Article
Computer Science, Artificial Intelligence
Bahram Jafrasteh, Daniel Hernandez-Lobato, Simon Pedro Lubian-Lopez, Isabel Benavente-Fernandez
Summary: A missing value indicates that a particular attribute of an instance is not recorded, and it is common in real-life datasets. Most machine learning methods cannot handle missing values, so it is necessary to impute them before training. We propose a hierarchical composition of sparse Gaussian Processes called Missing GP (MGP) to predict missing values, and MGP outperforms other state-of-the-art methods in imputing missing values.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Carlos Villacampa-Calvo, Gonzalo Hernandez-Munoz, Daniel Hernandez-Lobato
Summary: This paper proposes a method for approximate inference in deep Gaussian processes by minimizing alpha-divergences. The method extends the existing approaches of variational inference and expectation propagation and demonstrates its feasibility and improvement through extensive experiments.
INTERNATIONAL JOURNAL OF APPROXIMATE REASONING
(2022)
Article
Computer Science, Interdisciplinary Applications
Italo Gomes Goncalves, Felipe Guadagnin, Diogo Peixoto Cordova
Summary: This paper introduces a variational Gaussian process (VGP) model specialized in spatial data by leveraging recent advances in the machine learning field. The model is highly modular and customizable, allowing for different assumptions about the data. The focus of this work is on multivariate robust regression using an adaptation of the e-insensitive loss function. The VGP model enables end-to-end modeling with normal score transformation, spatial pattern detection, and prediction. The paper also presents a methodology for handling large datasets and provides an open-source implementation.
COMPUTERS & GEOSCIENCES
(2022)
Article
Computer Science, Artificial Intelligence
Younghwan Jeon, Ganguk Hwang
Summary: This paper addresses the data association problem and proposes a Bayesian approach based on a mixture of Gaussian Processes (GPs) to adapt to changing observations. Experimental results and theoretical analysis demonstrate the effectiveness of the proposed method.
PATTERN RECOGNITION
(2022)
Article
Oncology
Mizuho Nishio, Hidetoshi Matsuo, Yasuhisa Kurata, Osamu Sugiyama, Koji Fujimoto
Summary: This study aimed to develop and evaluate an automatic prediction system for grading histopathological images of prostate cancer using a deep learning model and label distribution learning (LDL). Results showed that LDL improved the diagnostic performance of the automatic prediction system for cancer grading.
Article
Computer Science, Artificial Intelligence
Aristeidis Panos, Petros Dellaportas, Michalis K. Titsias
Summary: This study introduces a Gaussian process latent factor model for multi-label classification, which can capture correlations among class labels. To address computational challenges, several techniques are introduced, including variational sparse Gaussian process and stochastic optimization. The results demonstrate the practicality of this method in large-scale multi-label learning problems.
Article
Computer Science, Artificial Intelligence
Daniele Gammelli, Kasper Pryds Rolsted, Dario Pacino, Filipe Rodrigues
Summary: In this paper, a new model is proposed to deal with censored observations. By exploiting correlations between multiple outputs, the model combines the flexibility of Gaussian process with the ability to leverage information from correlated outputs, and achieves better estimation of the true process.
PATTERN RECOGNITION
(2022)
Article
Automation & Control Systems
Jeremias Knoblauch, Jack Jewson, Theodoros Damoulas
Summary: The paper advocates for an optimization-centric view of Bayesian inference, introducing the Rule of Three (ROT) as a generalized method for Bayesian posteriors. It also explores the applications of Generalized Variational Inference (GVI) posteriors and their potential to improve robustness and posterior marginals in Bayesian Neural Networks and Deep Gaussian Processes.
JOURNAL OF MACHINE LEARNING RESEARCH
(2022)
Article
Statistics & Probability
Andrew C. Miller, Lauren Anderson, Boris Leistedt, John P. Cunningham, David W. Hogg, David M. Blei
Summary: Accounting for interstellar dust is crucial to accurately measure physical properties of stars. This study presents a model based on Gaussian processes to capture the dust distribution and develops a likelihood model and inference method that can handle millions of astronomical observations. The proposed approach, named ZIGGY, successfully infers the spatial dust map with well-calibrated posterior uncertainties.
ANNALS OF APPLIED STATISTICS
(2022)
Article
Biology
A. Ben Hamida, M. Devanne, J. Weber, C. Truntzer, V Derangere, F. Ghiringhelli, G. Forestier, C. Wemmert
Summary: Digital pathology plays a major role in the diagnosis and prognosis of tumors. Deep Learning methods show promise for tissue classification and segmentation in histopathological images. This study focuses on using DL architectures to classify and highlight colon cancer regions in sparsely annotated data.
COMPUTERS IN BIOLOGY AND MEDICINE
(2021)
Article
Computer Science, Artificial Intelligence
Massimo Salvi, Martino Bosco, Luca Molinaro, Alessandro Gambella, Mauro Papotti, U. Rajendra Acharya, Filippo Molinari
Summary: The RINGS algorithm presented a new image segmentation method for prostate gland segmentation, achieving a high dice score of 90.16% and outperforming other state-of-the-art techniques. The hybrid segmentation strategy based on stroma detection maintained high sensitivity even in the presence of severe glandular degeneration, making it a valuable tool for accurate diagnosis and treatment in prostate cancer detection.
ARTIFICIAL INTELLIGENCE IN MEDICINE
(2021)
Article
Computer Science, Artificial Intelligence
Max-Heinrich Laves, Malte Tolle, Alexander Schlaefer, Sandy Engelhardt
Summary: POTOBIM is an approach to inverse problems in medical imaging that optimizes both the prior distribution and posterior temperature using Bayesian optimization. This leads to improved reconstruction accuracy and uncertainty estimation.
MEDICAL IMAGE ANALYSIS
(2022)
Article
Statistics & Probability
Seokhyun Chung, Raed Al Kontar
Summary: This article proposes a federated analytics framework called FedMGP, which utilizes edge computing power to learn a multi-output Gaussian process (MGP) in a decentralized manner. It overcomes the challenges of traditional MGP approaches, such as massive computing and storage demands, significant communication traffic, and compromised privacy.
Article
Computer Science, Artificial Intelligence
Tianling Liu, Ran Su, Changming Sun, Xiuting Li, Leyi Wei
Summary: Ovarian cancer is a serious disease that poses a threat to women's health worldwide, especially epithelial ovarian cancer (EOC). This study proposes a deep framework, named EOCSA, which utilizes pathological whole slide images (WSIs) to accurately predict the prognosis of EOC patients. The framework effectively extracts features and applies survival analysis models to estimate survival time. The experimental results demonstrate state-of-the-art performance.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Computer Science, Artificial Intelligence
Meghavi Rana, Megha Bhushan
Summary: This study demonstrates the accuracy of transfer learning-based deep learning methods in Computer-Aided Design (CAD) systems for the early detection and analysis of diseases such as lung cancer, brain tumor, and breast cancer. By utilizing pre-trained models, the time for deep learning-based tasks in computer vision can be reduced. The effectiveness of transfer learning models for tumor classification is explained, and the best performing models on a specific dataset are identified.
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Computer Science, Information Systems
Arne Schmidt, Julio Silva-Rodriguez, Rafael Molina, Valery Naranjo
Summary: In this research, SSL and MIL are combined to create a deep learning classifier that can learn from large datasets. By utilizing global diagnosis and labeled/unlabeled patches, the proposed model achieves competitive performance on various tasks with only a small percentage of patch labels.
Article
Computer Science, Interdisciplinary Applications
Miguel Lopez-Perez, Arne Schmidt, Yunan Wu, Rafael Molina, Aggelos K. Katsaggelos
Summary: This study presents a novel ICH detection model based on Multiple Instance Learning and Deep Gaussian Processes, which can be trained with scan-level annotations and achieves good results in experiments. The DGPMIL model demonstrates superior performance in multiple experiments and shows great potential for applications in medical image classification.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2022)
Article
Computer Science, Interdisciplinary Applications
Elena Paya, Lorena Bori, Adrian Colomer, Marcos Meseguer, Valery Naranjo
Summary: This study presents a novel methodology based on deep learning to automatically evaluate the morphological appearance of human embryos from time-lapse imaging. The proposed methods outperform conventional approaches and improve state-of-the-art embryology results. It demonstrates excellent potential for the inclusion of the models in clinical practice.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2022)
Article
Engineering, Biomedical
Fernando Perez-Bueno, Juan G. Serra, Miguel Vega, Javier Mateos, Rafael Molina, Aggelos K. Katsaggelos
Summary: This study presents a blind color deconvolution framework based on Bayesian modeling and inference. By separating multi-stained images into single stained bands and normalizing the stain colors, the generalization error of computer aided diagnosis systems can be reduced.
COMPUTERIZED MEDICAL IMAGING AND GRAPHICS
(2022)
Article
Computer Science, Artificial Intelligence
Pablo Ruiz, Pablo Morales-Alvarez, Scott Coughlin, Rafael Molina, Aggelos K. Katsaggelos
Summary: The acquisition of training labels is costly in machine learning classification tasks. Crowdsourcing has become a popular approach to label a training set, involving a large number of annotators. The GravitySpy project combines crowdsourced labels with expert labels to enhance the detection of gravitational waves. This study proposes a new probabilistic crowdsourcing model based on sparse Gaussian Processes (GPs) and demonstrates its effectiveness.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Chemistry, Multidisciplinary
Laetitia Launet, Yuandou Wang, Adrian Colomer, Jorge Igual, Cristian Pulgarin-Ospina, Spiros Koulouzis, Riccardo Bianchi, Andres Mosquera-Zamudio, Carlos Monteagudo, Valery Naranjo, Zhiming Zhao
Summary: Deep learning-based algorithms have made significant progress in recent years, but their development heavily relies on access to large datasets. Cross-silo federated learning has emerged as a solution to train collaborative models among multiple institutions without sharing raw data. This paper introduces the Notebook Federator, a solution that leverages the Jupyter environment to simplify the automation of federated learning and bridge the gap between federated learning and notebook users. The feasibility of this approach is demonstrated with a collaborative model achieving high accuracy on a digital pathology image analysis task.
APPLIED SCIENCES-BASEL
(2023)
Article
Computer Science, Information Systems
Miguel Lopez-Perez, Pablo Morales-Alvarez, Lee A. D. Cooper, Rafael Molina, Aggelos K. Katsaggelos
Summary: This article explores the use of crowdsourcing methods for medical applications and develops the Deep Gaussian Processes for Crowdsourcing (DGPCR) model to address the labeling data challenge. The results demonstrate that DGPCR outperforms other state-of-the-art deep-learning crowdsourcing methods and achieves better results in breast cancer classification.
Review
Oncology
Andres Mosquera-Zamudio, Laetitia Launet, Zahra Tabatabaei, Rafael Parra-Medina, Adrian Colomer, Javier Oliver Moll, Carlos Monteagudo, Emiel Janssen, Valery Naranjo
Summary: Deep learning has shown promising outcomes in surgical pathology, particularly in dermatopathology, for melanocytic tumors in whole slide images. This study analyzes previously published studies on deep learning techniques for automatic image analysis of melanocytic tumors. The analysis reveals research trends in diagnostic prediction, prognosis, and regions of interest, and emphasizes the considerations for implementing these models in real scenarios. The rise of artificial intelligence as a support tool in clinical pathology workflows is also highlighted.
Article
Computer Science, Artificial Intelligence
Rocio del Amor, Jose Perez-Cano, Miguel Lopez-Perez, Liria Terradez, Jose Aneiros-Fernandez, Sandra Morales, Javier Mateos, Rafael Molina, Valery Naranjo
Summary: Digital pathology has become an essential tool for tumor diagnosis and prognosis, and Deep Learning algorithms combined with Whole Slide Images have enabled the development of Artificial Intelligence systems to support this process. This study presents a crowdsourcing-multiple instance learning protocol for creating and utilizing a dataset of Cutaneous Spindle Cell neoplasms. The proposed method outperforms traditional aggregation methods and supervised models on the test set, demonstrating its effectiveness in crowd-based classification tasks.
ARTIFICIAL INTELLIGENCE IN MEDICINE
(2023)
Article
Biotechnology & Applied Microbiology
Zahra Tabatabaei, Yuandou Wang, Adrian Colomer, Javier Oliver Moll, Zhiming Zhao, Valery Naranjo
Summary: The paper proposes a Federated Content-Based Medical Image Retrieval (FedCBMIR) tool that utilizes federated learning to address the challenges of acquiring diverse medical datasets for training. The tool distributes an unsupervised feature extractor to collaborative centers for training, resulting in shorter training times and higher performance.
BIOENGINEERING-BASEL
(2023)
Article
Computer Science, Artificial Intelligence
Arne Schmidt, Pablo Morales-Alvarez, Rafael Molina
Summary: Multiple Instance Learning (MIL) is increasingly popular as a weakly supervised learning paradigm, requiring less labeling effort than fully supervised methods. In this work, a probabilistic attention mechanism called the attention Gaussian process (AGP) is introduced for deep MIL, providing accurate predictions, instance-level explainability, and uncertainty estimations. The proposed AGP model outperforms state-of-the-art MIL approaches, including deterministic deep learning ones, in both synthetic and real-world cancer detection experiments, showcasing its superior performance and reliability.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Interdisciplinary Applications
Alireza Karimi, Reza Razaghi, Siddharth Daniel D'costa, Saeed Torbati, Sina Ebrahimi, Seyed Mohammadali Rahmati, Mary J. Kelley, Ted S. Acott, Haiyan Gong
Summary: This study investigated the biomechanical properties of the conventional aqueous outflow pathway using fluid-structure interaction. The results showed that the distribution of aqueous humor wall shear stress within this pathway is not uniform, which may contribute to our understanding of the underlying selective mechanisms.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2024)
Article
Computer Science, Interdisciplinary Applications
Robert V. Bergen, Jean-Francois Rajotte, Fereshteh Yousefirizi, Arman Rahmim, Raymond T. Ng
Summary: This article introduces a 3D generative model called TrGAN, which can generate medical images with important features and statistical properties while protecting privacy. By evaluating through a membership inference attack, the fidelity, utility, and privacy trade-offs of the model were studied.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2024)
Article
Computer Science, Interdisciplinary Applications
Hoda Mashayekhi, Mostafa Nazari, Fatemeh Jafarinejad, Nader Meskin
Summary: In this study, a novel model-free adaptive control method based on deep reinforcement learning (DRL) is proposed for cancer chemotherapy drug dosing. The method models the state variables and control action in their original infinite spaces, providing a more realistic solution. Numerical analysis shows the superior performance of the proposed method compared to the state-of-the-art RL-based approach.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2024)
Article
Computer Science, Interdisciplinary Applications
Hao Sun, Bao Li, Liyuan Zhang, Yanping Zhang, Jincheng Liu, Suqin Huang, Xiaolu Xi, Youjun Liu
Summary: In cases of moderate stenosis in the internal carotid artery, the A1 segment of the anterior cerebral artery or the posterior communicating artery within the Circle of Willis may show a hemodynamic environment with high OSI and low TAWSS, increasing the risk of atherosclerosis development and stenosis in the CoW.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2024)
Article
Computer Science, Interdisciplinary Applications
Ilaria Toniolo, Paola Pirini, Silvana Perretta, Emanuele Luigi Carniel, Alice Berardo
Summary: This study compared the outcomes of endoscopic sleeve gastroplasty (ESG) and laparoscopic sleeve gastrectomy (LSG) in weight loss surgery using computational models of specific patients. The results showed significant differences between the two procedures in terms of stomach volume reduction and mechanical stimulation. A predictive model was proposed to support surgical planning and estimation of volume reduction after ESG.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2024)
Article
Computer Science, Interdisciplinary Applications
Chun-You Chen, Ya-Lin Chen, Jeremiah Scholl, Hsuan-Chia Yang, Yu-Chuan (Jack) Li
Summary: This study evaluated the overall performance of a machine learning-based CDSS (MedGuard) in triggering clinically relevant alerts and intercepting inappropriate drug errors and LASA drug errors. The results showed that MedGuard has the ability to improve patients' safety by triggering clinically valid alerts.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2024)
Article
Computer Science, Interdisciplinary Applications
Lingzhi Tang, Xueqi Wang, Jinzhu Yang, Yonghuai Wang, Mingjun Qu, HongHe Li
Summary: In this paper, a dynamical local feature fusion net for automatically recognizing aortic valve calcification (AVC) from echocardiographic images is proposed. The network segments high-echo areas and adjusts the selection of local features to better integrate global and local semantic representations. Experimental results demonstrate the effectiveness of the proposed approach.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2024)
Article
Computer Science, Interdisciplinary Applications
You-Lei Fu, Wu Song, Wanni Xu, Jie Lin, Xuchao Nian
Summary: This study investigates the combination of surface electromyographic signals (sEMG) and deep learning-based CNN networks to study the interaction between humans and products and the impact on body comfort. It compares the advantages and disadvantages of different CNN networks and finds that DenseNet has unique advantages over other algorithms in terms of accuracy and ease of training, while mitigating issues of gradient disappearance and model degradation.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2024)
Article
Computer Science, Interdisciplinary Applications
Moritz Rempe, Florian Mentzel, Kelsey L. Pomykala, Johannes Haubold, Felix Nensa, Kevin Kroeninger, Jan Egger, Jens Kleesiek
Summary: In this study, a deep learning-based skull stripping algorithm for MRI was proposed, which works directly in the complex valued k-space and preserves the phase information. The results showed that the algorithm achieved similar results to the ground truth, with higher accuracy in the slices above the eye region. This approach not only preserves valuable information for further diagnostics, but also enables immediate anonymization of patient data before being transformed into the image domain.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2024)
Article
Computer Science, Interdisciplinary Applications
Ziyang Chen, Laura Cruciani, Elena Lievore, Matteo Fontana, Ottavio De Cobelli, Gennaro Musi, Giancarlo Ferrigno, Elena De Momi
Summary: In this paper, a deep learning-based approach is proposed to recover 3D information of intra-operative scenes, which can enhance the safety of robot-assisted surgery by implementing depth estimation using stereo images.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2024)
Article
Computer Science, Interdisciplinary Applications
Ao Leng, Bolun Zeng, Yizhou Chen, Puxun Tu, Baoxin Tao, Xiaojun Chen
Summary: This study presents a novel training system for zygomatic implant surgery, which offers a more realistic simulation and training solution. By integrating visual, haptic, and auditory feedback, the system achieves global rigid-body collisions and soft tissue simulation, effectively improving surgeons' proficiency.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2024)
Article
Computer Science, Interdisciplinary Applications
Yingjie Wang, Xueqing Yin
Summary: This study developed an integrated computational model combining coronary flow and myocardial perfusion models to achieve physiologically accurate simulations. The model has the potential for clinical application in diagnosing insufficient myocardial perfusion.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2024)
Article
Computer Science, Interdisciplinary Applications
Nitzan Avidan, Moti Freiman
Summary: This study aims to enhance the generalization capabilities of DNN-based MRI reconstruction methods for undersampled k-space data. By introducing a mask-aware DNN architecture and training method, the under-sampled data and mask are encoded within the model structure, leading to improved performance. Rigorous testing on the widely accessible fastMRI dataset reveals that this approach demonstrates better generalization capabilities and robustness compared to traditional DNN methods.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2024)
Article
Computer Science, Interdisciplinary Applications
Enhao Zhang, Saeed Miramini, Lihai Zhang
Summary: This study investigates the combined effects of osteoporosis and diabetes on fracture healing process by developing numerical models. The results show that osteoporotic fractures have higher instability and disruption in mesenchymal stem cells' proliferation and differentiation compared to non-osteoporotic fractures. Moreover, when osteoporosis coexists with diabetes, the healing process of fractures can be severely impaired.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2024)
Article
Computer Science, Interdisciplinary Applications
Yunhao Bai, Wenqi Li, Jianpeng An, Lili Xia, Huazhen Chen, Gang Zhao, Zhongke Gao
Summary: This study proposes an effective MIL method for classifying WSI of esophageal cancer. The use of self-supervised learning for feature extractor pretraining enhances feature extraction from esophageal WSI, leading to more robust and accurate performance. The proposed framework outperforms existing methods, achieving an accuracy of 93.07% and AUC of 95.31% on a comprehensive dataset of esophageal slide images.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2024)