Radiology report generation using multi–layer visual representation

Chenyu Wang; Vladimir Janjic; Stephen McKenna

doi:10.3389/978-2-8325-1231-9

Radiology report generation using multi–layer visual representation

Computing

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

157 Downloads (Pure)

Abstract

Chest X-ray images are crucial for diagnostics and treatment of various diseases, but analysing them requires highly skilled and experienced medical professionals. This leads to a high burden on radiologists. This paper investigates a method for automatic generation of radiology reports and incorporates visual features extracted from multiple layers of a convolutional visual encoder network. The use of multi-level features can provide more information to the text generator, potentially improving generated reports. Experimental results are reported on the MIMIC-CXR dataset using natural language generation metrics.

Original language	English
Title of host publication	27th Conference on Medical Image Understanding and Analysis 2023
Editors	Gordon Waiter, Tryphon Lambrou, Georgios Leontidis, Nir Oren, Teresa Morris, Sharon Gordon, Jade McGowan, Cara Nicolson
Publisher	Frontiers Media S.A.
Pages	251-256
Number of pages	6
ISBN (Print)	9782832512319
DOIs	https://doi.org/10.3389/978-2-8325-1231-9
Publication status	Published - 2023

Access to Document

10.3389/978-2-8325-1231-9Licence: CC BY

Final Published VersionFinal published version, 421 KBLicence: CC BY

Cite this

@inproceedings{9b53acf6fc434ca98be400555df7b77b,

title = "Radiology report generation using multi{\textendash}layer visual representation",

abstract = "Chest X-ray images are crucial for diagnostics and treatment of various diseases, but analysing them requires highly skilled and experienced medical professionals. This leads to a high burden on radiologists. This paper investigates a method for automatic generation of radiology reports and incorporates visual features extracted from multiple layers of a convolutional visual encoder network. The use of multi-level features can provide more information to the text generator, potentially improving generated reports. Experimental results are reported on the MIMIC-CXR dataset using natural language generation metrics.",

author = "Chenyu Wang and Vladimir Janjic and Stephen McKenna",

note = "Copyright: {\textcopyright} 2023 The Author(s).",

year = "2023",

doi = "10.3389/978-2-8325-1231-9",

language = "English",

isbn = "9782832512319",

pages = "251--256",

editor = "Gordon Waiter and Tryphon Lambrou and Georgios Leontidis and Nir Oren and Teresa Morris and Sharon Gordon and Jade McGowan and Cara Nicolson",

booktitle = "27th Conference on Medical Image Understanding and Analysis 2023",

publisher = "Frontiers Media S.A.",

address = "Switzerland",

}

Radiology report generation using multi–layer visual representation. / Wang, Chenyu; Janjic, Vladimir ; McKenna, Stephen.
27th Conference on Medical Image Understanding and Analysis 2023. ed. / Gordon Waiter; Tryphon Lambrou; Georgios Leontidis; Nir Oren; Teresa Morris; Sharon Gordon; Jade McGowan; Cara Nicolson. Frontiers Media S.A., 2023. p. 251-256.

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - Radiology report generation using multi–layer visual representation

AU - Wang, Chenyu

AU - Janjic, Vladimir

AU - McKenna, Stephen

PY - 2023

Y1 - 2023

N2 - Chest X-ray images are crucial for diagnostics and treatment of various diseases, but analysing them requires highly skilled and experienced medical professionals. This leads to a high burden on radiologists. This paper investigates a method for automatic generation of radiology reports and incorporates visual features extracted from multiple layers of a convolutional visual encoder network. The use of multi-level features can provide more information to the text generator, potentially improving generated reports. Experimental results are reported on the MIMIC-CXR dataset using natural language generation metrics.

AB - Chest X-ray images are crucial for diagnostics and treatment of various diseases, but analysing them requires highly skilled and experienced medical professionals. This leads to a high burden on radiologists. This paper investigates a method for automatic generation of radiology reports and incorporates visual features extracted from multiple layers of a convolutional visual encoder network. The use of multi-level features can provide more information to the text generator, potentially improving generated reports. Experimental results are reported on the MIMIC-CXR dataset using natural language generation metrics.

U2 - 10.3389/978-2-8325-1231-9

DO - 10.3389/978-2-8325-1231-9

M3 - Conference contribution

SN - 9782832512319

SP - 251

EP - 256

BT - 27th Conference on Medical Image Understanding and Analysis 2023

A2 - Waiter, Gordon

A2 - Lambrou, Tryphon

A2 - Leontidis, Georgios

A2 - Oren, Nir

A2 - Morris, Teresa

A2 - Gordon, Sharon

A2 - McGowan, Jade

A2 - Nicolson, Cara

PB - Frontiers Media S.A.

ER -

Radiology report generation using multi–layer visual representation

Abstract

Access to Document

Fingerprint

Cite this