Development and validation of a deep learning algorithm detecting 10 common abnormalities on chest radiographs

Ju Gang Nam, Minchul Kim, Jongchan Park, Eui Jin Hwang, Jong Hyuk Lee, Jung Hee Hong, Jin Mo Goo, Chang Min Park

Source: Eur Respir J, 57 (5) 2003061; 10.1183/13993003.03061-2020

Journal Issue: May

Abstract

We aimed to develop a deep learning algorithm detecting 10 common abnormalities (DLAD-10) on chest radiographs, and to evaluate its impact in diagnostic accuracy, timeliness of reporting and workflow efficacy.

DLAD-10 was trained with 146?717 radiographs from 108?053 patients using a ResNet34-based neural network with lesion-specific channels for 10 common radiological abnormalities (pneumothorax, mediastinal widening, pneumoperitoneum, nodule/mass, consolidation, pleural effusion, linear atelectasis, fibrosis, calcification and cardiomegaly). For external validation, the performance of DLAD-10 on a same-day computed tomography (CT)-confirmed dataset (normal:abnormal 53:147) and an open-source dataset (PadChest; normal:abnormal 339:334) was compared with that of three radiologists. Separate simulated reading tests were conducted on another dataset adjusted to real-world disease prevalence in the emergency department, consisting of four critical, 52 urgent and 146 nonurgent cases. Six radiologists participated in the simulated reading sessions with and without DLAD-10.

DLAD-10 exhibited area under the receiver operating characteristic curve values of 0.895–1.00 in the CT-confirmed dataset and 0.913–0.997 in the PadChest dataset. DLAD-10 correctly classified significantly more critical abnormalities (95.0% (57/60)) than pooled radiologists (84.4% (152/180); p=0.01). In simulated reading tests for emergency department patients, pooled readers detected significantly more critical (70.8% (17/24) versus 29.2% (7/24); p=0.006) and urgent (82.7% (258/312) versus 78.2% (244/312); p=0.04) abnormalities when aided by DLAD-10. DLAD-10 assistance shortened the mean±sd time-to-report critical and urgent radiographs (640.5±466.3 versus 3371.0±1352.5 s and 1840.3±1141.1 versus 2127.1±1468.2 s, respectively; all p<0.01) and reduced the mean±sd interpretation time (20.5±22.8 versus 23.5±23.7 s; p<0.001).

DLAD-10 showed excellent performance, improving radiologists' performance and shortening the reporting time for critical and urgent cases.

Rating:

You must login to grade this presentation.

Share or cite this content

Citations should be made in the following way:
Ju Gang Nam, Minchul Kim, Jongchan Park, Eui Jin Hwang, Jong Hyuk Lee, Jung Hee Hong, Jin Mo Goo, Chang Min Park. Development and validation of a deep learning algorithm detecting 10 common abnormalities on chest radiographs. Eur Respir J, 57 (5) 2003061; 10.1183/13993003.03061-2020

You must login to share this Presentation/Article on Twitter, Facebook, LinkedIn or by email.

Member's Comments

No comment yet.

You must Login to comment this presentation.

Related content which might interest you:

The frequency of the errors in the interpretation of digital radiographs of the chest by radiologists when detecting nodule and mass in the lungs. Source: Virtual Congress 2020 – New imaging techniques applied to old problems Year: 2020
A new method for evaluation of severity in COPD using dynamic chest x-ray examination Source: Annual Congress 2012 - Functional imaging in pulmonary oncology and COPD. Radiation dose in chest CT: survey and real life Year: 2012
Clinical utility of CT in children with persistent focal chest abnormality Source: Eur Respir J 2005; 26: 751 Year: 2005
Electromagnetic navigation – a new approach in diagnosing peripheral lung lesions: evaluation of the learning curve Source: Eur Respir J 2006; 28: Suppl. 50, 595s Year: 2006
Fully automatic detection and quantification of emphysema on computed tomography of the chest by a new software – comparison with lung functional criteria Source: Annual Congress 2008 - Imaging and measurement techniques in the evaluation of pulmonary embolism, COPD and pleural diseases Year: 2008
Development and validation of a low-cost chest wall motion assessment system Source: International Congress 2015 – Lung function: exploring the boundaries of the respiratory system Year: 2015
A robust lung segmentation algorithm using fuzzy C-means method from HRCT scans Source: International Congress 2016 – Imaging of COPD and airways: structural and functional assessments Year: 2016
The use of a simplified chest X-ray evaluation score for the assessment of tuberculosis Source: International Congress 2019 – Tuberculosis: from diagnosis to complications Year: 2019
The value of chest CT as a COVID-19 screening tool in children Source: Eur Respir J, 55 (6) 2001241; 10.1183/13993003.01241-2020 Year: 2020
Novel quantitative bronchiectasis scoring technique for chest computed tomography: BEST-CT. A study within the iABC project Source: International Congress 2019 – Imaging for prognostication and disease characterisation: 2019 update Year: 2019
Chest radiography: performance, indications and interpretation Source: Eur Respir Monogr 2015; 70: 34-46 Year: 2015
Computer-assisted detection of pulmonary embolism: performance evaluation in consensus with experienced and inexperienced chest radiologists Source: Eur Respir J 2007; 30: Suppl. 51, 253s Year: 2007
The validity of classic symptoms, medical history, and chest radiographs in predicting pulmonary tuberculosis (the TB scoring system) Source: Eur Respir J 2004; 24: Suppl. 48, 654s Year: 2004
A new method for detection of flow limitation in COPD using dynamic chest X-ray examination Source: Annual Congress 2011 - Morphological and functional imaging in obstructive airway disease Year: 2011
Deep learning architecture for the classification of COVID-19 and others pneumonias sources on lung CT imaging Source: Virtual Congress 2021 – Imaging Year: 2021
Late Breaking Abstract - Identifying and phenotyping COVID-19 patients using machine learning on chest x-rays Source: Virtual Congress 2020 – Covering COVID - the best abstracts Year: 2020
Diagnostic accuracy of COPD severity grading using machine learning features and lung sounds. Source: International Congress 2019 – Innovations in primary care assessment and management Year: 2019
Interobserver variability in visual evaluation of thoracic CT scans and comparison with automatic computer measurements of CT lung density Source: Annual Congress 2012 - Functional imaging in pulmonary oncology and COPD. Radiation dose in chest CT: survey and real life Year: 2012
Feasibility of low dose chest CT for virtual bronchoscopy navigation in a porcine model Source: International Congress 2019 – Interventional pulmonology techniques: targeting selective lung tissue Year: 2019
Diagnostic accuracy of asthma severity grading using machine learning features and lung sounds Source: International Congress 2018 – Innovations in equipment and their application Year: 2018