Inthis study, we proposed a closed-loop transformer neural network decoder-based signal generation method to record, decode, and stimulate the brain to intercept a previously functional brain area. The dataset consists of the auditory cortex output of computational modeling of the human auditory system, following voice input. The system is comprised of three phases. In Phase I artificial brain signal was generated based on a generative sequence-to-sequence machine learning model. In Phase II the cortical area downstream was stimulated with the generated brain signal, effectively replacing the upstream cortical area on which the machine learning model is based. Finally, in Phase III, incremental learning is applied to fine-tune the model to unlock differences between what is being recorded by our electrodes and the hidden neural patterns of the brain. As an evaluation, we computed RMSE and similarity index between the generated stimulation output and inputted original computational model output without any interception. The generative regression model using our proposed method shows an average RMSE = 0.009845, and similarity index (SI) = 0.9416 across the models without an word embedding layer, and SI = 0.855283 and RMSE = 0.033068 for the model with a word embedding layer. This could be evidence to support that the Transformer-Decoder model has extracted distinguishable features during heard speech.
The diagnosis of the western medicine often relies on intricate and expensive equipment. Instead, Traditional Chinese Medicine (TCM) relies on four methods and one of them (inspection) investigates various visual features to understand the patient's health condition. A lot of work were proposed to digitize the process and the first step is to segment the tongue from a given photo. In this paper, we present DoggyTongue, a system to automate this tongue segmentation process. The system contains a mobile application for taking photos and presenting segmented results, and a backend server with machine learning algorithms to perform tongue segmentation. The combination of a mobile application and machine learning algorithms not only has a potential to support TCM diagnosis, but it also allows the public to easily use it. The results of our work have shown a promising direction to apply mobile and web engineering into the TCM domain.
With the burgeoning prevalence of diabetes and the urgent imperative for effective dietary management, harnessing the power of image analysis and computer vision technology presents a promising solution to the automatic nutrition estimation challenge. However, constructing a comprehensive frame-work for this endeavor presents its own set of challenges, encompassing multiple intricate procedures, including data collection, image segmentation, food recognition, and volume estimation. In this paper, we introduce a sophisticated and automated framework for Chinese food nutrition estimation, leveraging food images sourced from diabetic patients. In addition to developing models for the various procedures, the framework necessitates label images for model training. These label images encompass ground truth masks delineating precise object boundaries for image segmentation, food type labels for food recognition, and comprehensive nutrition tables for each food item. Through rigorous experimentation and validation, our framework has demonstrated its efficacy, offering a convenient and practical tool for managing dietary requirements in diabetes care.
Lithium-ion batteries are widely used in portable devices and mobile medical equipment due to their high energy density and long cycle life. However, long charging times for lithium-ion batteries can limit their usability. This paper proposes a multi-stage constant current charging protocol.
The developments in eHealth technology have driven a significant shift towards increased digitalization in healthcare, transforming eHealth from standalone tools into intricate infrastructures of multiple interconnected data systems. These infrastructures introduce multifaceted challenges in legal, ethical, financial, and technological (LEFT) domains, highlighting the critical need for a comprehensive framework to navigate the complexities within these domains and facilitate successful implementation. This viewpoint paper presents version 1.0 of a roadmap for eHealth implementation concerning LEFT domains and provides a preliminary overview of key preconditions essential for successful implementation. The roadmap is constructed based on the LEFT barriers and facilitators identified in an earlier study, systematically translating these insights into specific preconditions for successful implementation. Legal preconditions emphasize the necessity of effective data privacy and security measures, along with a comprehensive regulatory compliance framework. Ethical preconditions focus on facilitating patient understanding and control, enhancing consent processes, and implementing robust validations mechanisms for eHealth technologies. Financial preconditions advocate for fostering public-private stakeholder collaboration, establishing a financially sustainable business model, and developing eHealth reimbursement strategies. Technological preconditions include promoting interoperability between different data systems, enhancing data and system quality, and improving usability and accessibility of eHealth technologies. While this paper introduces version 1.0 of the roadmap and outlines the LEFT preconditions for eHealth implementation, effective implementation also requires collaborative decision-making among multidisciplinary stakeholders. Recognizing the need for a holistic approach, these preconditions necessitate careful consideration and systematic planning, with the roadmap serving as guidance to address the challenges of creating a successful eHealth infrastructure.
Problematic smartphone use (PSU) is a social issue affecting daily lives without a definitive treatment. Current PSU treatments rely on self-reported data, which can be inaccurate and lead to ineffective treatments. To address this, we propose using a hidden Markov model to objectively analyze smartphone usage patterns from log data. The model is trained on data from various patients and then individualized, allowing for meaningful interpretation and comparison of usage states. We conduct two case studies on patients attending outpatient clinics for Internet addiction due to PSU. These case studies compare the model-estimated usage state of patients with 1) a daily activity log as reported by the patients and 2) a several-month clinical history of the patients as reported by their attending psychiatrists. Through each case study, we validate the appropriateness of the definitions assigned to each usage state and the effectiveness of the state estimations performed using the proposed method. Our approach demonstrated that it is possible to objectively understand and track changes in PSU, which self-reports alone cannot achieve.
With the growing importance of privacy in data-driven applications, ensuring the security and confidentiality of personal information has become a significant challenge. Federated Learning (FL) offers a promising solution by enabling collaborative model training across multiple clients while keeping individual data localized and private. In FL, outputs computed by various devices are aggregated at a central server, which uses iterative algorithms to develop a globally shared model. However, the presence of malicious participants can result in the intentional manipulation of training data or the model, compromising the system's accuracy and reliability. In this study, we propose a novel FL technique based on Split Learning (SL) to enhance robustness against data poisoning attacks. Our approach aims to develop a robust FL system based on SL, integrate it with existing aggregation methods, and compare its performance with traditional FL approaches.
Internet of Things (IoT) is crucial for the hierarchical medical system, and enables the real-time monitoring and collection of data, thereby improving patient treatment outcomes. However, achieving secure, efficient, timely, and controllable medical IoT data sharing between higher-lever hospital (HLH) and lower-level hospital (LLH) is a challenging task for the hierarchical medical system. Consortium blockchain, which is an effective way to achieve secure and trustworthy data sharing, has the potential to address these issues. In this article, we propose a novel cloud-chain sharing scheme for medical IoT data based on consortium blockchain. HLH and LLH establish a consortium blockchain, where medical IoT data is stored both on-chain and off-chain. On-chain data adopt a proxy re-encryption based on elliptic curve cryptography (ECC-PRE) strategy and attribute-based strategy to facilitate secure access and controlled sharing of data. Off-chain data sharing provides three different modes: private data collection (PDC), direct channel, and cloud storage (CS), according to the urgency of patient and the sensitivity of the data. Furthermore, a file security breakpoint resume scheme, rooted in the consortium blockchain, and a file weighting strategy are employed to enhance the efficiency and timeliness of data sharing. Finally, the performance of our proposed solution is verified by experimental results, and the results demonstrate our solution is feasible and efficient. In future work, we plan to use searchable encryption technology to make this scheme more versatile and gradually implement dynamic adjustment of permissions.
This study aims to address the problem of medical text data privacy protection in the context of extensive development of cloud computing. This paper comprehensively researches the relevant literature and technical progress in recent years on privacy protection in medical scenarios using homomorphic encryption and other advanced cryptography techniques, and synthesizes the scalability of cloud computing, the timeliness of edge computing, and the inerrancy of blockchain technology to achieve the privacy protection of medical text data through the cloud storage of medical data and the dense state computation, and protects the privacy of patient's medical data in order to achieve the purpose of protecting individual privacy, property security, and security. It protects the privacy of patients' medical data to realize the purpose of protecting personal privacy and property security. By changing the encryption parameters of the existing DGHV multi-key homomorphic encryption scheme to realize the characteristics of the adapted scheme, and at the same time, the public key parity restriction can be eliminated. In addition, this paper also discusses the application of blockchain technology in homomorphic encryption key management, and proposes a blockchain-based key security storage system to ensure the key's tamperability and security. The privacy security problem in medical data transmission, storage and processing is solved by constructing this medical text data privacy protection system based on cloud-edge cooperative computing and homomorphic encryption. Simulation experiments verify that this scheme is practically feasible, while the correct parameters and the choice of the edge computing approach take into account both security and feasibility.
Accurate classification of brain tumor images is crucial for accurate prognosis and effective treatment planning. This study involved an analysis of the application of machine learning in the classification of brain tumors. We employed segmentation using mU-Net, as well as feature extraction and feature selection techniques to identify the most prominent features. We identified 14 significant features using the Information Gain algorithm to classify brain tumors from 33 available features. The K-NN, SVM, Decision Tree, and Random Forest classification models were employed, and the SVM model yielded the highest performance with an accuracy of 0.929.
Recently, there has been a surge in cyberattacks targeting the Internet of Health Things (IoHT), increasing the urgency for advancing network intrusion detection systems (IDS). Machine learning techniques, especially deep neural networks (DNNs), are demonstrating potential in improving the precision of detection methods. Despite their advantages, the complexity of DNNs can obscure their decision-making process, impacting their acceptance in security-critical environments. To improve the transparency of DNN models, we propose a novel post-hoc interpretation method that applies a perturbation-based approach with an optimized mask applied to an autoencoder model for IDS. More importantly, we leverage contrastive learning to maintain perturbed samples within the original feature space, reducing the risk of misclassification due to sample drift and ensuring a clear interpretation of the mask. We validate our approach using the NSL-KDD and UNSW15 datasets, showing that it provides clearer and more robust explanations compared to existing methods. This enhancement in interpretability is pivotal for healthcare cybersecurity experts to gain insights into the decision-making processes of black-box models.
Urinary tract infections (UTIs) are a significant health concern, particularly for people living with dementia (PLWD), as they can lead to severe complications if not detected and treated early. This study builds on previous work that utilised machine learning (ML) to detect UTIs in PLWD by analysing in-home activity and physiological data collected through low-cost, passive sensors. The current research focuses on improving the performance of previous models, particularly by refining the Multilayer Perceptron (MLP), to better handle variations in home environments and improve sex fairness in predictions by making use of concepts from multitask learning. This study implemented three primary model designs: feature clustering, loss-dependent clustering, and participant ID embedding which were compared against a baseline MLP model. The results demonstrated that the loss-dependent MLP achieved the most significant improvements, increasing validation precision from 48.92 sensitivity from 27.44 sexes. These findings suggest that the refined models offer a more reliable and equitable approach to early UTI detection in PLWD, addressing participant-specific data variations and enabling clinicians to detect and screen for UTI risks more effectively, thereby facilitating earlier and more accurate treatment decisions.
In the context of healthcare for elderly patients, especially in areas located far from rehabilitation clinics, pro-viding remotely assisted prevention and treatment enabled by reliable and secure connectivity through micro-networks based on technologies like 5G and beyond, could be a viable alternative to regularly transporting the elderly to the clinics, potentially improving their quality of life. This paper analyses the needs of such a system featuring fall detection and risk assessment as well as rehabilitation at home and presents the design of the End-to- End interoperable and secure software system and the planned validation.
This paper presents a cooperative project, supported by the German government, between the University of Wuppertal and a company in the medical sector. The paper includes the construction of the drive unit and explains the main features of the mechanics and electronics being developed. These pumps are designed for a niche between medical devices and consumer hardware. To enable mobile usage, the new pump must be lighter and smaller than the existing ones. It should also have a water-proof housing for faster disinfection. The pump is designed so that both breasts can be treated individually. An integrated touch display and Bluetooth Low-Energy connectivity shall provide an easy operation and documentation. For predictive maintenance aspects the pump transmits data to a server via LTE.
This paper presents a comprehensive approach for detecting respiratory diseases using IoMTs and Machine Learning (ML) algorithms, leveraging audio recordings from multiple sensor locations on the body. By capturing, extracting, and analyzing diverse audio features, such as MFCC, STFT, and Mel-spectrogram, we aim to detect and classify six respi-ratory conditions: Bronchiectasis, Bronchiolitis, COPD, Healthy, Pneumonia, and URTI. We used a publicly annotated dataset to conduct the experiment and analyze the performance of our proposed approach. This dataset underwent preprocessing, which included feature extraction, removal of rare diseases, data flattening, and encoding for model training. Our findings demonstrate that Deep Learning (DL), such as the Convolutional Neural Network (CNN) model achieved the highest accuracy of 92.4 % and an AU C of 97 %, highlighting its potential in audio-based diagnostics. Our experimental results prove that DL, particularly CNN, outperforms traditional ML techniques in detection accuracy, which makes them a good choice in developing non-invasive, efficient, and cost-effective solutions for respiratory disease detection.
Skin cancer is a prevalent and potentially fatal disease that requires early detection for effective treatment. We trained and evaluated five YOLOv8 classification model variants (YOLOv8n-cls, YOLOv8s-cls, YOLOv8m-cls, YOLOv8l-cls, and YOLOv8x-cls) on the HAM10000 dataset, which contains 10,015 dermatoscopic images of common pigmented skin lesions. The models were trained for 30 epochs using data augmentation techniques to enhance generalization. Performance was assessed using metrics including accuracy, precision, recall, F1-score, and inference time. The YOLOv8x-cls model achieved the highest accuracy of 86.2% and precision of 82.1%, while the YOLOv81-cIs model demonstrated the best balance with the highest F1-score of 77.0%. Compared to previous ensemble approaches, our single YOLOv8 models achieved superior performance with lower computational overhead. The YOLOv8n-cls variant showed the fastest inference time of 0.5 ms, making it suitable for real-time applications. Our results demonstrate the potential of YOLOv8-based models for accurate and efficient skin lesion classification, which could aid in early skin cancer detection and improve patient outcomes.
Machine learning (ML) and its applications have expanded over the past few years. ML inevitably makes its way to the world of prosthetics and amputees. Smart prosthetics have been studied and growing recently. Through the collection of data from these devices, results can help the user in numerous ways. On the other hand, mobile devices and applications are widely used. However, how to combine mobile applications and ML to enhance the prosthetic device was less addressed. In this research, we studied ML with an loT -based prosthetic device paired with a mobile application. The hypothesis was that the ML methods could help evaluate the user's status. The study results showed that some evaluated ML methods were able to see through the average temperature, humidity and contraction percentage of the people who wear a designed prosthetic device. The results also indicated that the users could tell if the contractions reached a concerning level.
Cardiovascular diseases (CVDs) constitute the primary cause of human mortality globally in recent decades. To effectively detect CVDs, heart auscultation plays an important role in early diagnosis. With the development of artificial intelligence (AI), many studies have designed varying AI-assisted diagnosis systems helping people discriminate abnormal heart sounds. Yet, a robust system usually requires a noise-less input signal, which is critical as heart sounds are often affected by some unavoidable noise. Therefore, many heart sound classification models use filters or other methods to obtain the clean signals. However, these classic techniques are not adaptable enough to distinguish the meaningful murmurs and real noises. Thus, we propose a novel approach to transfer an audio source separation model to denoise the heart sound. In this paper, we test different denoisers on synthesis heart sound with additive white Gaussian noises. Our method performs well on the noise reduction metrics. Meanwhile, we evaluate the classification performance of each denoiser with some classifiers on the PhysioNet dataset. Experimental results demonstrate that our method can outperform other denoising techniques by achieving the highest unweighted average recall (UAR) at 95.7% with the smallest standard deviation. The results confirm that our method is robust and adaptable in improving audio's denoising.
This paper introduces a new system for non-contact hand movement sensing, utilizing ultrasound sensors with millimeter precision. Acting akin to a sonar, the system detects pressure waves reflected from the moving hand. Due to the high spatio/temporal resolution of the system, the proposed solution has been tested for finger tapping monitoring. Finger tapping serves as a key clinical test for diagnosing neurodegenerative diseases, such as Parkinson's disease, where symptoms like bradykinesia, rigidity, and tremor are focal points of investigation. Currently, clinicians visually assess bradykinesia by observing subjects' finger tapping gestures, rating the severity qualitatively. The idea is to provide a quantitative assessment of the tapping providing a support for diagnosis. In detail, through subsequent time/frequency analysis and specialized processing, tapping peaks are identified, enabling evaluation of tapping stability by measuring interpeak distances. This feasibility study presents the prototype and initial measurement findings.
Hypophonia is a common speech symptom related to Parkinson's disease, affecting human comprehension and effective communication. Unlike dysarthric speech, hypophonic speech is characterized by its low volume and breathy voice, which makes it challenging to be heard and understood by human and voice-controllable systems especially in noisy environments. Conventional speech enhancement techniques, primarily focusing on amplifying audio power or cancelling environmental noise, fall short in improving the intelligibility and perception for hypophonic speech. To enhance hypophonic speech, we present ClearAI, an innovative AI-powered technology to improve speech quality for individuals suffering from hypophonia. ClearAI first leverages voice conversion technology to create a parallel dataset composed of normal and corresponding hypophonic speech samples. Then, ClearAI incorporates a predictive model trained on augmented parallel data to estimate the optimal audio style from hypophonic speech to strengthen the audio intensity and enhance the speech patterns. Next, a speech restoration model is built on the generated parallel speech data to reconstruct clear speech from the style transferred speech. Our experimental results reveal that ClearAI leads to substantial improvements in audio intensity in both digital formats and over-the-air transmission. In addition, ClearAI successfully reduces the hypophonic speech recognition error rate by more than 30% in noisy environments. Our human test results also validate ClearAI enhanced speech has the best human perceptual quality compared with other baseline methods.
