Journal of Medical Systems最新文献

Hospitals around the world are deploying increasingly advanced systems to collect and store high-resolution physiological patient data for quality improvement and research. However, data accuracy, completeness, consistency, and contextual validity remain issues. This report highlights a data artifact known as waveform clipping in our hospital's physiological data capture system that went unnoticed for years, limiting data analysis and delaying several research projects. We aim to raise awareness in the medical informatics community about the importance of careful system setup, ongoing data validation, and close cooperation between clinicians and data scientists.

Feedback on cognitive workload may reduce decision-making mistakes. Machine learning-based models can produce feedback from physiological data such as electroencephalography (EEG) and electrocardiography (ECG). Supervised machine learning requires large training data sets that are (1) relevant and decontaminated and (2) carefully labeled for accurate approximation, a costly and tedious procedure. Commercial over-the-counter devices are low-cost resolutions for the real-time collection of physiological modalities. However, they produce significant artifacts when employed outside of laboratory settings, compromising machine learning accuracies. Additionally, the physiological modalities that most successfully machine-approximate cognitive workload in everyday settings are unknown. To address these challenges, a first-ever hybrid implementation of feature selection and self-supervised machine learning techniques is introduced. This model is employed on data collected outside controlled laboratory settings to (1) identify relevant physiological modalities to machine approximate six levels of cognitive-physical workloads from a seven-modality repository and (2) postulate limited labeling experiments and machine approximate mental-physical workloads using self-supervised learning techniques.

In recent years, Electronic health records (EHR) has gradually become the mainstream in the healthcare field. However, due to the fact that EHR systems are provided by different vendors, data is dispersed and stored, which leads to the phenomenon of data silos, making medical information too fragmented and bringing some challenges to current medical services. Therefore, in view of the difficulties in sharing EHR between medical institutions, the risk of privacy leakage, and the lack of EHR usage control by patients, an EHR sharing model based on consortium blockchain is proposed in this paper. Firstly, the Interplanetary File System is combined with consortium blockchain, which forms a hybrid storage scheme of EHR, this technology effectively improves data security, privacy protection, and operational efficiency. Secondly, the model combines unidirectional multi-hop conditional proxy re-encryption based on type and identity with distributed key generation technology to achieve secure EHR sharing with fine grained control. At the same time, users are required to link the operation records of EHR, so as to realize the traceability of EHR usage. A dynamic Byzantine fault-tolerant algorithm based on reputation and clustering is then proposed to solve the problems of arbitrary master node selection, high latency and low throughput of PBFT, enabling the nodes to reach consensus more efficiently. Finally, the model is analyzed in terms of security and user control, showing that the model is less energy intensive in terms of communication overhead and time consumption, and can effectively achieve secure sharing between medical data.

In this issue of Journal of Medical Systems, Neri et al. share results from their study in which they compared the YouCare device to a standard Holter monitor. The wearable used in the study incorporates a single electrocardiogram lead in a crop top garment that is customized for each patient. This editorial discusses the YouCare device, the study findings, and their clinical relevance and impact in the context of wearable technology.

This research evaluates the readability and quality of patient information material about female urinary incontinence (fUI) in ten popular artificial intelligence (AI) supported chatbots. We used the most recent versions of 10 widely-used chatbots, including OpenAI's GPT-4, Claude-3 Sonnet, Grok 1.5, Mistral Large 2, Google Palm 2, Meta's Llama 3, HuggingChat v0.8.4, Microsoft's Copilot, Gemini Advanced, and Perplexity. Prompts were created to generate texts about UI, stress type UI, urge type UI, and mix type UI. The modified Ensuring Quality Information for Patients (EQIP) technique and QUEST (Quality Evaluating Scoring Tool) were used to assess the quality, and the average of 8 well-known readability formulas, which is Average Reading Level Consensus (ARLC), were used to evaluate readability. When comparing the average scores, there were significant differences in the mean mQEIP and QUEST scores across ten chatbots (p = 0.049 and p = 0.018). Gemini received the greatest mean scores for mEQIP and QUEST, whereas Grok had the lowest values. The chatbots exhibited significant differences in mean ARLC, word count, and sentence count (p = 0.047, p = 0.001, and p = 0.001, respectively). For readability, Grok is the easiest to read, while Mistral is highly complex to understand. AI-supported chatbot technology needs to be improved in terms of readability and quality of patient information regarding female UI.

Stroke registries are tools for improving care and advancing research. We aim to describe the methodology and resourcing of existing national stroke registries. We conducted a systematic search of the published, peer-reviewed literature and grey literature examining descriptions of data collection methods and resourcing of national stroke registries published from 2012 to 2023. The systematic review was registered in PROSPERO (CRD42023393841). 101 records relating to 21 registries in 19 countries were identified. They universally employed web-based platforms for data collection. The principal profession of data collectors was nursing. All included the acute phase of care, 28% (6) registered the pre-hospital (ambulance) phase and 14% (3) included rehabilitation. 80% (17) collected outcome data. The registries varied in their approach to outcome data collection; in 9 registries it was collected by hospitals, in 2 it was collected by the registry, and 7 had linkage to national administrative databases allowing follow-up of a limited number of end points. Coverage of the total number of strokes varies from 6 to 95%. Despite widespread use of Electronic Health Records (EHRs) the ability to automatically populate variables remained limited. Governance and management structures are diverse, making it challenging to compare their resourcing. Data collection for clinical registries requires time and necessary skills and imposes a significant administrative burden on the professionals entering data. We highlight the role of clinical registries as powerful instruments for quality improvement. Future work should involve creating a central repository of stroke registries to enable the development of new registries and facilitate international collaboration.

In recent years a significant demand to develop computer-assisted diagnostic tools to assess prostate cancer using whole slide images has been observed. In this study we develop and validate a machine learning system for cancer assessment, inclusive of detection of perineural invasion and measurement of cancer portion to meet clinical reporting needs. The system analyses the whole slide image in three consecutive stages: tissue detection, classification, and slide level analysis. The whole slide image is divided into smaller regions (patches). The tissue detection stage relies upon traditional machine learning to identify WSI patches containing tissue, which are then further assessed at the classification stage where deep learning algorithms are employed to detect and classify cancer tissue. At the slide level analysis stage, entire slide level information is generated by aggregating all the patch level information of the slide. A total of 2340 haematoxylin and eosin stained slides were used to train and validate the system. A medical team consisting of 11 board certified pathologists with prostatic pathology subspeciality competences working independently in 4 different medical centres performed the annotations. Pixel-level annotation based on an agreed set of 10 annotation terms, determined based on medical relevance and prevalence, was created by the team. The system achieved an accuracy of 99.53% in tissue detection, with sensitivity and specificity respectively of 99.78% and 99.12%. The system achieved an accuracy of 92.80% in classifying tissue terms, with sensitivity and specificity respectively 92.61% and 99.25%, when 5x magnification level was used. For 10x magnification, these values were respectively 91.04%, 90.49%, and 99.07%. For 20x magnification they were 84.71%, 83.95%, 90.13%.