检索结果-内蒙古大学图书馆

Bias in Machine Learning Models Can Be Significantly Mitigated by Careful Training: Evidence from Neuroimaging Studies

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Wang, Rongguang Chaudhari, Pratik Davatzikos, Christos Department of Electrical and Systems Engineering University of Pennsylvania United States Center for AI and Data Science for Integrated Diagnostics University of Pennsylvania United States Center for Biomedical Image Computing and Analytics University of Pennsylvania United States Department of Computer and Information Science University of Pennsylvania United States Department of Radiology Perelman School of Medicine University of Pennsylvania United States

Despite the great promise that machine learning has offered in many fields of medicine, it has also raised concerns about potential biases and poor generalization across genders, age distributions, races and ethnicities, hospitals, and data acquisition equipment and protocols. In the current study, and in the context of three brain diseases, we provide evidence which suggests that when properly trained, machine learning models can generalize well across diverse conditions and do not necessarily suffer from bias. Specifically, by using multi-study magnetic resonance imaging consortia for diagnosing Alzheimer's disease, schizophrenia, and autism spectrum disorder, we find that well-trained models have a high area-under-the-curve (AUC) on subjects across different subgroups pertaining to attributes such as gender, age, racial groups, and different clinical studies and are unbiased under multiple fairness metrics such as demographic parity difference, equalized odds difference, equal opportunity difference etc. We find that models that incorporate multi-source data from demographic, clinical, genetic factors and cognitive scores are also unbiased. These models have better predictive AUC across subgroups than those trained only with imaging features but there are also situations when these additional features do not help. © 2022, CC BY.

关键词： Magnetic resonance imaging

Recurrent Neural Networks for Modelling Gross Primary Production

学校读者我要写书评

暂无评论

Recurrent Neural Networks for Modelling Gross Primary Produc...

IEEE International Symposium on Geoscience and Remote Sensing (IGARSS)

作者： David Montero Miguel D. Mahecha Francesco Martinuzzi César Aybar Anne Klosterhalfen Alexander Knohl Franziska Koebsch Jesús Anaya Sebastian Wieneke Remote Sensing Centre for Earth System Research (RSC4Earth) Leipzig University Institute of Earth System Science & Remote Sensing Leipzig University German Centre for Integrative Biodiversity Research (iDiv) Halle-Jena-Leipzig Department of Remote Sensing Helmholtz Centre for Environmental Research (UFZ) Center for Scalable Data Analytics and Artificial Intelligence (ScaDS.AI) Leipzig University Image Processing Laboratory Universitat de València Bioclimatology University of Göttingen Grupo de Investigaciones y Mediciones Ambientales (GEMA) Universidad de Medellín

ISBN: (数字)9798350360325

ISBN: (纸本)9798350360332

Accurate quantification of Gross Primary Production (GPP) is crucial for understanding terrestrial carbon dynamics. It represents the largest atmosphere-to-land CO 2 flux, especially significant for forests. Eddy Covariance (EC) measurements are widely used for ecosystem-scale GPP quantification but are globally sparse. In areas lacking local EC measurements, remote sensing (RS) data are typically utilised to estimate GPP after statistically relating them to in-situ data. Deep learning offers novel perspectives, and the potential of recurrent neural network architectures for estimating daily GPP remains underexplored. This study presents a comparative analysis of three architectures: Recurrent Neural Networks (RNNs), Gated Recurrent Units (GRUs), and Long-Short Term Memory (LSTMs). Our findings reveal comparable performance across all models for full-year and growing season predictions. Notably, LSTMs outperform in predicting climate-induced GPP extremes. Furthermore, our analysis highlights the importance of incorporating radiation and RS inputs (optical, temperature, and radar) for accurate GPP predictions, particularly during climate extremes.

关键词： Radar remote sensing Recurrent neural networks Laser radar Atmospheric measurements Ultraviolet sources Biological system modeling Production

Applications of Generative Adversarial Networks in Neuroimaging and Clinical Neuroscience

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Wang, Rongguang Bashyam, Vishnu Yang, Zhijian Yu, Fanyang Tassopoulou, Vasiliki Chintapalli, Sai Spandana Skampardoni, Ioanna Sreepada, Lasya P. Sahoo, Dushyant Nikita, Konstantina Abdulkadir, Ahmed Wen, Junhao Davatzikos, Christos Center for AI and Data Science for Integrated Diagnostics University of Pennsylvania Philadelphia United States Center for Biomedical Image Computing and Analytics University of Pennsylvania Philadelphia United States School of Electrical and Computer Engineering National Technical University of Athens Athens Greece Department of Clinical Neurosciences Lausanne University Hospital University of Lausanne Lausanne Switzerland Department of Radiology Perelman School of Medicine University of Pennsylvania Philadelphia United States

Generative adversarial networks (GANs) are one powerful type of deep learning models that have been successfully utilized in numerous fields. They belong to the broader family of generative methods, which learn to generate realistic data with a probabilistic model by learning distributions from real samples. In the clinical context, GANs have shown enhanced capabilities in capturing spatially complex, nonlinear, and potentially subtle disease effects compared to traditional generative methods. This review critically appraises the existing literature on the applications of GANs in imaging studies of various neurological conditions, including Alzheimer’s disease, brain tumors, brain aging, and multiple sclerosis. We provide an intuitive explanation of various GAN methods for each application and further discuss the main challenges, open questions, and promising future directions of leveraging GANs in neuroimaging. We aim to bridge the gap between advanced deep learning methods and neurology research by highlighting how GANs can be leveraged to support clinical decision making and contribute to a better understanding of the structural and functional patterns of brain diseases. © 2022, CC BY.

关键词： Generative adversarial networks

Transcriptional and Neurochemical Signatures of Cerebral Blood Flow Alterations in Individuals With Schizophrenia or at Clinical High Risk for Psychosis

学校读者我要写书评

暂无评论

Biological Psychiatry 2025年第2期98卷 144-155页

作者： Knight, Samuel R. Abbasova, Leyla Zeighami, Yashar Hansen, Justine Y. Martins, Daniel Zelaya, Fernando Dipasquale, Ottavia Liu, Thomas Shin, David Bossong, Matthijs Azis, Matilda Antoniades, Mathilde Howes, Oliver D. Bonoldi, Ilaria Egerton, Alice Allen, Paul McGuire, Philip Modinos, Gemma Department of Psychological Medicine Institute of Psychiatry Psychology and Neuroscience King's College London London United Kingdom Centre for Developmental Neurobiology Institute of Psychiatry Psychology and Neuroscience King's College London London United Kingdom Medical Research Council Centre for Neurodevelopmental Disorders King's College London London United Kingdom Douglas Research Centre Department of Psychiatry McGill University Montreal QC Canada Montreal Neurological Institute McGill University Montreal QC Canada Department of Neuroimaging Institute of Psychiatry Psychology and Neuroscience King's College London London United Kingdom Olea Medical La Ciotat France Centre for Functional MRI University of California San Diego San Diego CA United States Global MR Applications and Workflow GE Healthcare Menlo Park CA United States Department of Psychosis Studies Institute of Psychiatry Psychology and Neuroscience King's College London London United Kingdom Department of Psychiatry Brain Center Rudoph Magnus University Medical Center Utrecht Utrecht Netherlands Center for AI and Data Science for Integrated Diagnostics and Center for Biomedical Image Computing and Analytics University of Pennsylvania Philadelphia PA United States Department of Psychiatry Oxford University Oxford United Kingdom

Background: The brain integrates multiple scales of description, from the level of cells and molecules to large-scale networks and behavior. Understanding relationships across these scales may be fundamental to advancing understanding of brain function in health and disease. Recent neuroimaging research has shown that functional brain alterations that are associated with schizophrenia spectrum disorders (SSDs) are already present in young adults at clinical high risk for psychosis (CHR-P), but the cellular and molecular determinants of these alterations remain unclear. Methods: Here, we used regional cerebral blood flow (rCBF) data from 425 individuals (122 with an SSD compared with 116 healthy control participants [HCs] and 129 individuals at CHR-P compared with 58 HCs) and applied a novel pipeline to integrate brainwide rCBF case-control maps with publicly available transcriptomic data (17,205 gene maps) and neurotransmitter atlases (19 maps) from 1074 healthy volunteers. Results: We identified significant correlations between astrocyte, oligodendrocyte, oligodendrocyte precursor cell, and vascular leptomeningeal cell gene modules for both SSD and CHR-P rCBF phenotypes. Additionally, endothelial cell genes were correlated in SSD, and microglia in CHR-P. Receptor distribution significantly predicted case-control rCBF differences, with dominance analysis highlighting dopamine (D₁, D₂, dopamine transporter), acetylcholine (VAChT, M₁), gamma-aminobutyric acid A (GABA_A), and glutamate (NMDA) receptors as key predictors for SSD (R²_adjusted = 0.58, false discovery rate [FDR]–corrected p <.05) and CHR-P (R²_adjusted = 0.6, p_FDR <.05) rCBF phenotypes. These associations were primarily localized in subcortical regions and implicate cell types involved in stress response and inflammation, alongside specific neuroreceptor systems, in shared and distinct rCBF phenotypes in psychos

关键词： Arterial spin labeling Biomarkers Gene expression Mental health Neuroimaging Neurotransmitters

An international study of stain variability in histopathology using qualitative and quantitative analysis

学校读者我要写书评

暂无评论

Journal of Pathology Informatics 2025年 17卷 100423-100423页

作者： Dunn, Catriona Brettle, David Hodgson, Chantell Hughes, Robert Treanor, Darren National Pathology Imaging Co-operative Leeds Teaching Hospitals NHS Trust Beckett Street Leeds United Kingdom UK NEQAS Cellular Pathology Technique Haylofts St Thomas Street Haymarket Newcastle United Kingdom Department of Histopathology Leeds Teaching Hospitals NHS Trust Beckett Street Leeds United Kingdom Department of Pathology and Data Analytics University of Leeds Beckett Street Leeds United Kingdom Department of Clinical Pathology and Clinical and Experimental Medicine Linköping University Linköping Sweden Centre for Medical Image Science and Visualisation Linköping University Linköping Sweden

Hematoxylin and eosin (H&E) staining accounts for over 80% of slides stained worldwide. Although routinely used, there are high levels of variation between labs due to different staining methods. Staining is a pivotal part of slide preparation, but quality control is largely subjective, with overall clinical assurance provided by external quality assessment (EQA) services, underpinned by expert assessment. Digital pathology offers the potential to provide objective quantification of stain, through color analysis, to augment EQA assessment. This large-scale study evaluated H&E staining in 247 international labs participating in the UK NEQAS CPT EQA programme. Tissue sections were circulated to each lab to stain using their routine H&E staining protocol. The slides were reviewed by independent expert UK NEQAS CPT assessors, and quantitative digital analysis was conducted, comprising of H&E color deconvolution and color difference determination (ΔE). Most labs (69%) achieved an EQA score indicating good or excellent staining, with high inter-observer concordance to support this (92.5% within one mark of each other). H&E color difference, ΔE, showed 60% of labs were within 2 ΔE of the mean, which is considered as only perceptible through close observation. There was little correlation found between H&E intensity and assessor score, however, the H&E intensity ratio indicated a trend with assessor score suggesting there may be an optimal stain relationship that should be investigated further. The presented hybrid analysis combines expert analysis with objective data. This has the potential to inform upon optimal tissue staining and allows us to consider quantitative standards of H&E staining in pathology practice. © 2025 The Authors

关键词： Digital pathology H&E Quality assurance Stain variation

Embedding Space Augmentation for Weakly Supervised Learning in Whole-Slide images

学校读者我要写书评

暂无评论

Embedding Space Augmentation for Weakly Supervised Learning ...

IEEE International Symposium on Biomedical Imaging

作者： Imaad Zaffar Guillaume Jaume Nasir Rajpoot Faisal Mahmood Department of Computer Science University College London UK Department of Pathology Harvard Medical School Brigham and Women’s Hospital Boston MA USA Department of Pathology Harvard Medical School Massachusetts General Hospital Boston MA USA Cancer Program Broad Institute of Harvard and MIT Cambridge MA USA Data Science Program Dana-Farber Cancer Institute Boston MA USA Department of Computer Science Tissue Image Analytics Centre University of Warwick Coventry UK Department of Pathology University Hospitals Coventry and Warwickshire NHS Trust Coventry UK The Alan Turing Institute London UK

Multiple Instance Learning (MIL) is a widely employed framework for learning on gigapixel whole-slide images (WSIs) from WSI-level annotations. In most MIL-based analytical pipelines for WSI-level analysis, the WSIs are divided into patches, and deep features for patches (i.e., patch embeddings) are extracted prior to training to reduce the overall computational cost and cope with the GPUs’ limited RAM. Because of this bottleneck, incorporating patch-level data augmentations during training adds an extra computational burden. To overcome this limitation, we present EmbAugmenter, a data augmentation generative adversarial network (DA-GAN) that can synthesize data augmentations in the embedding space rather than in the pixel space, thereby significantly reducing the computational requirements. Experiments on the SICAPv2 dataset show that our approach outperforms MIL without augmentation and is on par with traditional patch-level augmentation for MIL training while being substantially faster.

关键词：

Completely self-supervised crowd counting via distribution matching

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Sam, Deepak Babu Agarwalla, Abhinav Joseph, Jimmy Sindagi, Vishwanath A. Babu, R. Venkatesh Patel, Vishal M. Video Analytics Lab Department of Computational and Data Sciences Indian Institute of Science Bangalore India Vision & Image Understanding Lab Department of Electrical and Computer Engineering Johns Hopkins University Baltimore United States

Dense crowd counting is a challenging task that demands millions of head annotations for training models. Though existing self-supervised approaches could learn good representations, they require some labeled data to map these features to the end task of density estimation. We mitigate this issue with the proposed paradigm of complete self-supervision, which does not need even a single labeled image. The only input required to train, apart from a large set of unlabeled crowd images, is the approximate upper limit of the crowd count for the given dataset. Our method dwells on the idea that natural crowds follow a power law distribution, which could be leveraged to yield error signals for backpropagation. A density regressor is first pretrained with self-supervision and then the distribution of predictions is matched to the prior by optimizing Sinkhorn distance between the two. Experiments show that this results in effective learning of crowd features and delivers significant counting performance. Furthermore, we establish the superiority of our method in less data setting as well. The code and models for our approach is available at https://***/val-iisc/css-ccnn. Copyright © 2020, The Authors. All rights reserved.

关键词： Deep learning

Restoration of marker occluded hematoxylin and eosin stained whole slide histology images using generative adversarial networks

学校读者我要写书评

暂无评论

arXiv 2019年

作者： Venkatesh, Bairavi Shah, Tosha Chen, Antong Ghafurian, Soheil Image Data Analytics Informatics IT Merck & Co. Inc. West PointPA19486 United States Image Data Analytics Informatics IT Merck & Co. Inc. RahwayNJ07065 United States

It is common for pathologists to annotate specific regions of the tissue, such as tumor, directly on the glass slide with markers. Although this practice was helpful prior to the advent of histology whole slide digitization, it often occludes important details which are increasingly relevant to immuno-oncology due to recent advancements in digital pathology imaging techniques. The current work uses a generative adversarial network with cycle loss to remove these annotations while still maintaining the underlying structure of the tissue by solving an image-to-image translation problem. We train our network on up to 300 whole slide images with marker inks and show that 70% of the corrected image patches are indistinguishable from originally uncontaminated image tissue to a human expert. This portion increases 97% when we replace the human expert with a deep residual network. We demonstrated the fidelity of the method to the original image by calculating the correlation between image gradient magnitudes. We observed a revival of up to 94,000 nuclei per slide in our dataset, the majority of which were located on tissue border. Copyright © 2019, The Authors. All rights reserved.

关键词： Restoration

Novel genomic loci shape two neuroanatomical dimensions of AD in non-demented populations

学校读者我要写书评

暂无评论

Alzheimer's & Dementia 2023年第S10期19卷

作者： Junhao Wen Zhijian Yang Ioanna Skampardoni Marilena De Pian Guray Erus Elizabeth Mamourian Yuhan Cui Bingxin Zhao Ilya M. Nasrallah Arthur W. Toga Haochang Shou Li Shen Christos Davatzikos Artificial Intelligence in Biomedical Imaging Laboratory (AIBIL) Center for and Data Science for Integrated Diagnostics (AI2D) Perelman School of Medicine University of Pennsylvania Philadelphia PA USA Center for Biomedical Image Computing and Analytics University of Pennsylvania Philadelphia PA USA Centre for Biomedical Image Computing and Analytics University of Pennsylvania Philadelphia PA USA University of Pennsylvania philadelphia PA USA Upenn Philadelphia PA USA University of Pennsylvania Philadelphia PA USA Laboratory of Neuro Imaging (LONI) University of Southern California Los Angeles CA USA Department of Biostatistics Epidemiology & Informatics University of Pennsylvania Philadelphia PA USA University of Pennsylvania Philadelphia PA Greece

Background Artificial intelligence has been in burgeoning demand to be applied to magnetic imaging resonance (MRI) data to parse the neuroanatomical heterogeneity of Alzheimer’s disease (AD) and/or mild cognitive impairment (MCI) (Wen et al., 2021a, 2021b; Yang et al., 2021). However, the genetic heterogeneity leading to this neuroanatomical heterogeneity, especially at the early asymptomatic stages, is not fully elucidated. Method A deep semi-supervised representation learning method, called Surreal-GAN (Yang et al., 2022), was applied to T1-weighted MRI from AD/MCI patients from ADNI to define the two neuroanatomical dimensions of AD at the symptomatic stages. The R1 dimension was characterized by widespread brain atrophy without an exclusive focus on the medial temporal lobe, whereas the R2 dimension showed focal medial temporal lobe atrophy (Wen et al., 2022). We then applied the trained model to 33,540 non-demented individuals from UK Biobank. Imputed genotyping data were analyzed in genome-wide association studies (GWAS). We performed multiple linear regressions controlling for confounding factors (e.g., age) via Plink (Purcell et al., 2007). We defined a genomic locus (represented by the top lead SNP) considering linkage disequilibrium. Finally, we manually queried all leading SNPs on GWAS Catalog (Buniello et al., 2019) to determine if the genomic loci were novel, provided they were not previously associated with any clinical traits. Result GWAS discovered 24 genomic loci, of which 7 were novel for R1 and R2, respectively (Fig. 1). For R1, 8 mapped genes, based on physical positions, eQTL, and chromatin interactions, were associated with AD-related traits in the literature. For example, the KDM2B gene from the genomic locus (rs9795600, P-value = 1.13e-08) was previously associated with AD (Gouveia et al., 2022). For R2, 13 mapped genes were associated with AD-related traits. For example, the TESC gene from the genomic locus (rs77956314, P-value = 9.99e-29) w

关键词：