Federated deep learning for detecting COVID-19 lung abnormalities in CT: a privacy-preserving multinational validation study

npj Digital Medicine, Год журнала: 2021, Номер 4(1)

Опубликована: Март 29, 2021

Data privacy mechanisms are essential for rapidly scaling medical training databases to capture the heterogeneity of patient data distributions toward robust and generalizable machine learning systems. In current COVID-19 pandemic, a major focus artificial intelligence (AI) is interpreting chest CT, which can be readily used in assessment management disease. This paper demonstrates feasibility federated method detecting related CT abnormalities with external validation on patients from multinational study. We recruited 132 seven different centers, three internal hospitals Hong Kong testing, four external, independent datasets Mainland China Germany, validating model generalizability. also conducted case studies longitudinal scans automated estimation lesion burden hospitalized patients. explore algorithms develop privacy-preserving AI image diagnosis good generalization capability unseen datasets. Federated could provide an effective mechanism during pandemics clinically useful across institutions countries overcoming central aggregation large amounts sensitive data.

Язык: Английский

---

Weakly Supervised Segmentation of COVID19 Infection with Scribble Annotation on CT Images

Pattern Recognition, Год журнала: 2021, Номер 122, С. 108341 - 108341

Опубликована: Сен. 20, 2021

Язык: Английский

---

Learning with limited annotations: A survey on deep semi-supervised learning for medical image segmentation

Computers in Biology and Medicine, Год журнала: 2023, Номер 169, С. 107840 - 107840

Опубликована: Дек. 16, 2023

Язык: Английский

---

Label-Free Segmentation of COVID-19 Lesions in Lung CT

IEEE Transactions on Medical Imaging, Год журнала: 2021, Номер 40(10), С. 2808 - 2819

Опубликована: Март 24, 2021

Scarcity of annotated images hampers the building automated solution for reliable COVID-19 diagnosis and evaluation from CT. To alleviate burden data annotation, we herein present a label-free approach segmenting lesions in CT via voxel-level anomaly modeling that mines out relevant knowledge normal lung scans. Our is inspired by observation parts tracheae vessels, which lay high-intensity range where belong to, exhibit strong patterns. facilitate learning such patterns at voxel level, synthesize 'lesions' using set simple operations insert synthesized into scans to form training pairs, learn normalcy-recognizing network (NormNet) recognizes tissues separate them possible lesions. experiments on three different public datasets validate effectiveness NormNet, conspicuously outperforms variety unsupervised detection (UAD) methods.

Язык: Английский

---

COVID-Net CT-2: Enhanced Deep Neural Networks for Detection of COVID-19 From Chest CT Images Through Bigger, More Diverse Learning

Frontiers in Medicine, Год журнала: 2022, Номер 8

Опубликована: Март 10, 2022

The COVID-19 pandemic continues to rage on, with multiple waves causing substantial harm health and economies around the world. Motivated by use of computed tomography (CT) imaging at clinical institutes world as an effective complementary screening method RT-PCR testing, we introduced COVID-Net CT, a deep neural network tailored for detection cases from chest CT images, along large curated benchmark dataset comprising 1,489 patient part open-source initiative. However, one potential limiting factor is restricted data quantity diversity given single nation cohort used in study. To address this limitation, study introduce enhanced networks images which are trained using large, diverse, multinational cohort. We accomplish through introduction two new datasets, largest comprises 4,501 patients least 16 countries. best our knowledge, represents largest, most diverse open-access form. Additionally, novel lightweight architecture called S, significantly smaller faster than previously architecture. leverage explainability investigate decision-making behavior models ensure that decisions based on relevant indicators, results select reviewed reported board-certified radiologists over 10 30 years experience, respectively. best-performing achieved accuracy, sensitivity, positive predictive value, specificity, negative value 99.0%/99.1%/98.0%/99.4%/99.7%, Moreover, explainability-driven performance validation shows consistency radiologist interpretation leveraging correct, clinically critical factors. promising suggest strong tool computer-aided assessment. While not production-ready solution, hope open-source, release CT-2 associated datasets will continue enable researchers, clinicians, citizen scientists alike build upon them.

Язык: Английский

---

PDAtt-Unet: Pyramid Dual-Decoder Attention Unet for Covid-19 infection segmentation from CT-scans

Medical Image Analysis, Год журнала: 2023, Номер 86, С. 102797 - 102797

Опубликована: Март 21, 2023

Since the emergence of Covid-19 pandemic in late 2019, medical imaging has been widely used to analyze this disease. Indeed, CT-scans lungs can help diagnose, detect, and quantify infection. In paper, we address segmentation infection from CT-scans. To improve performance Att-Unet architecture maximize use Attention Gate, propose PAtt-Unet DAtt-Unet architectures. aims exploit input pyramids preserve spatial awareness all encoder layers. On other hand, is designed guide inside lung lobes. We also combine these two architectures into a single one, which refer as PDAtt-Unet. overcome blurry boundary pixels infection, hybrid loss function. The proposed were tested on four datasets with evaluation scenarios (intra cross datasets). Experimental results showed that both segmenting infections. Moreover, combination PDAtt-Unet led further improvement. Compare methods, three baseline (Unet, Unet++, Att-Unet) state-of-the-art (InfNet, SCOATNet, nCoVSegNet) tested. comparison superiority trained (PDEAtt-Unet) over methods. PDEAtt-Unet able various challenges infections scenarios.

Язык: Английский

---

LDANet: Automatic lung parenchyma segmentation from CT images

Computers in Biology and Medicine, Год журнала: 2023, Номер 155, С. 106659 - 106659

Опубликована: Фев. 10, 2023

Язык: Английский

---

Review on the Evaluation and Development of Artificial Intelligence for COVID-19 Containment

Sensors, Год журнала: 2023, Номер 23(1), С. 527 - 527

Опубликована: Янв. 3, 2023

Artificial intelligence has significantly enhanced the research paradigm and spectrum with a substantiated promise of continuous applicability in real world domain. intelligence, driving force current technological revolution, been used many frontiers, including education, security, gaming, finance, robotics, autonomous systems, entertainment, most importantly healthcare sector. With rise COVID-19 pandemic, several prediction detection methods using artificial have employed to understand, forecast, handle, curtail ensuing threats. In this study, recent related publications, methodologies medical reports were investigated purpose studying intelligence's role pandemic. This study presents comprehensive review specific attention machine learning, deep image processing, object detection, segmentation, few-shot learning studies that utilized tasks COVID-19. particular, genetic analysis, clinical data sound biomedical classification, socio-demographic anomaly health monitoring, personal protective equipment (PPE) observation, social control, patients' mortality risk approaches forecast threatening factors demonstrates artificial-intelligence-based algorithms integrated into Internet Things wearable devices quite effective efficient forecasting insights which actionable through wide usage. The results produced by prove is promising arena can be applied for disease prognosis, forecasting, drug discovery, development sector on global scale. We indeed played important helping fight against COVID-19, insightful knowledge provided here could extremely beneficial practitioners experts domain implement systems curbing next pandemic or disaster.

Язык: Английский

---

Self-supervised deep learning model for COVID-19 lung CT image segmentation highlighting putative causal relationship among age, underlying disease and COVID-19

Journal of Translational Medicine, Год журнала: 2021, Номер 19(1)

Опубликована: Июль 26, 2021

Coronavirus disease 2019 (COVID-19) is very contagious. Cases appear faster than the available Polymerase Chain Reaction test kits in many countries. Recently, lung computerized tomography (CT) has been used as an auxiliary COVID-19 testing approach. Automatic analysis of CT images needed to increase diagnostic efficiency and release human participant. Deep learning successful automatically solving computer vision problems. Thus, it can be introduced automatic rapid diagnosis. Many advanced deep learning-based vison techniques were developed model performance but have not medical image analysis.In this study, we propose a self-supervised two-stage segment lesions (ground-glass opacity consolidation) from chest support The proposed integrates several such generative adversarial inpainting, focal loss, lookahead optimizer. Two real-life datasets evaluate model's compared previous related works. To explore clinical biological mechanism predicted lesion segments, extract some engineered features lesions. We their mediation effects on relationship age with severity, well underlying diseases severity using statistic analysis.The best overall F1 score observed segmentation (0.63) two baseline models (0.55, 0.49). also identified phenotypes that mediate potential causal between severity.This work contributes promising provides segments interpretability. could raw higher accuracy these are associated through mediating known (age diseases).

Язык: Английский

---

xViTCOS: Explainable Vision Transformer Based COVID-19 Screening Using Radiography

IEEE Journal of Translational Engineering in Health and Medicine, Год журнала: 2021, Номер 10, С. 1 - 10

Опубликована: Дек. 8, 2021

Objective: Since its outbreak, the rapid spread of COrona VIrus Disease 2019 (COVID-19) across globe has pushed health care system in many countries to verge collapse. Therefore, it is imperative correctly identify COVID-19 positive patients and isolate them as soon possible contain disease reduce ongoing burden on healthcare system. The primary screening test, RT-PCR although accurate reliable, a long turn-around time. In recent past, several researchers have demonstrated use Deep Learning (DL) methods chest radiography (such X-ray CT) for detection. However, existing CNN based DL fail capture global context due their inherent image-specific inductive bias. Methods: Motivated by this, this work, we propose vision transformers (instead convolutional networks) using CT images. We employ multi-stage transfer learning technique address issue data scarcity. Furthermore, show that features learned our transformer networks are explainable. Results: demonstrate method not only quantitatively outperforms benchmarks but also focuses meaningful regions images detection (as confirmed Radiologists), aiding diagnosis localization infected area. code implementation can be found here - https://github.com/arnabkmondal/xViTCOS. Conclusion: proposed will help timely identification efficient utilization limited resources.

Язык: Английский

---