Bilateral adaptive graph convolutional network on CT based Covid-19 diagnosis with uncertainty-aware consensus-assisted multiple instance learning DOI Creative Commons
Yanda Meng, Joshua Bridge,

Cliff Addison

et al.

Medical Image Analysis, Journal Year: 2022, Volume and Issue: 84, P. 102722 - 102722

Published: Dec. 15, 2022

Coronavirus disease (COVID-19) has caused a worldwide pandemic, putting millions of people's health and lives in jeopardy. Detecting infected patients early on chest computed tomography (CT) is critical combating COVID-19. Harnessing uncertainty-aware consensus-assisted multiple instance learning (UC-MIL), we propose to diagnose COVID-19 using new bilateral adaptive graph-based (BA-GCN) model that can use both 2D 3D discriminative information CT volumes with arbitrary number slices. Given the importance lung segmentation for this task, have created largest manual annotation dataset so far 7,768 slices from patients, used it train segment lungs individual mask as regions interest subsequent analyses. We then UC-MIL estimate uncertainty each prediction consensus between predictions slice automatically select fixed reliable reasoning. Finally, adaptively constructed BA-GCN vertices different granularity levels (2D 3D) aggregate multi-level features final diagnosis benefits graph convolution network's superiority tackle cross-granularity relationships. Experimental results three datasets demonstrated our produce accurate any slices, which outperforms existing approaches terms generalisation ability. To promote reproducible research, made datasets, including annotations cleaned dataset, well implementation code, available at https://doi.org/10.5281/zenodo.6361963.

Language: Английский

Deep learning for unsupervised domain adaptation in medical imaging: Recent advancements and future perspectives DOI

Suruchi Kumari,

Pravendra Singh

Computers in Biology and Medicine, Journal Year: 2023, Volume and Issue: 170, P. 107912 - 107912

Published: Dec. 28, 2023

Language: Английский

Citations

22
Multi-objective optimization determines when, which and how to fuse deep networks: An application to predict COVID-19 outcomes DOI Creative Commons
Valerio Guarrasi, Paolo Soda

Computers in Biology and Medicine, Journal Year: 2023, Volume and Issue: 154, P. 106625 - 106625

Published: Feb. 2, 2023

The COVID-19 pandemic has caused millions of cases and deaths the AI-related scientific community, after being involved with detecting signs in medical images, been now directing efforts towards development methods that can predict progression disease. This task is multimodal by its very nature and, recently, baseline results achieved on publicly available AIforCOVID dataset have shown chest X-ray scans clinical information are useful to identify patients at risk severe outcomes. While deep learning superior performance several fields, most it considers unimodal data only. In this respect, when, which how fuse different modalities an open challenge learning. To cope these three questions here we present a novel approach optimizing setup end-to-end model. It exploits Pareto multi-objective optimization working metric diversity score multiple candidate neural networks be fused. We test our method dataset, attaining state-of-the-art results, not only outperforming but also robust external validation. Moreover, exploiting XAI algorithms figure out hierarchy among extract features' intra-modality importance, enriching trust predictions made

Language: Английский

Citations

17
A systematic meta‐analysis of immune signatures in patients with COVID‐19 DOI Creative Commons
Kun Liu, Tong Yang, Xue‐Fang Peng

et al.

Reviews in Medical Virology, Journal Year: 2020, Volume and Issue: 31(4)

Published: Nov. 20, 2020

Summary Currently severe acute respiratory syndrome coronavirus 2 (SARS‐CoV‐2) transmission has been on the rise worldwide. Predicting outcome in COVID‐19 remains challenging, and search for more robust predictors continues. We made a systematic meta‐analysis current literature from 1 January 2020 to 15 August that independently evaluated 32 circulatory immunological signatures were compared between patients with different disease severity was made. Their roles as of determined well. A total 149 distinct studies ten cytokines, four antibodies, T cells, B NK neutrophils, monocytes, eosinophils basophils included. Compared non‐severe COVID‐19, serum levels Interleukins (IL)‐2, IL‐2R, IL‐4, IL‐6, IL‐8, IL‐10 tumor necrosis factor α significantly up‐regulated patients, largest inter‐group differences observed IL‐6 IL‐10. In contrast, IL‐5, IL‐1β Interferon (IFN)‐γ did not show significant difference. Four mediators cells count, including CD3 + T, CD4 CD8 CD25 CD127 ‐ Treg, together CD19 count CD16 CD56 all consistently depressed group than group. SARS‐CoV‐2 specific IgA IgG antibodies higher group, while IgM antibody slightly lower those IgE showed no differences. The combination especially IL‐10, cell related immune can be used biomarkers predict following infection.

Language: Английский

Citations

49
An explainable AI system for automated COVID-19 assessment and lesion categorization from CT-scans DOI Open Access
Matteo Pennisi, Isaak Kavasidis, Concetto Spampinato

et al.

Artificial Intelligence in Medicine, Journal Year: 2021, Volume and Issue: 118, P. 102114 - 102114

Published: May 21, 2021

Language: Английский

Citations

37
Bilateral adaptive graph convolutional network on CT based Covid-19 diagnosis with uncertainty-aware consensus-assisted multiple instance learning DOI Creative Commons
Yanda Meng, Joshua Bridge,

Cliff Addison

et al.

Medical Image Analysis, Journal Year: 2022, Volume and Issue: 84, P. 102722 - 102722

Published: Dec. 15, 2022

Coronavirus disease (COVID-19) has caused a worldwide pandemic, putting millions of people's health and lives in jeopardy. Detecting infected patients early on chest computed tomography (CT) is critical combating COVID-19. Harnessing uncertainty-aware consensus-assisted multiple instance learning (UC-MIL), we propose to diagnose COVID-19 using new bilateral adaptive graph-based (BA-GCN) model that can use both 2D 3D discriminative information CT volumes with arbitrary number slices. Given the importance lung segmentation for this task, have created largest manual annotation dataset so far 7,768 slices from patients, used it train segment lungs individual mask as regions interest subsequent analyses. We then UC-MIL estimate uncertainty each prediction consensus between predictions slice automatically select fixed reliable reasoning. Finally, adaptively constructed BA-GCN vertices different granularity levels (2D 3D) aggregate multi-level features final diagnosis benefits graph convolution network's superiority tackle cross-granularity relationships. Experimental results three datasets demonstrated our produce accurate any slices, which outperforms existing approaches terms generalisation ability. To promote reproducible research, made datasets, including annotations cleaned dataset, well implementation code, available at https://doi.org/10.5281/zenodo.6361963.

Language: Английский

Citations

28