month = "1",
day = "1",
doi = "10.1117/12.2295456",
language = "English (US)",
volume = "10581",
booktitle = "Medical Imaging 2018",
publisher = "SPIE",

}

TY - GEN

T1 - Deep variational auto-encoders for unsupervised glomerular classification

AU - Lutnick, Brendon

AU - Yacoub, Rabi

AU - Jen, Kuang-Yu

AU - Tomaszewski, John E.

AU - Jain, Sanjay

AU - Sarder, Pinaki

PY - 2018/1/1

Y1 - 2018/1/1

N2 - The adoption of deep learning techniques in medical applications has thus far been limited by the availability of the large labeled datasets required to robustly train neural networks, as well as difficulty interpreting these networks. However, recent techniques for unsupervised training of neural networks promise to address these issues, leveraging only structure to model input data. We propose the use of a variational autoencoder (VAE) which utilizes data from an animal model to augment the training set and non-linear dimensionality reduction to map this data to human sets. This architecture utilizes variational inference, performed on latent parameters, to statistically model the probability distribution of training data in a latent feature space. We show the feasibility of VAEs, using images of mouse and human renal glomeruli from various pathological stages of diabetic nephropathy (DN), to model the progression of structural changes which occur in DN. When plotted in a 2-dimentional latent space, human and mouse glomeruli, show separation with some overlap, suggesting that the data is continuous, and can be statistically correlated. When DN stage is plotted in this latent space, trends in disease pathology are visualized.

AB - The adoption of deep learning techniques in medical applications has thus far been limited by the availability of the large labeled datasets required to robustly train neural networks, as well as difficulty interpreting these networks. However, recent techniques for unsupervised training of neural networks promise to address these issues, leveraging only structure to model input data. We propose the use of a variational autoencoder (VAE) which utilizes data from an animal model to augment the training set and non-linear dimensionality reduction to map this data to human sets. This architecture utilizes variational inference, performed on latent parameters, to statistically model the probability distribution of training data in a latent feature space. We show the feasibility of VAEs, using images of mouse and human renal glomeruli from various pathological stages of diabetic nephropathy (DN), to model the progression of structural changes which occur in DN. When plotted in a 2-dimentional latent space, human and mouse glomeruli, show separation with some overlap, suggesting that the data is continuous, and can be statistically correlated. When DN stage is plotted in this latent space, trends in disease pathology are visualized.

KW - Cross species modeling

KW - Knowledge transfer

KW - Unsupervised modeling

KW - Variational autoencoder

UR - http://www.scopus.com/inward/record.url?scp=85049189634&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85049189634&partnerID=8YFLogxK

U2 - 10.1117/12.2295456

DO - 10.1117/12.2295456

M3 - Conference contribution

AN - SCOPUS:85049189634

VL - 10581

BT - Medical Imaging 2018

PB - SPIE

ER -