Artificial intelligence in medicine: mitigating risks and maximizing benefits via quality assurance, quality control, and acceptance testing.

The adoption of artificial intelligence (AI) tools in medicine poses challenges to existing clinical workflows. This commentary discusses the necessity of context-specific quality assurance (QA), emphasizing the need for robust QA measures with quality control (QC) procedures that encompass (1) acceptance testing (AT) before clinical use, (2) continuous QC monitoring, and (3) adequate user training. The discussion also covers essential components of AT and QA, illustrated with real-world examples. We also highlight what we see as the shared responsibility of manufacturers or vendors, regulators, healthcare systems, medical physicists, and clinicians to enact appropriate testing and oversight to ensure a safe and equitable transformation of medicine through AI.

Task design for crowdsourced glioma cell annotation in microscopy images.

Crowdsourcing has been used in computational pathology to generate cell and cell nuclei annotations for machine learning. Herein, we broaden its scope to the previously unsolved challenging task of glioma cell detection. This requires multiplexed immunofluorescence microscopy due to diffuse invasiveness and exceptional similarity between glioma cells and reactive astrocytes. In four pilot experiments, we iteratively developed a task design enabling high-quality annotations by crowdworkers on Amazon Mechanical Turk. We applied majority or weighted vote and validated them against ground truth in the final setting. On the base of a YOLO convolutional neural network architecture, we used these consensus labels for training with different image representations regarding colors, intensities, and immmunohistochemical marker combinations. A crowd of 712 workers defined aggregated point annotations in 235 images with an average [Formula: see text] score of 0.627 for majority vote. The networks resulted in acceptable [Formula: see text] scores up to 0.69 for YOLOv8 on average and indicated first evidence for transferability to images lacking tumor markers, especially in IDH-wildtype glioblastoma. Our work confirms feasibility of crowdsourcing to generate labels suitable for training of machine learning tools in the challenging and clinically relevant use case of glioma microenvironment.

Fake paper identification in the pool of withdrawn and rejected manuscripts submitted to Naunyn-Schmiedeberg's Archives of Pharmacology.

Honesty of publications is fundamental in science. Unfortunately, science has an increasing fake paper problem with multiple cases having surfaced in recent years, even in renowned journals. There are companies, the so-called paper mills, which professionally fake research data and papers. However, there is no easy way to systematically identify these papers. Here, we show that scanning for exchanged authors in resubmissions is a simple approach to detect potential fake papers. We investigated 2056 withdrawn or rejected submissions to Naunyn-Schmiedeberg's Archives of Pharmacology (NSAP), 952 of which were subsequently published in other journals. In six cases, the stated authors of the final publications differed by more than two thirds from those named in the submission to NSAP. In four cases, they differed completely. Our results reveal that paper mills take advantage of the fact that journals are unaware of submissions to other journals. Consequently, papers can be submitted multiple times (even simultaneously), and authors can be replaced if they withdraw from their purchased authorship. We suggest that publishers collaborate with each other by sharing titles, authors, and abstracts of their submissions. Doing so would allow the detection of suspicious changes in the authorship of submitted and already published papers. Independently of such collaboration across publishers, every scientific journal can make an important contribution to the integrity of the scientific record by analyzing its own pool of withdrawn and rejected papers versus published papers according to the simple algorithm proposed in the present paper.

Assessment of Driver's Stress using Multimodal Biosignals and Regularized Deep Kernel Learning.

In this work, we classify the stress state of car drivers using multimodal physiological signals and regularized deep kernel learning. Using a driving simulator in a controlled environment, we acquire electrocardiography (ECG), electrodermal activity (EDA), photoplethysmography (PPG), and respiration rate (RESP) from N = 10 healthy drivers in experiments of 25min duration with different stress states (5min resting, 10min driving, 10min driving + answering cognitive questions). We manually remove unusable segments and approximately 4h of data remain. Multimodal time and frequency features are extracted and employed to regularized deep kernel machine learning based on a fusion framework. Task-specific representations of different physiological signals are combined using intermediate fusion. Subsequently, the fused multimodal features are fed a support vector machine (SVM) and a random forest (RF) for stress classification. The experimental results show that the proposed approach can discriminate between stress states. The combination of PPG and ECG using RF as classifier yields the highest F1-score of 0.97 in the test set. PPG only and RF yield a maximum F1-score of 0.90. Furthermore, subject-specific cross-validation improves performance. ECG and PPG signals are reliable in classifying the stress state of a car driver. In summary, the proposed framework could be extended to real-time stress state assessment in driving conditions.

Crowdsourcing image segmentation for deep learning: integrated platform for citizen science, paid microtask, and gamification.

OBJECTIVES: Segmentation is crucial in medical imaging. Deep learning based on convolutional neural networks showed promising results. However, the absence of large-scale datasets and a high degree of inter- and intra-observer variations pose a bottleneck. Crowdsourcing might be an alternative, as many non-experts provide references. We aim to compare different types of crowdsourcing for medical image segmentation. METHODS: We develop a crowdsourcing platform that integrates citizen science (incentive: participating in the research), paid microtask (incentive: financial reward), and gamification (incentive: entertainment). For evaluation, we choose the use case of sclera segmentation in fundus images as a proof-of-concept and analyze the accuracy of crowdsourced masks and the generalization of learning models trained with crowdsourced masks. RESULTS: The developed platform is suited for the different types of crowdsourcing and offers an easy and intuitive way to implement crowdsourcing studies. Regarding the proof-of-concept study, citizen science, paid microtask, and gamification yield a median F-score of 82.2, 69.4, and 69.3 % compared to expert-labeled ground truth, respectively. Generating consensus masks improves the gamification masks (78.3 %). Despite the small training data (50 images), deep learning reaches median F-scores of 80.0, 73.5, and 76.5 % for citizen science, paid microtask, and gamification, respectively, indicating sufficient generalizability. CONCLUSIONS: As the platform has proven useful, we aim to make it available as open-source software for other researchers.

Digital Emergency Management for a Complex One Health Landscape: the Need for Standardization, Integration, and Interoperability.

OBJECTIVE: Planning reliable long-term planning actions to handle disruptive events requires a timely development of technological infrastructures, as well as the set-up of focused strategies for emergency management. The paper aims to highlight the needs for standardization, integration, and interoperability between Accident & Emergency Informatics (A&EI) and One Digital Health (ODH), as fields capable of dealing with peculiar dynamics for a technology-boosted management of emergencies under an overarching One Health panorama. METHODS: An integrative analysis of the literature was conducted to draw attention to specific foci on the correlation between ODH and A&EI, in particular: (i) the management of disruptive events from private smart spaces to diseases spreading, and (ii) the concepts of (health-related) quality of life and well-being. RESULTS: A digitally-focused management of emergency events that tackles the inextricable interconnectedness between humans, animals, and surrounding environment, demands standardization, integration, and systems interoperability. A consistent and finalized process of adoption and implementation of methods and tools from the International Standard Accident Number (ISAN), via findability, accessibility, interoperability, and reusability (FAIR) data principles, to Medical Informatics and Digital Health Multilingual Ontology (MIMO) - capable of looking at different approaches to encourage the integration between the ODH framework and the A&EI vision, provides a first answer to these needs. CONCLUSIONS: ODH and A&EI look at different scales but with similar goals for converging health and environmental-related data management standards to enable multi-sources, interdisciplinary, and real-time data integration and interoperability. This allows holistic digital health both in routine and emergency events.

Machine and Deep Learning Dominate Recent Innovations in Sensors, Signals and Imaging Informatics.

OBJECTIVES: This review presents research papers highlighting notable developments and trends in sensors, signals, and imaging informatics (SSII) in 2022. METHOD: We performed a bibliographic search in PubMed combining Medical Subject Heading (MeSH) terms and keywords to create particular queries for sensors, signals, and imaging informatics. Only papers published in journals containing greater than three articles in the search query were considered. Using a three-point Likert scale (1 = not include, 2 = perhaps include, 3 = include), we reviewed the titles and abstracts of all database results. Only articles that scored three times Likert scale 3, or two times Likert scale 3, and one time Likert scale 2 were considered for full paper review. On this pre-selection, only papers with a total of at least eight points of the three section co-editors were considered for external review. Based on the external reviewers, we selected the top two papers representing significant research in SSII. RESULTS: Among the 469 returned papers published in 2022 in the various areas of SSII, 90, 31, and 348 papers for sensors, signals, and imaging informatics, and then, the full review process selected the two best papers. From the 469 papers, the section co-editors identified 29 candidate papers with at least 8 Likert points in total, of which 9 were nominated as the best contributions after a full paper assessment. Five external reviewers evaluated the nominated papers, and the two highest-scoring papers were selected based on the overall scores of all external reviewers. A consensus of the International Medical Informatics Association (IMIA) Yearbook editorial board finally approved the nominated papers. Machine and deep learning-based techniques continue to be the dominant theme in this field. CONCLUSIONS: Sensors, signals, and imaging informatics is a dynamic field of intensive research with increasing practical applications to support medical decision-making on a personalized basis.

Robust in-vehicle heartbeat detection using multimodal signal fusion.

A medical check-up during driving enables the early detection of diseases. Heartbeat irregularities indicate possible cardiovascular diseases, which can be determined with continuous health monitoring. Therefore, we develop a redundant sensor system based on electrocardiography (ECG) and photoplethysmography (PPG) sensors attached to the steering wheel, a red, green, and blue (RGB) camera behind the steering wheel. For the video, we integrate the face recognition engine SeetaFace to detect landmarks of face segments continuously. Based on the green channel, we derive colour changes and, subsequently, the heartbeat. We record the ECG, PPG, video, and reference ECG with body electrodes of 19 volunteers during different driving scenarios, each lasting 15 min: city, highway, and countryside. We combine early, signal-based late, and sensor-based late fusion with a hybrid convolutional neural network (CNN) and integrated majority voting to deliver the final heartbeats that we compare to the reference ECG. Based on the measured and the reference heartbeat positions, the usable time was 51.75%, 58.62%, and 55.96% for the driving scenarios city, highway, and countryside, respectively, with the hybrid algorithm and combination of ECG and PPG. In conclusion, the findings suggest that approximately half the driving time can be utilised for in-vehicle heartbeat monitoring.

Robust in-vehicle respiratory rate detection using multimodal signal fusion.

Continuous health monitoring in private spaces such as the car is not yet fully exploited to detect diseases in an early stage. Therefore, we develop a redundant health monitoring sensor system and signal fusion approaches to determine the respiratory rate during driving. To recognise the breathing movements, we use a piezoelectric sensor, two accelerometers attached to the seat and the seat belt, and a camera behind the windscreen. We record data from 15 subjects during three driving scenarios (15 min each) city, highway, and countryside. An additional chest belt provides the ground truth. We compare the four convolutional neural network (CNN)-based fusion approaches: early, sensor-based late, signal-based late, and hybrid fusion. We evaluate the performance of fusing for all four signals to determine the portion of driving time and the signal combination. The hybrid algorithm fusing all four signals is most effective in detecting respiratory rates in the city ([Formula: see text]), highway ([Formula: see text]), and countryside ([Formula: see text]). In summary, 60% of the total driving time can be used to measure the respiratory rate. The number of signals used in the multi-signal fusion improves reliability and enables continuous health monitoring in a driving vehicle.

Technological Assessment of Smart Wearables and 5G-Integrated Edge Computing for Real-Time Health Monitoring.

Smart wearables advance to reliably and continuously measure vital signs. Analyzing the produced data requires complex algorithms, which would unreasonably increase the energy consumption of mobile devices and exceed their computing power. Fifth-generation (5G) mobile networks provide low latencies, high bandwidth, and many connected devices and introduced multi-access edge computing, which brings high computation power close to the clients. We propose an architecture for evaluating smart wearables in real-time and evaluate it exemplary with electrocardiography signals and binary classification of myocardial infarctions. Our solution shows that real-time infarct classification is feasible with 44 clients and secured transmissions. Future releases of 5G will increase real-time capability and enable capacity for more data.

Data Sharing Platform for MIMIC-IV and MIMIC-ED Data Marts: Designing a Data Retrieving System Based on the Intra-Hospital Patient Transfer Pathway.

Accessibility to high-quality historical data for patients in hospitals may facilitate related predictive model development and data analysis experiments. This study provides a design for a data-sharing platform based on all possible criteria for Medical Information Mart for Intensive Care (MIMIC) IV and Emergency MIMIC-ED. Tables containing columns of medical attributions and outcomes were studied by a team of 5 experts in Medical Informatics. They completely agreed about the columns connection using subject-id, HDM-id, and stay-id as foreign keys. The tables of two marts were considered in the intra-hospital patient transfer path with various outcomes. Using the constraints, queries were generated and applied to the backend of the platform. The suggested user interface was drawn to retrieve records based on various entry criteria and present the output in the frame of a dashboard or a graph. This design is a step toward platform development that is useful for studies aimed at patient trajectory analysis, medical outcome prediction, or studies that require heterogeneous data entries.

Towards Prediction of Injuries in Traffic Accidents.

For people involved in road traffic accidents, the time necessary to respond is crucial and it is hard to discern, which persons in which cars most urgently need help. To plan the rescue operation before arriving at the scene, digital information regarding the severity of the accident is vital. Our framework aims to transmit available data from the in-car sensors and to simulate the forces enacted on occupants using injury models. To avoid data security and privacy issues, we install low-cost hardware in the car for aggregation and preprocessing. Our framework can be retrofitted to existing cars and therefore could extend the benefits to a wide range of people.

A protocol for a systematic review of electronic early warning/track-and-trigger systems (EW/TTS) to predict clinical deterioration: Focus on automated features, technologies, and algorithms.

BACKGROUND: This is a systematic review protocol to identify automated features, applied technologies, and algorithms in the electronic early warning/track and triage system (EW/TTS) developed to predict clinical deterioration (CD). METHODOLOGY: This study will be conducted using PubMed, Scopus, and Web of Science databases to evaluate the features of EW/TTS in terms of their automated features, technologies, and algorithms. To this end, we will include any English articles reporting an EW/TTS without time limitation. Retrieved records will be independently screened by two authors and relevant data will be extracted from studies and abstracted for further analysis. The included articles will be evaluated independently using the JBI critical appraisal checklist by two researchers. DISCUSSION: This study is an effort to address the available automated features in the electronic version of the EW/TTS to shed light on the applied technologies, automated level of systems, and utilized algorithms in order to smooth the road toward the fully automated EW/TTS as one of the potential solutions of prevention CD and its adverse consequences. TRIAL REGISTRATION: Systematic review registration: PROSPERO CRD42022334988.

Recognizing Human Activity of Daily Living Using a Flexible Wearable for 3D Spine Pose Tracking.

The World Health Organization recognizes physical activity as an influencing domain on quality of life. Monitoring, evaluating, and supervising it by wearable devices can contribute to the early detection and progress assessment of diseases such as Alzheimer's, rehabilitation, and exercises in telehealth, as well as abrupt events such as a fall. In this work, we use a non-invasive and non-intrusive flexible wearable device for 3D spine pose measurement to monitor and classify physical activity. We develop a comprehensive protocol that consists of 10 indoor, 4 outdoor, and 8 transition states activities in three categories of static, dynamic, and transition in order to evaluate the applicability of the flexible wearable device in human activity recognition. We implement and compare the performance of three neural networks: long short-term memory (LSTM), convolutional neural network (CNN), and a hybrid model (CNN-LSTM). For ground truth, we use an accelerometer and strips data. LSTM reached an overall classification accuracy of 98% for all activities. The CNN model with accelerometer data delivered better performance in lying down (100%), static (standing = 82%, sitting = 75%), and dynamic (walking = 100%, running = 100%) positions. Data fusion improved the outputs in standing (92%) and sitting (94%), while LSTM with the strips data yielded a better performance in bending-related activities (bending forward = 49%, bending backward = 88%, bending right = 92%, and bending left = 100%), the combination of data fusion and principle components analysis further strengthened the output (bending forward = 100%, bending backward = 89%, bending right = 100%, and bending left = 100%). Moreover, the LSTM model detected the first transition state that is similar to fall with the accuracy of 84%. The results show that the wearable device can be used in a daily routine for activity monitoring, recognition, and exercise supervision, but still needs further improvement for fall detection.

AAPM task group report 273: Recommendations on best practices for AI and machine learning for computer-aided diagnosis in medical imaging.

Rapid advances in artificial intelligence (AI) and machine learning, and specifically in deep learning (DL) techniques, have enabled broad application of these methods in health care. The promise of the DL approach has spurred further interest in computer-aided diagnosis (CAD) development and applications using both "traditional" machine learning methods and newer DL-based methods. We use the term CAD-AI to refer to this expanded clinical decision support environment that uses traditional and DL-based AI methods. Numerous studies have been published to date on the development of machine learning tools for computer-aided, or AI-assisted, clinical tasks. However, most of these machine learning models are not ready for clinical deployment. It is of paramount importance to ensure that a clinical decision support tool undergoes proper training and rigorous validation of its generalizability and robustness before adoption for patient care in the clinic. To address these important issues, the American Association of Physicists in Medicine (AAPM) Computer-Aided Image Analysis Subcommittee (CADSC) is charged, in part, to develop recommendations on practices and standards for the development and performance assessment of computer-aided decision support systems. The committee has previously published two opinion papers on the evaluation of CAD systems and issues associated with user training and quality assurance of these systems in the clinic. With machine learning techniques continuing to evolve and CAD applications expanding to new stages of the patient care process, the current task group report considers the broader issues common to the development of most, if not all, CAD-AI applications and their translation from the bench to the clinic. The goal is to bring attention to the proper training and validation of machine learning algorithms that may improve their generalizability and reliability and accelerate the adoption of CAD-AI systems for clinical decision support.

One Digital Health Intervention for Monitoring Human and Animal Welfare in Smart Cities: Viewpoint and Use Case.

Smart cities and digital public health are closely related. Managing digital transformation in urbanization and living spaces is challenging. It is critical to prioritize the emotional and physical health and well-being of humans and their animals in the dynamic and ever-changing environment they share. Human-animal bonds are continuous as they live together or share urban spaces and have a mutual impact on each other's health as well as the surrounding environment. In addition, sensors embedded in the Internet of Things are everywhere in smart cities. They monitor events and provide appropriate responses. In this regard, accident and emergency informatics (A&EI) offers tools to identify and manage overtime hazards and disruptive events. Such manifold focuses fit with One Digital Health (ODH), which aims to transform health ecosystems with digital technology by proposing a comprehensive framework to manage data and support health-oriented policies. We showed and discussed how, by developing the concept of ODH intervention, the ODH framework can support the comprehensive monitoring and analysis of daily life events of humans and animals in technologically integrated environments such as smart homes and smart cities. We developed an ODH intervention use case in which A&EI mechanisms run in the background. The ODH framework structures the related data collection and analysis to enhance the understanding of human, animal, and environment interactions and associated outcomes. The use case looks at the daily journey of Tracy, a healthy woman aged 27 years, and her dog Mego. Using medical Internet of Things, their activities are continuously monitored and analyzed to prevent or manage any kind of health-related abnormality. We reported and commented on an ODH intervention as an example of a real-life ODH implementation. We gave the reader examples of a "how-to" analysis of Tracy and Mego's daily life activities as part of a timely implementation of the ODH framework. For each activity, relationships to the ODH dimensions were scored, and relevant technical fields were evaluated in light of the Findable, Accessible, Interoperable, and Reusable principles. This "how-to" can be used as a template for further analyses. An ODH intervention is based on Findable, Accessible, Interoperable, and Reusable data and real-time processing for global health monitoring, emergency management, and research. The data should be collected and analyzed continuously in a spatial-temporal domain to detect changes in behavior, trends, and emergencies. The information periodically gathered should serve human, animal, and environmental health interventions by providing professionals and caregivers with inputs and "how-to's" to improve health, welfare, and risk prevention at the individual and population levels. Thus, ODH complementarily combined with A&EI is meant to enhance policies and systems and modernize emergency management.

Best Research Papers in the Field of Sensors, Signals, and Imaging Informatics 2021.

OBJECTIVES: In this synopsis, we identify and highlight research papers representing noteworthy developments in signals, sensors, and imaging informatics in 2021. METHODS: A broad literature search was conducted on PubMed and Scopus databases. We combined Medical Subject Heading (MeSH) terms and keywords to construct particular queries for sensors, signals, and imaging informatics. Except for the sensor section, we only consider papers that have been published in journals providing at least three articles in the query response. Using a three-point Likert scale (1=not include, 2=maybe include, and 3=include), we reviewed the titles and abstracts of all database returns. Only those papers which reached two times three points were further considered for full paper review using the same Likert scale. Again, we only considered works with two times three points and provided these for external reviews. Based on the external reviews, we selected three best papers, as it happens that the three highest ranked papers represent works from all three parts of this section: sensors, signals, and imaging informatics. RESULTS: The search for papers was executed in January 2022. After removing duplicates and conference proceedings, the query returned a set of 88, 376, and 871 papers for sensors, signals, and imaging informatics, respectively. For signals and images, we filtered out journals that had less than three papers in the query results, reducing the number of papers to 215 and 512, respectively. From this total of 815 papers, the section co-editors identified 35 candidate papers with two times three Likert points, from which nine candidate best papers were nominated after full paper assessment. At least three external reviewers then rated the remaining papers and the three best-ranked papers were selected using the composite rating of all external reviewers. By accident, these three papers represent each of the three fields of sensor, signal, and imaging informatics. They were approved by consensus of the International Medical Informatics Association (IMIA) Yearbook editorial board. Deep and machine learning techniques are still a dominant topic as well as concepts beyond the state-of-the-art. CONCLUSIONS: Sensors, signals, and imaging informatics is a dynamic field of intense research. Current research focuses on creating and processing heterogeneous sensor data towards meaningful decision support in clinical settings.

Unobtrusive Heart Rate Monitoring using Near-Infrared Imaging During Driving.

In-vehicle health monitoring allows for continuous vital sign measurement in everyday life. Eventually, this could lead to early detection of cardiovascular diseases. In this work, we propose non-contact heart rate (HR) monitoring utilizing near-infrared (NIR) camera technology. Ten healthy volunteers are monitored in a realistic driving simulator during resting (5 min) and driving (10 min). We synchronously acquire videos using an out-of-the-shelf, low-cost NIR camera and 3-lead electrocardiography (ECG) serves as ground truth. The MediaPipe face detector delivers the region of interest (ROI) and we determine the HR from the peak with maximum amplitude within the power spectrum of skin color changes. We compare video-based with ECG-based HR, resulting in a mean absolute error (MAE) of 7.8 bpm and 13.0 bpm in resting and driving condition, respectively. As we apply only a simple signal processing pipeline without sophisticated filtering, we conclude that NIR camera-based HR measurements enables unobtrusive and non-contact monitoring to a certain extent, but artifacts from subject movement pose a challenge. If these issues can be addressed, continuous vital sign measurement in everyday life could become reality.

Human activity recognition from textile electrocardiograms.

Textile sensors for physiological signals bear the potential of unobtrusive and continuous application in daily life. Recently, textile electrocardiography (ECG) sensors became available which are of particular interest for physical activity monitoring due to the high effect of exercise on the heart rate. In this work, we evaluate the effectiveness of a single-lead ECG signal acquired using a non-medical-grade ECG shirt for human activity recognition (HAR). Healthy volunteers (N=10) wore the shirt during four different activities (sleeping, sitting, walking, running) in an uncontrolled environment and ECG data (256 Hz, 12 Bit) was stored, manually checked, and unusable segments (e.g. no sensor contact) were removed, resulting in a total of 228 hours of recording. Signals were split in short segments of different duration (10, 30, 60s), transformed using the Short-time Fourier Transform (STFT) to a spectrogram image and fed into a state-of-the-art convolutional neural network (CNN). The best configuration results in an F'l-Score of 73% and an accuracy of 77% on the test set. Results with leave-one-subject-out cross-validation show F'l-Scores ranging from 41 % to 80%. Thus, a single-lead, wearable-generated ECG has an informative value for HAR to a certain extent. In future work, we aim at using more sensors of the smart shirt and sensor fusion.