Evaluating a Fully Automated Pulmonary Nodule Detection Approach and                     Its Impact on Radiologist Performance

Li, Kai; Li, Qiong; Ma, Jiechao; Zhou, Zijian; Sun, Mengmeng; Deng, Yufeng; Tu, Wenting; Wang, Yun; Fan, Li; Chen, Xia; Xiao, Yi; Zhang, Rongguo; Liu, Shiyuan

doi:10.1148/ryai.2019180084

Cited by 73 publications

(61 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Deep learning has demonstrated superior performance in the field of radiology [3,14,15]. Previous studies have successfully applied deep learning techniques to detect pneumonia in chest images [14,17] and to further differentiate viral and other types of pneumonia in chest radiographs or CT imaging [15,18,19].…”

Section: Discussionmentioning

confidence: 99%

“…Artificial intelligence (AI) using deep learning technology has demonstrated great success in the field of medical imaging due to its prominent capability of feature extraction [14,15]. A recent study [15] showed that a machine learning approach using the convolutional neural networks model was able to distinguish COVID-19 from community-acquired pneumonia (CAP).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

COVID-19 Pneumonia Severity Grading: Test of A Trained Deep Learning Model

Zhang

Song³

et al. 2020

Preprint

View full text Add to dashboard Cite

Purpose: To detect affected lung lobes and conduct severity grading of coronavirus disease 2019 (COVID-19) pneumonia on chest CT images using an artificial intelligence (AI) technique.Materials and Methods: We used a deep learning model which was previously developed and trained to extract visual features from chest CT exams for the detection and severity grading of COVID-19 pneumonia. In this retrospective study, we tested this model with COVID-19 pneumonia cases in our institution. The numbers of affected lung lobes and severity grading values were compared for the AI method and manual method via the paired Chi-square test. The severity grading capability of the AI method was evaluated using receiver operating characteristic analysis.Results: A total of 24 cases of confirmed COVID-19 were included (13 men and 11 women). The most frequent CT observation was bilateral ground-glass opacities with consolidation and more than one affected lung lobe. Most cases were mild cases. Compared with the manual method, the AI method presented excellent sensitivity (97.2%) and accuracy (80.8%) but poor specificity (57.1%) in detecting affected lung lobes and good ability (area under the curve=0.795, accuracy=91.6%) in severity grading of COVID-19. Additionally, the time consumed in checking the accuracy of the AI detected lesions within the whole lung was significantly shorter than that of severity assessment by the manual method (t=9.434, p<0.001).Conclusion: The AI method with our model is useful in evaluating the severity grading of COVID-19 pneumonia.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

COVID-19 Pneumonia Severity Grading: Test of A Trained Deep Learning Model

Zhang

Song³

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…Indeed, the AI algorithm already achieved excellent performance in the detection of active tuberculosis on chest X-ray [19]. Detection of active tuberculosis on CT is still a challenging task, and CT achieved decent performance in the detection of a pulmonary nodule [20]. Although we did not use AI algorithms to automatize and objectivize to read the CT images, the AI algorithm can potentially be used to read CT images of CM-predominant TB in the near future.…”

Section: Plos Onementioning

confidence: 99%

Clustered micronodules as predominant manifestation on CT: A sign of active but indolently evolving pulmonary tuberculosis

et al. 2020

View full text Add to dashboard Cite

To investigate the prevalence, patient characteristics, and natural history of clustered micronodules (CMs) in active pulmonary tuberculosis. Materials and methods From January 2013 through July 2018, 833 consecutive patients with bacteriologically or polymerase chain reaction-proven active pulmonary tuberculosis were retrospectively evaluated. CMs were defined as a localized aggregation of multiple dense discrete micronodules, which primarily distributed around small airways distal to the level of the segmental bronchus: small airways surrounded by CMs maintained luminal patency and the CMs might coalesce into a larger nodule. The patients were dichotomized according to whether the predominant computed tomography (CT) abnormalities were CMs. We analyzed radiologic and pathologic findings in patients whose predominant diagnostic CT abnormalities were CMs, along with those of incidental pre-diagnostic CT scans, if available. Chi-square, McNemar, Student t-test and Wilcoxon-signed rank test were performed. Results CMs were the predominant CT abnormality in 2.6% of the patients (22/833, 95% CI, 1.8-4.0%) with less sputum smear-positivity (4.8% vs 31.0%; p = .010) and a similar proportion of immunocompromised status (40.9% vs 46.0%; p = .637) than those without having CMs as the predominant CT abnormality. The time interval for minimal radiologic progression was 6.4 months. The extent of CMs increased with disease progression, frequently accompanied by consolidation and small airway wall thickening. Pathologically, smaller CMs were non-caseating granulomas confined to the peribronchiolar interstitium, whereas larger CMs were caseating granulomas involving lung parenchyma. Two of the five patients with a prediagnostic CT scan obtained more than 50 months pre-diagnosis showed an incipient stage of CMs, in which they were small peribronchiolar nodules.

show abstract

“…Eine Herausforderung für die künstliche Intelligenz stellen multiple Herde in der Lungenkrebs-Diagnostik dar. In einer kleineren Arbeit an 53 Patienten mit 108 Läsionen ließ sich aber immerhin eine Übereinstimmung zwischen künstlicher Intelligenz und Pathologie von 89 % erreichen [50]. Solide Läsionen und größere Herde wurden durch KI am häufigsten übersehen.…”

Section: Datensätze Zum Maschinellen Lernenunclassified

Expertentreffen COPD: Technologische Innovationen in der Pneumologie – Facetten aus Diagnostik und Therapie

et al. 2020

View full text Add to dashboard Cite

ZusammenfassungEin wesentlicher Anteil der aktuellen technologischen Entwicklungen in der Pneumologie liegt in den verschiedenen Bereichen der Informationstechnologie. Das Spektrum reicht dabei von Smartphone-Apps, die im täglichen Leben oder der Praxis von Patienten oder Ärzten angewandt werden sollen, bis hin zum Einsatz der künstlichen Intelligenz in der Früherkennung. Die Diagnose-Genauigkeit von Apps zur Symptomanalyse ist dabei zurzeit noch sehr limitiert. Forschungsprojekte beschäftigen sich mit der Integration von Symptomen und Funktionsparametern in der Früherkennung, aber auch mit der Mobilitätserfassung als prognostischem Marker bei der COPD. Eine große Herausforderung stellt das Lungenkrebs-Screening mittels Computertomografie dar. Hier kann künstliche Intelligenz helfen, riesige Datenmengen zu bewältigen. Die Qualität hängt jedoch vom suffizienten Training der Systeme ab. Technologische Entwicklungen prägen alle Felder der Pneumologie. Sie erlauben in der diagnostischen und interventionellen Endoskopie die verbesserte Biopsietechnik und mikrostrukturelle Bildgebung. Methoden der Lungenfunktionsdiagnostik ermöglichen die differenzierte Analyse von atemmechanischen Störungen und können in die Beatmungstechnologie überführt werden. Die Translation von Grundlagenerkenntnissen zum Mikrobiom kann perspektivisch helfen, COPD-Exazerbationen besser zu verstehen und zielgerichteter zu behandeln.

show abstract

Evaluating a Fully Automated Pulmonary Nodule Detection Approach and Its Impact on Radiologist Performance

Cited by 73 publications

References 27 publications

COVID-19 Pneumonia Severity Grading: Test of A Trained Deep Learning Model

COVID-19 Pneumonia Severity Grading: Test of A Trained Deep Learning Model

Clustered micronodules as predominant manifestation on CT: A sign of active but indolently evolving pulmonary tuberculosis

Expertentreffen COPD: Technologische Innovationen in der Pneumologie – Facetten aus Diagnostik und Therapie

Contact Info

Product

Resources

About