Deep learning models for triaging hospital head MRI examinations

David A Wood; Sina Kafiabadi; Ayisha Al Busaidi; Emily Guilhem; Antanas Montvila; Jeremy Lynch; Matthew Townend; Siddharth Agarwal; Asif Mazumder; Gareth J Barker; Sebastien Ourselin; James H Cole; Thomas C Booth

doi:10.1016/j.media.2022.102391

Deep learning models for triaging hospital head MRI examinations

David A Wood, Sina Kafiabadi, Ayisha Al Busaidi, Emily Guilhem, Antanas Montvila, Jeremy Lynch, Matthew Townend, Siddharth Agarwal, Asif Mazumder, Gareth J Barker, Sebastien Ourselin, James H Cole, Thomas C Booth

Wrightington Wigan & Leigh NHS Fdn Trust, Prosser White Dermatol Ctr
GSTT Guy's and St Thomas' NHS Foundation Trust
Melbourne Dementia Research Centre
High Dimensional Neurology group, UCL Queen's Square Institute of Neurology
Univ London, Birkbeck University London, University College London, University of London Royal Veterinary College, University of London, St Georges University London, Imperial College London, Kings College London, Royal Holloway University London, Queen Mary University London, Div Populat Hlth Sci & Educ
King's College Hospital

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

25 Citations (Scopus)

308 Downloads (Pure)

Abstract

The growing demand for head magnetic resonance imaging (MRI) examinations, along with a global shortage of radiologists, has led to an increase in the time taken to report head MRI scans in recent years. For many neurological conditions, this delay can result in poorer patient outcomes and inflated healthcare costs. Potentially, computer vision models could help reduce reporting times for abnormal examinations by flagging abnormalities at the time of imaging, allowing radiology departments to prioritise limited resources into reporting these scans first. To date, however, the difficulty of obtaining large, clinically-representative labelled datasets has been a bottleneck to model development. In this work, we present a deep learning framework, based on convolutional neural networks, for detecting clinically-relevant abnormalities in minimally processed, hospital-grade axial T2-weighted and axial diffusion-weighted head MRI scans. The models were trained at scale using a Transformer-based neuroradiology report classifier to generate a labelled dataset of 70,206 examinations from two large UK hospital networks, and demonstrate fast (< 5 s), accurate (area under the receiver operating characteristic curve (AUC) > 0.9), and interpretable classification, with good generalisability between hospitals (ΔAUC ≤ 0.02). Through a simulation study we show that our best model would reduce the mean reporting time for abnormal examinations from 28 days to 14 days and from 9 days to 5 days at the two hospital networks, demonstrating feasibility for use in a clinical triage environment.

Original language	English
Article number	102391
Pages (from-to)	102391
Journal	Medical Image Analysis
Volume	78
Early online date	12 Feb 2022
DOIs	https://doi.org/10.1016/j.media.2022.102391
Publication status	Published - May 2022

Access to Document

10.1016/j.media.2022.102391

Wood_2021_MIA_final_cleanAccepted author manuscript, 4.81 MB
1-s2.0-S1361841522000433-main
1361-8415/© 2022 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/)
Final published version, 6.14 MBLicence: CC BY

Cite this

@article{484ef9a26ae4485d998eddd201df0f5b,

title = "Deep learning models for triaging hospital head MRI examinations",

abstract = "The growing demand for head magnetic resonance imaging (MRI) examinations, along with a global shortage of radiologists, has led to an increase in the time taken to report head MRI scans in recent years. For many neurological conditions, this delay can result in poorer patient outcomes and inflated healthcare costs. Potentially, computer vision models could help reduce reporting times for abnormal examinations by flagging abnormalities at the time of imaging, allowing radiology departments to prioritise limited resources into reporting these scans first. To date, however, the difficulty of obtaining large, clinically-representative labelled datasets has been a bottleneck to model development. In this work, we present a deep learning framework, based on convolutional neural networks, for detecting clinically-relevant abnormalities in minimally processed, hospital-grade axial T2-weighted and axial diffusion-weighted head MRI scans. The models were trained at scale using a Transformer-based neuroradiology report classifier to generate a labelled dataset of 70,206 examinations from two large UK hospital networks, and demonstrate fast (< 5 s), accurate (area under the receiver operating characteristic curve (AUC) > 0.9), and interpretable classification, with good generalisability between hospitals (ΔAUC ≤ 0.02). Through a simulation study we show that our best model would reduce the mean reporting time for abnormal examinations from 28 days to 14 days and from 9 days to 5 days at the two hospital networks, demonstrating feasibility for use in a clinical triage environment.",

author = "Wood, {David A} and Sina Kafiabadi and Busaidi, {Ayisha Al} and Emily Guilhem and Antanas Montvila and Jeremy Lynch and Matthew Townend and Siddharth Agarwal and Asif Mazumder and Barker, {Gareth J} and Sebastien Ourselin and Cole, {James H} and Booth, {Thomas C}",

note = "Funding Information: This work was supported by the Royal College of Radiologists , King's College Hospital Research and Innovation, King's Health Partners Challenge Fund, NVIDIA (through the unrestricted use of a GPU obtained in a competition), and the Wellcome/Engineering and Physical Sciences Research Council center for Medical Engineering (WT 203,148/Z/16/Z). Publisher Copyright: {\textcopyright} 2022",

year = "2022",

month = may,

doi = "10.1016/j.media.2022.102391",

language = "English",

volume = "78",

pages = "102391",

journal = "Medical Image Analysis",

issn = "1361-8415",

publisher = "Elsevier",

}

TY - JOUR

T1 - Deep learning models for triaging hospital head MRI examinations

AU - Wood, David A

AU - Kafiabadi, Sina

AU - Busaidi, Ayisha Al

AU - Guilhem, Emily

AU - Montvila, Antanas

AU - Lynch, Jeremy

AU - Townend, Matthew

AU - Agarwal, Siddharth

AU - Mazumder, Asif

AU - Barker, Gareth J

AU - Ourselin, Sebastien

AU - Cole, James H

AU - Booth, Thomas C

N1 - Funding Information: This work was supported by the Royal College of Radiologists , King's College Hospital Research and Innovation, King's Health Partners Challenge Fund, NVIDIA (through the unrestricted use of a GPU obtained in a competition), and the Wellcome/Engineering and Physical Sciences Research Council center for Medical Engineering (WT 203,148/Z/16/Z). Publisher Copyright: © 2022

PY - 2022/5

Y1 - 2022/5

N2 - The growing demand for head magnetic resonance imaging (MRI) examinations, along with a global shortage of radiologists, has led to an increase in the time taken to report head MRI scans in recent years. For many neurological conditions, this delay can result in poorer patient outcomes and inflated healthcare costs. Potentially, computer vision models could help reduce reporting times for abnormal examinations by flagging abnormalities at the time of imaging, allowing radiology departments to prioritise limited resources into reporting these scans first. To date, however, the difficulty of obtaining large, clinically-representative labelled datasets has been a bottleneck to model development. In this work, we present a deep learning framework, based on convolutional neural networks, for detecting clinically-relevant abnormalities in minimally processed, hospital-grade axial T2-weighted and axial diffusion-weighted head MRI scans. The models were trained at scale using a Transformer-based neuroradiology report classifier to generate a labelled dataset of 70,206 examinations from two large UK hospital networks, and demonstrate fast (< 5 s), accurate (area under the receiver operating characteristic curve (AUC) > 0.9), and interpretable classification, with good generalisability between hospitals (ΔAUC ≤ 0.02). Through a simulation study we show that our best model would reduce the mean reporting time for abnormal examinations from 28 days to 14 days and from 9 days to 5 days at the two hospital networks, demonstrating feasibility for use in a clinical triage environment.

AB - The growing demand for head magnetic resonance imaging (MRI) examinations, along with a global shortage of radiologists, has led to an increase in the time taken to report head MRI scans in recent years. For many neurological conditions, this delay can result in poorer patient outcomes and inflated healthcare costs. Potentially, computer vision models could help reduce reporting times for abnormal examinations by flagging abnormalities at the time of imaging, allowing radiology departments to prioritise limited resources into reporting these scans first. To date, however, the difficulty of obtaining large, clinically-representative labelled datasets has been a bottleneck to model development. In this work, we present a deep learning framework, based on convolutional neural networks, for detecting clinically-relevant abnormalities in minimally processed, hospital-grade axial T2-weighted and axial diffusion-weighted head MRI scans. The models were trained at scale using a Transformer-based neuroradiology report classifier to generate a labelled dataset of 70,206 examinations from two large UK hospital networks, and demonstrate fast (< 5 s), accurate (area under the receiver operating characteristic curve (AUC) > 0.9), and interpretable classification, with good generalisability between hospitals (ΔAUC ≤ 0.02). Through a simulation study we show that our best model would reduce the mean reporting time for abnormal examinations from 28 days to 14 days and from 9 days to 5 days at the two hospital networks, demonstrating feasibility for use in a clinical triage environment.

UR - http://www.scopus.com/inward/record.url?scp=85124685453&partnerID=8YFLogxK

U2 - 10.1016/j.media.2022.102391

DO - 10.1016/j.media.2022.102391

M3 - Article

C2 - 35183876

SN - 1361-8415

VL - 78

SP - 102391

JO - Medical Image Analysis

JF - Medical Image Analysis

M1 - 102391

ER -

Deep learning models for triaging hospital head MRI examinations

Abstract

Access to Document

Other files and links

Fingerprint

Cite this