Extraction of radiographic findings from unstructured thoracoabdominal computed tomography reports using convolutional neural network based natural language processing.
<h4>Background</h4>Heart failure (HF) is a major cause of morbidity and mortality. However, much of the clinical data is unstructured in the form of radiology reports, while the process of data collection and curation is arduous and time-consuming.<h4>Purpose</h4>We utilized...
Main Authors: | , , , , , , , , , , , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
Public Library of Science (PLoS)
2020-01-01
|
Series: | PLoS ONE |
Online Access: | https://doi.org/10.1371/journal.pone.0236827 |
id |
doaj-c23fb8fb87a24e859f2068195955f9c1 |
---|---|
record_format |
Article |
spelling |
doaj-c23fb8fb87a24e859f2068195955f9c12021-04-23T04:30:30ZengPublic Library of Science (PLoS)PLoS ONE1932-62032020-01-01157e023682710.1371/journal.pone.0236827Extraction of radiographic findings from unstructured thoracoabdominal computed tomography reports using convolutional neural network based natural language processing.Mohit PandeyZhuoran XuEvan SholleGabriel MaliakalGurpreet SinghZahra FatimaDaria LarineBenjamin C LeeJing WangAlexander R van RosendaelLohendran BaskaranLeslee J ShawJames K MinSubhi J Al'Aref<h4>Background</h4>Heart failure (HF) is a major cause of morbidity and mortality. However, much of the clinical data is unstructured in the form of radiology reports, while the process of data collection and curation is arduous and time-consuming.<h4>Purpose</h4>We utilized a machine learning (ML)-based natural language processing (NLP) approach to extract clinical terms from unstructured radiology reports. Additionally, we investigate the prognostic value of the extracted data in predicting all-cause mortality (ACM) in HF patients.<h4>Materials and methods</h4>This observational cohort study utilized 122,025 thoracoabdominal computed tomography (CT) reports from 11,808 HF patients obtained between 2008 and 2018. 1,560 CT reports were manually annotated for the presence or absence of 14 radiographic findings, in addition to age and gender. Thereafter, a Convolutional Neural Network (CNN) was trained, validated and tested to determine the presence or absence of these features. Further, the ability of CNN to predict ACM was evaluated using Cox regression analysis on the extracted features.<h4>Results</h4>11,808 CT reports were analyzed from 11,808 patients (mean age 72.8 ± 14.8 years; 52.7% (6,217/11,808) male) from whom 3,107 died during the 10.6-year follow-up. The CNN demonstrated excellent accuracy for retrieval of the 14 radiographic findings with area-under-the-curve (AUC) ranging between 0.83-1.00 (F1 score 0.84-0.97). Cox model showed the time-dependent AUC for predicting ACM was 0.747 (95% confidence interval [CI] of 0.704-0.790) at 30 days.<h4>Conclusion</h4>An ML-based NLP approach to unstructured CT reports demonstrates excellent accuracy for the extraction of predetermined radiographic findings, and provides prognostic value in HF patients.https://doi.org/10.1371/journal.pone.0236827 |
collection |
DOAJ |
language |
English |
format |
Article |
sources |
DOAJ |
author |
Mohit Pandey Zhuoran Xu Evan Sholle Gabriel Maliakal Gurpreet Singh Zahra Fatima Daria Larine Benjamin C Lee Jing Wang Alexander R van Rosendael Lohendran Baskaran Leslee J Shaw James K Min Subhi J Al'Aref |
spellingShingle |
Mohit Pandey Zhuoran Xu Evan Sholle Gabriel Maliakal Gurpreet Singh Zahra Fatima Daria Larine Benjamin C Lee Jing Wang Alexander R van Rosendael Lohendran Baskaran Leslee J Shaw James K Min Subhi J Al'Aref Extraction of radiographic findings from unstructured thoracoabdominal computed tomography reports using convolutional neural network based natural language processing. PLoS ONE |
author_facet |
Mohit Pandey Zhuoran Xu Evan Sholle Gabriel Maliakal Gurpreet Singh Zahra Fatima Daria Larine Benjamin C Lee Jing Wang Alexander R van Rosendael Lohendran Baskaran Leslee J Shaw James K Min Subhi J Al'Aref |
author_sort |
Mohit Pandey |
title |
Extraction of radiographic findings from unstructured thoracoabdominal computed tomography reports using convolutional neural network based natural language processing. |
title_short |
Extraction of radiographic findings from unstructured thoracoabdominal computed tomography reports using convolutional neural network based natural language processing. |
title_full |
Extraction of radiographic findings from unstructured thoracoabdominal computed tomography reports using convolutional neural network based natural language processing. |
title_fullStr |
Extraction of radiographic findings from unstructured thoracoabdominal computed tomography reports using convolutional neural network based natural language processing. |
title_full_unstemmed |
Extraction of radiographic findings from unstructured thoracoabdominal computed tomography reports using convolutional neural network based natural language processing. |
title_sort |
extraction of radiographic findings from unstructured thoracoabdominal computed tomography reports using convolutional neural network based natural language processing. |
publisher |
Public Library of Science (PLoS) |
series |
PLoS ONE |
issn |
1932-6203 |
publishDate |
2020-01-01 |
description |
<h4>Background</h4>Heart failure (HF) is a major cause of morbidity and mortality. However, much of the clinical data is unstructured in the form of radiology reports, while the process of data collection and curation is arduous and time-consuming.<h4>Purpose</h4>We utilized a machine learning (ML)-based natural language processing (NLP) approach to extract clinical terms from unstructured radiology reports. Additionally, we investigate the prognostic value of the extracted data in predicting all-cause mortality (ACM) in HF patients.<h4>Materials and methods</h4>This observational cohort study utilized 122,025 thoracoabdominal computed tomography (CT) reports from 11,808 HF patients obtained between 2008 and 2018. 1,560 CT reports were manually annotated for the presence or absence of 14 radiographic findings, in addition to age and gender. Thereafter, a Convolutional Neural Network (CNN) was trained, validated and tested to determine the presence or absence of these features. Further, the ability of CNN to predict ACM was evaluated using Cox regression analysis on the extracted features.<h4>Results</h4>11,808 CT reports were analyzed from 11,808 patients (mean age 72.8 ± 14.8 years; 52.7% (6,217/11,808) male) from whom 3,107 died during the 10.6-year follow-up. The CNN demonstrated excellent accuracy for retrieval of the 14 radiographic findings with area-under-the-curve (AUC) ranging between 0.83-1.00 (F1 score 0.84-0.97). Cox model showed the time-dependent AUC for predicting ACM was 0.747 (95% confidence interval [CI] of 0.704-0.790) at 30 days.<h4>Conclusion</h4>An ML-based NLP approach to unstructured CT reports demonstrates excellent accuracy for the extraction of predetermined radiographic findings, and provides prognostic value in HF patients. |
url |
https://doi.org/10.1371/journal.pone.0236827 |
work_keys_str_mv |
AT mohitpandey extractionofradiographicfindingsfromunstructuredthoracoabdominalcomputedtomographyreportsusingconvolutionalneuralnetworkbasednaturallanguageprocessing AT zhuoranxu extractionofradiographicfindingsfromunstructuredthoracoabdominalcomputedtomographyreportsusingconvolutionalneuralnetworkbasednaturallanguageprocessing AT evansholle extractionofradiographicfindingsfromunstructuredthoracoabdominalcomputedtomographyreportsusingconvolutionalneuralnetworkbasednaturallanguageprocessing AT gabrielmaliakal extractionofradiographicfindingsfromunstructuredthoracoabdominalcomputedtomographyreportsusingconvolutionalneuralnetworkbasednaturallanguageprocessing AT gurpreetsingh extractionofradiographicfindingsfromunstructuredthoracoabdominalcomputedtomographyreportsusingconvolutionalneuralnetworkbasednaturallanguageprocessing AT zahrafatima extractionofradiographicfindingsfromunstructuredthoracoabdominalcomputedtomographyreportsusingconvolutionalneuralnetworkbasednaturallanguageprocessing AT darialarine extractionofradiographicfindingsfromunstructuredthoracoabdominalcomputedtomographyreportsusingconvolutionalneuralnetworkbasednaturallanguageprocessing AT benjaminclee extractionofradiographicfindingsfromunstructuredthoracoabdominalcomputedtomographyreportsusingconvolutionalneuralnetworkbasednaturallanguageprocessing AT jingwang extractionofradiographicfindingsfromunstructuredthoracoabdominalcomputedtomographyreportsusingconvolutionalneuralnetworkbasednaturallanguageprocessing AT alexanderrvanrosendael extractionofradiographicfindingsfromunstructuredthoracoabdominalcomputedtomographyreportsusingconvolutionalneuralnetworkbasednaturallanguageprocessing AT lohendranbaskaran extractionofradiographicfindingsfromunstructuredthoracoabdominalcomputedtomographyreportsusingconvolutionalneuralnetworkbasednaturallanguageprocessing AT lesleejshaw extractionofradiographicfindingsfromunstructuredthoracoabdominalcomputedtomographyreportsusingconvolutionalneuralnetworkbasednaturallanguageprocessing AT jameskmin extractionofradiographicfindingsfromunstructuredthoracoabdominalcomputedtomographyreportsusingconvolutionalneuralnetworkbasednaturallanguageprocessing AT subhijalaref extractionofradiographicfindingsfromunstructuredthoracoabdominalcomputedtomographyreportsusingconvolutionalneuralnetworkbasednaturallanguageprocessing |
_version_ |
1714662251776966656 |