PocketNet: A Smaller Neural Network for Medical Image Analysis

Adrian Celaya; Jonas A. Actor; Rajarajesawari Muthusivarajan; Evan Gates; Caroline Chung; Dawid Schellingerhout; Beatrice Riviere; David Fuentes

doi:10.1109/TMI.2022.3224873

PocketNet: A Smaller Neural Network for Medical Image Analysis

Adrian Celaya, Jonas A. Actor, Rajarajesawari Muthusivarajan, Evan Gates, Caroline Chung, Dawid Schellingerhout, Beatrice Riviere, David Fuentes

Research output: Contribution to journal › Article › peer-review

1 Scopus citations

Abstract

Medical imaging deep learning models are often large and complex, requiring specialized hardware to train and evaluate these models. To address such issues, we propose the PocketNet paradigm to reduce the size of deep learning models by throttling the growth of the number of channels in convolutional neural networks. We demonstrate that, for a range of segmentation and classification tasks, PocketNet architectures produce results comparable to that of conventional neural networks while reducing the number of parameters by multiple orders of magnitude, using up to 90% less GPU memory, and speeding up training times by up to 40%, thereby allowing such models to be trained and deployed in resource-constrained settings.

Original language	English (US)
Pages (from-to)	1172-1184
Number of pages	13
Journal	IEEE Transactions on Medical Imaging
Volume	42
Issue number	4
DOIs	https://doi.org/10.1109/TMI.2022.3224873
State	Published - Apr 1 2023

Keywords

Neural network
pattern recognition and classification
segmentation

ASJC Scopus subject areas

Software
Radiological and Ultrasound Technology
Computer Science Applications
Electrical and Electronic Engineering

Access to Document

10.1109/TMI.2022.3224873

Cite this

@article{f90eacd9aab34d849245683b89f76d7a,

title = "PocketNet: A Smaller Neural Network for Medical Image Analysis",

abstract = "Medical imaging deep learning models are often large and complex, requiring specialized hardware to train and evaluate these models. To address such issues, we propose the PocketNet paradigm to reduce the size of deep learning models by throttling the growth of the number of channels in convolutional neural networks. We demonstrate that, for a range of segmentation and classification tasks, PocketNet architectures produce results comparable to that of conventional neural networks while reducing the number of parameters by multiple orders of magnitude, using up to 90% less GPU memory, and speeding up training times by up to 40%, thereby allowing such models to be trained and deployed in resource-constrained settings.",

keywords = "Neural network, pattern recognition and classification, segmentation",

author = "Adrian Celaya and Actor, {Jonas A.} and Rajarajesawari Muthusivarajan and Evan Gates and Caroline Chung and Dawid Schellingerhout and Beatrice Riviere and David Fuentes",

note = "Publisher Copyright: {\textcopyright} 1982-2012 IEEE.",

year = "2023",

month = apr,

day = "1",

doi = "10.1109/TMI.2022.3224873",

language = "English (US)",

volume = "42",

pages = "1172--1184",

journal = "IEEE Transactions on Medical Imaging",

issn = "0278-0062",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

number = "4",

}

TY - JOUR

T1 - PocketNet

T2 - A Smaller Neural Network for Medical Image Analysis

AU - Celaya, Adrian

AU - Actor, Jonas A.

AU - Muthusivarajan, Rajarajesawari

AU - Gates, Evan

AU - Chung, Caroline

AU - Schellingerhout, Dawid

AU - Riviere, Beatrice

AU - Fuentes, David

PY - 2023/4/1

Y1 - 2023/4/1

N2 - Medical imaging deep learning models are often large and complex, requiring specialized hardware to train and evaluate these models. To address such issues, we propose the PocketNet paradigm to reduce the size of deep learning models by throttling the growth of the number of channels in convolutional neural networks. We demonstrate that, for a range of segmentation and classification tasks, PocketNet architectures produce results comparable to that of conventional neural networks while reducing the number of parameters by multiple orders of magnitude, using up to 90% less GPU memory, and speeding up training times by up to 40%, thereby allowing such models to be trained and deployed in resource-constrained settings.

AB - Medical imaging deep learning models are often large and complex, requiring specialized hardware to train and evaluate these models. To address such issues, we propose the PocketNet paradigm to reduce the size of deep learning models by throttling the growth of the number of channels in convolutional neural networks. We demonstrate that, for a range of segmentation and classification tasks, PocketNet architectures produce results comparable to that of conventional neural networks while reducing the number of parameters by multiple orders of magnitude, using up to 90% less GPU memory, and speeding up training times by up to 40%, thereby allowing such models to be trained and deployed in resource-constrained settings.

KW - Neural network

KW - pattern recognition and classification

KW - segmentation

UR - http://www.scopus.com/inward/record.url?scp=85144048528&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85144048528&partnerID=8YFLogxK

U2 - 10.1109/TMI.2022.3224873

DO - 10.1109/TMI.2022.3224873

M3 - Article

C2 - 36427285

AN - SCOPUS:85144048528

SN - 0278-0062

VL - 42

SP - 1172

EP - 1184

JO - IEEE Transactions on Medical Imaging

JF - IEEE Transactions on Medical Imaging

IS - 4

ER -

PocketNet: A Smaller Neural Network for Medical Image Analysis

Abstract

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this