Development of a convolutional neural network to differentiate among the etiology of similar appearing pathological B lines on lung ultrasound: a deep learning study

Arntfield, Robert; VanBerlo, Blake; Alaifan, Thamer; Phelps, Nathan; White, Matthew; Chaudhary, Rushil; Ho, Jordan; Wu, Daniel

doi:10.1136/bmjopen-2020-045120

Cited by 50 publications

(42 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These studies focused on either detecting B-lines, namely artefacts appearing when patients suffer from pneumonia, or binary classifications of LUS frames into Covid-19 and non-Covid-19[ 15 , 31 ]. Moreover, only a few researchers have exploited data from reliable hospital sources[ [32] , [33] , [34] ], indicating the lack of a reliable dataset; several authors have described the inconsistent quality of their data and the need to rely on non-validated sources as limitations of their studies[ 35 ]. In addition, some researchers have worked with LUS from only one particular type of probe, thus lacking heterogeneous data to train the neural networks, posing another limitation on the soundness of their conclusions and DL algorithm usage[ 33 ].…”

Section: Introductionmentioning

confidence: 99%

“…Moreover, only a few researchers have exploited data from reliable hospital sources[ [32] , [33] , [34] ], indicating the lack of a reliable dataset; several authors have described the inconsistent quality of their data and the need to rely on non-validated sources as limitations of their studies[ 35 ]. In addition, some researchers have worked with LUS from only one particular type of probe, thus lacking heterogeneous data to train the neural networks, posing another limitation on the soundness of their conclusions and DL algorithm usage[ 33 ]. Only two studies have focused on DL systems for the purpose of detecting Covid-19 pneumonia and assessing the severity of lung engagement[ 32 , 34 ].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Deep learning and lung ultrasound for Covid-19 pneumonia detection and severity classification

Salvia

Secco

Torti

et al. 2021

Computers in Biology and Medicine

View full text Add to dashboard Cite

The Covid-19 European outbreak in February 2020 has challenged the world's health systems, eliciting an urgent need for effective and highly reliable diagnostic instruments to help medical personnel. Deep learning (DL) has been demonstrated to be useful for diagnosis using both computed tomography (CT) scans and chest X-rays (CXR), whereby the former typically yields more accurate results. However, the pivoting function of a CT scan during the pandemic presents several drawbacks, including high cost and cross-contamination problems. Radiation-free lung ultrasound (LUS) imaging, which requires high expertise and is thus being underutilised, has demonstrated a strong correlation with CT scan results and a high reliability in pneumonia detection even in the early stages. In this study, we developed a system based on modern DL methodologies in close collaboration with Fondazione IRCCS Policlinico San Matteo's Emergency Department (ED) of Pavia. Using a reliable dataset comprising ultrasound clips originating from linear and convex probes in 2908 frames from 450 hospitalised patients, we conducted an investigation into detecting Covid-19 patterns and ranking them considering two severity scales. This study differs from other research projects by its novel approach involving four and seven classes. Patients admitted to the ED underwent 12 LUS examinations in different chest parts, each evaluated according to standardised severity scales. We adopted residual convolutional neural networks (CNNs), transfer learning, and data augmentation techniques. Hence, employing methodological hyperparameter tuning, we produced state-of-the-art results meeting F1 score levels, averaged over the number of classes considered, exceeding 98%, and thereby manifesting stable measurements over precision and recall.

show abstract

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Deep learning and lung ultrasound for Covid-19 pneumonia detection and severity classification

Salvia

Secco

Torti

et al. 2021

Computers in Biology and Medicine

View full text Add to dashboard Cite

show abstract

“…Notably, LUS sensitivity was found to be higher than that of CXR for COVID-19 diagnosis, 76 and some even found comparable diagnostic accuracy to CT. 77,78 However, the role of LUS for the COVID-19 pandemic is still actively debated [79][80][81] and, regarding AI, with only one publicly available dataset, 26 more research is needed to narrow down the practical role of AI on LUS. 26,50,82,83 Additionally, studies using ML on multiple imaging modalities from the same cohort are certainly needed to shed light on comparative questions between modalities from the perspective of ML. The performance of AI-assisted radiologists in detecting COVID-19 might or might not confirm the current radiologic findings, for example that CXR is less sensitive than CT 84 and LUS (when compared with RT-PCR 76 or CT 85 ) or that B-lines are the most reliable pathological pattern across CT, CXR, and LUS.…”

Section: Imaging Modality Rivalrymentioning

confidence: 99%

On the role of artificial intelligence in medical imaging of COVID-19

Born¹,

Beymer²,

Rajan³

et al. 2021

Patterns

View full text Add to dashboard Cite

Although a plethora of research articles on AI methods on COVID-19 medical imaging are published, their clinical value remains unclear. We conducted the largest systematic review of the literature addressing the utility of AI in imaging for COVID-19 patient care. By keyword searches on PubMed and preprint servers throughout 2020, we identified 463 manuscripts and performed a systematic meta-analysis to assess their technical merit and clinical relevance. Our analysis evidences a significant disparity between clinical and AI communities, in the focus on both imaging modalities (AI experts neglected CT and ultrasound, favoring X-ray) and performed tasks (71.9% of AI papers centered on diagnosis). The vast majority of manuscripts were found to be deficient regarding potential use in clinical practice, but 2.7% (n = 12) publications were assigned a high maturity level and are summarized in greater detail. We provide an itemized discussion of the challenges in developing clinically relevant AI solutions with recommendations and remedies.THE BIGGER PICTURE During the COVID-19 pandemic, medical imaging (CT, X-ray, ultrasound) has played a key role in addressing the magnified need for speed, low cost, ubiquity, and precision in patient care. The contemporary digitization of medicine and rise of artificial intelligence (AI) induce a quantum leap in medical imaging: AI has proven equipollent to healthcare professionals across a diverse range of tasks, and hopes are high that AI can save time and cost and increase coverage by advancing rapid patient stratification and empowering clinicians. This review bridges medical imaging and AI in the context of COVID-19 and conducts the largest systematic review of the literature in the field. We identify several gaps and evidence significant disparities between clinicians and AI experts and foresee a need for improved, interdisciplinary collaboration to develop robust AI solutions that can be deployed in clinical practice.The key challenges on that roadmap are discussed alongside recommended solutions. ll

show abstract

“…Although LUS is a rapid, bedside, goal-oriented, diagnostic test, the IRR variation is a common problem in point of care ultrasound. To overcome this problem, we believe it is very important to perform a study on the automatic ultrasound judgement such as deep learning [44,45].…”

Section: Discussionmentioning

confidence: 99%

Do Serial Lung Ultrasound Scores Predict Prolonged Mechanical Ventilation in Patients With Severe COVID-19? A Single-centre Retrospective Cohort Study

Taniguchi

Ohya

Yamagata

et al. 2021

Preprint

View full text Add to dashboard Cite

Background: Some patients with severe coronavirus disease (COVID-19) who present with fibrosis on computed tomography (CT) require prolonged mechanical ventilation (PMV). Lung ultrasound (LUS), a rapid, bedside test, has been reported to have findings consistent with those of CT. Thus, this study aimed to assess whether serial LUS scores could predict PMV or successful extubation in severe COVID-19 patients.Methods: LUS was performed for 20 consecutive severe COVID-19 patients at three time points: admission (day 1), after 48 h (day 3), and seventh-day follow-up (day 7). We compared the LUS score with the results of chest X-rays and laboratory tests at three time points. Moreover, we assessed LUS score to determine the inter-rater reliability (IRR) of the results among examiners.Results: While there were no significant differences in mortality in each PMV and successful extubation groups, there were significant differences in LUS scores on day 3 and day 7; XP score on day 7; and P/F ratio on day 7 in the PMV group (p<0.05). There were significant differences in LUS scores on days 3 and 7, C-reactive protein (CRP) levels on day 7, and P/F ratio on day 7 in the successful extubation group (p<0.05). The areas under the curves (AUCs) of LUS score on days 3 and 7, XP score on day 7, and P/F ratio were 0.88, 0.98, 0.77, and 0.80, respectively in the PMV group; and the AUCs of LUS score on days 3 and 7, CRP levels on day 7, and P/F ratio 0.79, 0.90, 0.82, and 0.79, respectively, in the successful extubation group. Variations in serial LUS scores exhibited significant differences between the groups. The serial LUS score on day 7 was higher than that on day 1 in the PMV group but lower in the successful extubation group (p<0.05). However, there was slight IRR agreement in the LUS score changes on days 1 to 7 (κ= 0.15 [95% CI: 0-0.31]). Conclusions: The serial LUS score of severe COVID-19 patients could predicted PMV and successful extubation. To overcome IRR disagreement, the automatic ultrasound judgement, such as deep learning, would be needed.

show abstract

Development of a convolutional neural network to differentiate among the etiology of similar appearing pathological B lines on lung ultrasound: a deep learning study

Cited by 50 publications

References 29 publications

Deep learning and lung ultrasound for Covid-19 pneumonia detection and severity classification

Deep learning and lung ultrasound for Covid-19 pneumonia detection and severity classification

On the role of artificial intelligence in medical imaging of COVID-19

Do Serial Lung Ultrasound Scores Predict Prolonged Mechanical Ventilation in Patients With Severe COVID-19? A Single-centre Retrospective Cohort Study

Contact Info

Product

Resources

About