Fig 5 - available via license: Creative Commons Attribution 4.0 International
Content may be subject to copyright.
A taxonomy of different security threats on ML/DL models.
Source publication
Recent years have witnessed widespread adoption of machine learning (ML)/deep learning (DL) techniques due to their superior performance for a variety of healthcare applications ranging from the prediction of cardiac arrest from one-dimensional heart signals to computer-aided diagnosis (CADx) using multi-dimensional medical images. Notwithstanding...
Contexts in source publication
Context 1
... security threats on ML systems can be broadly categorized into three dimensions, i.e., influence attacks, security violations, and attack specificity [97]. A taxonomy of these security threats on ML systems is depicted in Fig. 5. a) Influence: Influence attacks can be of two types: (1) causative: the one that attempts to get control over training data; (2) exploratory: the one that exploits the missclassification of the ML model without intervening the model training. b) Security Violation: It is concerned with the availability and integrity of the services ...
Context 2
... or both. c) Attack Specificity: The specificity of an attack can be defined in two ways: (1) targeted attack: whether the attack is intended for a specific input sample or a group of samples; (2) indiscriminate attack: it causes the ML model to fail indiscriminately. The first axis in the taxonomy of the attacks on ML/DL systems (as shown in Fig. 5) defines the capabilities of the adversaries, e.g., whether they are able to modify training process by injecting poisoned data or not (i.e., attempting access to training data). If the attacker does not have access to the training data, the attacker can realize an exploratory attack, e.g., consider a disease classification problem, ...
Citations
... In recent years, the growing success of machine learning (ML) has led to its widespread deployment across a range of applications [1,58,49]. Once a machine learning model has been trained, it may be necessary to unlearn specific training data instances for various reasons, like complying with user data deletion requests [43,18,52,2], removing stale or corrupt data [5,53], etc. Retraining an ML model entirely from scratch with each deletion request is expensive, especially for modern large-scale models [9,1,58]. Machine unlearning [46,65] techniques focus on efficiently unlearning the influence of a data instance from a trained ML model. ...
Machine unlearning is the process of efficiently removing the influence of a training data instance from a trained machine learning model without retraining it from scratch. A popular subclass of unlearning approaches is exact machine unlearning, which focuses on techniques that explicitly guarantee the removal of the influence of a data instance from a model. Exact unlearning approaches use a machine learning model in which individual components are trained on disjoint subsets of the data. During deletion, exact unlearning approaches only retrain the affected components rather than the entire model. While existing approaches reduce retraining costs, it can still be expensive for an organization to retrain a model component as it requires halting a system in production, which leads to service failure and adversely impacts customers. To address these challenges, we introduce an exact unlearning framework -- Sequence-aware Sharded Sliced Training (S3T), designed to enhance the deletion capabilities of an exact unlearning system while minimizing the impact on model's performance. At the core of S3T, we utilize a lightweight parameter-efficient fine-tuning approach that enables parameter isolation by sequentially training layers with disjoint data slices. This enables efficient unlearning by simply deactivating the layers affected by data deletion. Furthermore, to reduce the retraining cost and improve model performance, we train the model on multiple data sequences, which allows S3T to handle an increased number of deletion requests. Both theoretically and empirically, we demonstrate that S3T attains superior deletion capabilities and enhanced performance compared to baselines across a wide range of settings.
... This data, when analyzed using machine learning techniques, holds significant potential for revolutionizing healthcare. Machine learning algorithms can uncover valuable insights from the vast amounts of data collected within the Metaverse environment [26][27][28]. ...
The Metaverse, a persistent, immersive virtual environment, has the immense potential to revolutionize healthcare by transforming patient care, medical education, and research. This paper explores the applications, benefits, and challenges associated with this transformative technology, highlighting its ability to improve patient engagement, communication, access to information, and health outcomes. The paper also examines how the analysis of Metaverse data using machine learning techniques can unlock insights to further enhance healthcare applications. The discussion summarizes key findings, analyzes the significance and practical implications of Metaverse integration, and identifies areas for future research. It underscores the role of major tech companies in developing Metaverse-based solutions and the importance of addressing emerging opportunities and challenges to unlock the transformative potential of this technology in healthcare. The paper concludes by emphasizing the need for collaboration between stakeholders to ensure the ethical and effective implementation of these technologies, ultimately leading to a more accessible, personalized, and efficient healthcare system.
... The low sample size can also lead to an unstable feature selection defined by an outcome specific to the experimental settings [9]. To increase robustness [10], ensemble approaches combine the strengths of several feature selection components, improving the results' stability and accuracy [11,12] through a more thorough exploration of the space of possible selections [13]. While ensemble approaches have been used for feature selection in the past, they have often been limited to simple techniques such as majority or weighted voting [14], hill climbing [15], ablation [16] or genetic algorithms [17]. ...
Motivation
A major hindrance towards using Machine Learning (ML) on medical datasets is the discrepancy between a large number of variables and small sample sizes. While multiple feature selection techniques have been proposed to avoid the resulting overfitting, overall ensemble techniques offer the best selection robustness. Yet, current methods designed to combine different algorithms generally fail to leverage the dependencies identified by their components. Here, we propose Graphical Ensembling (GE), a graph-theory-based ensemble feature selection technique designed to improve the stability and relevance of the selected features.
Results
Relying on four datasets, we show that GE increases classification performance with fewer selected features. For example, on rheumatoid arthritis patient stratification, GE outperforms the baseline methods by 9% Balanced Accuracy while relying on fewer features. We use data on sub-cellular networks to show that the selected features (proteins) are closer to the known disease genes, and the uncovered biological mechanisms are more diversified. By successfully tackling the complex correlations between biological variables, we anticipate that GE will improve the medical applications of ML.
Availability and implementation
https://github.com/ebattistella/auto_machine_learning.
... Meanwhile, societal issues involve low willingness of patients to choose community primary care and insufficient information reception, among others. 14 However, the existing digital health literature primarily delves into clinical applications and electronic healthcare, 15 such as the application of machine learning (ML)/deep learning (DL) technologies in various medical fields, 16 for instance, the use of virtual reality rehabilitation landscape (VRTL) technology in clinical treatment, 17 and the design and application of clinical decision support systems and electronic nursing records. 18 The purpose of this article is to explore the construction of an information-based bidirectional referral mechanism within the framework of a healthcare alliance model. ...
Background
Digital health technologies are progressively assuming significant roles in aspects encompassing in-hospital management, patient-centered design, and tiered referral systems. Nevertheless, current studies do not involve exploration into the potential value and mechanisms of digital health in a patient-centered context. This study aimed to explore the development of a framework of comprehensive, evidence-based digital health technologies for the construction of welfare-oriented healthcare.
Methods
From March to June 2023, a cross-sectional online study was performed, involving 335 respondents with prior referral experiences hailing from the Central China region. Data on welfare-oriented healthcare factors (clinical pathway management, medical structure configuration, healthcare service accessibility, two-way referrals) underwent factor analysis in advance, and correlation between these factors and their association with two-way referrals was evaluated by testing for direct and indirect (mediating) effects.
Results
Firstly, there existed a significant positive correlation between integrative medical indicators and welfare-centered healthcare (β = 0.02–0.16, p < 0.05). Furthermore, two-way referral had an direct association with integrative medical parameters and the welfare healthcare service system (β = 0.15–0.31, p < 0.05), but exerted a partial mediatory function in the welfare healthcare service system (β = 0.005–0.021, α < 0.05). Two-way referrals partially mediate the integrated medical indicators, mainly through direct effects, while also providing complementary support. Clinical pathways, medical structure, and accessibility are closely linked to welfare healthcare and significantly influence healthcare quality. Thus, improving these factors should be prioritized.
Conclusion
This study proposes a method combining integrated evaluation indicators with pathway mechanism design. This pathway mechanism design includes key steps such as patient registration, information extraction, hospital allocation or referral, diagnosis and treatment, rehabilitation plan monitoring, service feedback, and demand resolution. This design aims to change patients’ intentions in seeking healthcare, thereby increasing their acceptance of bidirectional referrals, and ultimately enhancing the effectiveness and realization of welfare healthcare.
... Ensemble learning is a technique used in ML to improve the performance of predictive models [20]. By combining the strengths of different models, ensemble learning can achieve better accuracy and robustness than any single model [21]. Deep Ensemble is a technique in which multiple deep neural networks are trained on different subsets of the data and the outputs are combined through averaging or voting [22]. ...
Retinal disorders, including diabetic retinopathy and macular degeneration due to aging, can lead to preventable blindness in diabetics. Vision loss caused by diseases that affect the retinal fundus cannot be reversed if not diagnosed and treated on time. This paper employs deep-learned feature extraction with ensemble learning models to improve the multi-disease classification of fundus images. This research presents a novel approach to the multi-classification of fundus images, utilizing deep-learned feature extraction techniques and ensemble learning to diagnose retinal disorders and diagnosing eye illnesses involving feature extraction, classification, and preprocessing of fundus images. The study involves analysis of deep learning and implementation of image processing. The ensemble learning classifiers have used retinal photos to increase the classification accuracy. The results demonstrate improved accuracy in diagnosing retinal disorders using DL feature extraction and ensemble learning models. The study achieved an overall accuracy of 87.2%, which is a significant improvement over the previous study. The deep learning models utilized in the study, including NASNetMobile, InceptionResNetV4, VGG16, and Xception, were effective in extracting relevant features from the Fundus images. The average F1-score for Extra Tree was 99%, while for Histogram Gradient Boosting and Random Forest, it was 98.8% and 98.4%, respectively. The results show that all three algorithms are suitable for the classification task. The combination of DenseNet feature extraction technique and RF, ET, and HG classifiers outperforms other techniques and classifiers. This indicates that using DenseNet for feature extraction can effectively enhance the performance of classifiers in the task of image classification.
... To overcome limitations and enhance the robustness of machine learning (ML) models in healthcare, various strategies have been proposed (Qayyum et, al., 2020). Firstly, optimizing algorithms for scalability is crucial to ensure that ML solutions can efficiently process large volumes of healthcare data as shown in figure 5 (Chen & Wang, 2021). ...
This review paper provides an overview of precision healthcare analytics, focusing on the integration of machine learning (ML) techniques for automating image interpretation, disease detection, and prognosis prediction across various medical imaging modalities, including X-rays, MRIs, and CT scans. Drawing upon existing literature and empirical evidence, we assess the impact of ML-driven automated image interpretation on diagnostic accuracy, highlighting its superiority over traditional methods. Additionally, we examine the effectiveness of ML algorithms in disease detection, emphasizing their potential for early intervention and improved patient outcomes. Furthermore, we explore the prognostic capabilities of ML-based models in forecasting disease progression and guiding treatment strategies. Through a comprehensive synthesis of research findings, we identify key factors influencing the performance of ML algorithms in healthcare applications and discuss strategies for addressing challenges related to data quality, interpretability, and scalability. By critically evaluating current trends and advancements in precision healthcare analytics, this review aims to provide insights into the potential benefits and limitations of ML integration in medical practice, contributing to the ongoing discourse on enhancing patient care and healthcare delivery.
... Medical data privacy can be compromised when used for ML, due to the centralized nature of data gathering. Concentrating sensitive health information in centralized databases or cloud servers poses a significant risk [10]. If these centralized repositories become targets for security breaches or unauthorized access, the potential impact is heightened. ...
Wearable devices and medical sensors revolutionize health monitoring, raising concerns about data privacy in Machine Learning (ML) for healthcare. This tutorial explores Federated Learning (FL) and Blockchain (BC) integration, offering a secure and privacy-preserving approach to healthcare analytics. FL enables decentralized model training on local devices at healthcare institutions, keeping patient data localized. This facilitates collaborative model development without compromising privacy. However, FL introduces vulnerabilities. BC, with its tamper-proof ledger and smart contracts, provides a robust framework for secure collaborative learning in FL. After presenting a taxonomy for the various types of data used in ML in medical applications, and a concise review of ML techniques for healthcare use cases, this tutorial explores three integration architectures for balancing decentralization, scalability, and reliability in healthcare data. Furthermore, it investigates how Blockchain-based Federated Learning (BCFL) enhances data security and collaboration in disease prediction, medical image analysis, patient monitoring, and drug discovery. By providing a tutorial on FL, blockchain, and their integration, along with a review of BCFL applications, this paper serves as a valuable resource for researchers and practitioners seeking to leverage these technologies for secure and privacy-preserving healthcare ML. It aims to accelerate advancements in secure and collaborative healthcare analytics, ultimately improving patient outcomes.
... Throughout the past years, machine learning has been used excessively in several problems including ecommerce (Rath, 2022), sports (Richter et al., 2021), and healthcare (Qayyum et al., 2020). Time series forecasting is also one of the main areas for machine learning algorithms (Ahmed et al., 2010), because efficient forecasting may lead to better trading returns and enhance utilization of healthcare infrastructure. ...
... → dem Management des Datenschutzes, → der "Robustheit" von ML in Bezug auf die Datenauswertung und → der AKZEPTANZ von e-Health aus Sicht der PatientInnen (Alt et al., 2023) *MARS = Mobile App Rating Scale (Reliabilität = Omega 0,79 -0,93; ICC = 0.82) (Terhorst et al., 2020) Abb. Entwicklung von eHealth, basierend auf Qayyum et al., 2021; ML = machine learning; LTSM = long short term memory (rekurrentes neuronales Netz) → lernt und speichert Informationen ...
... • Schwachstellen bei der Datenerhebung (Egger, et al., 2021), z.B. unsaubere Dokumentation und Extraktion von PatientInnen-Daten • Schwachstellen aufgrund von Datenverarbeitung (Liu et al., 2021), z.B. Uneinigkeit bei der Interpretation von Symptomen und Diagnosen -Problem, wenn es viele sind • Schwachstellen in der Einführungsphase → → Robustheit des Systems (Qayyum et al., 2021) ...
This presentation aims to explain the complexity of environmental healthcare and physiotherapy in particular.
It is not only patients who suffer from ineffective treatment methods, but society as a whole. Constantly rising health insurance premiums, primary care that is reminiscent of industrial assembly line production, and digitalization that defies data protection are at the forefront of this misery.
But that's not all, because the consequences of these developments are also having an impact on the environment, which is increasingly feeling the effects of our ecological footprint and confronting us with it.
It is time to turn lip service into action...
- from: IM FALSCHEN GESUNDHEITSSYSTEM DAS RICHTIGE TUN -
... 3) Existing AI solutions: The last few years have witnessed ubiquitous utilization of ML algorithms and DL architectures for a variety of healthcare and medical applications [157], such as physical activity recognition with time-series sensory data and diabetic retinopathy recognition with multimodal images. The authors in [158] proposed an intermediate fusion framework for human activity recognition using sensory data of wearable devices, in which the deep local features extracted by a deep convolutional network were combined with descriptive statistic features to improve the recognition rate. ...
Around 2020, 5G began its commercialization journey, and discussions about the next-generation networks (such as 6G) emerged. Researchers predict that 6G networks will have higher bandwidth, coverage, reliability, energy efficiency, and lower latency, and will be an integrated ``human-centric" network system powered by artificial intelligence (AI). This 6G network will lead to many real-time automated decisions, ranging from network resource allocation to collision avoidance for self-driving cars. However, there is a risk of losing control over decision-making due to the high-speed, data-intensive AI decision-making that may go beyond designers' and users' comprehension. To mitigate this risk, explainable AI (XAI) methods can be used to enhance the transparency of the black-box AI decision-making process. This paper surveys the application of XAI towards the upcoming 6G age, including 6G technologies (such as intelligent radio and zero-touch network management) and 6G use cases (such as industry 5.0). Additionally, the paper summarizes the lessons learned from recent attempts and outlines important research challenges in applying XAI for 6G use cases soon.
