Advances, challenges and opportunities in creating data for trustworthy AI

Exploring Spectrogram-Based Audio Classification for Parkinson's Disease: A Study on Speech Classification and Qualitative Reliability Verification

Patients suffering from Parkinson's disease suffer from voice impairment. In this study, we introduce models to classify normal and Parkinson's patients using their speech. We used an AST (audio spectrogram transformer), a transformer-based speech classification model that has recently outperformed CNN-based models in many fields, and a CNN-based PSLA (pretraining, sampling, labeling, and aggregation), a high-performance model in the existing speech classification field, for the study. This study compares and analyzes the models from both quantitative and qualitative perspectives. First, qualitatively, PSLA outperformed AST by more than 4% in accuracy, and the AUC was also higher, with 94.16% for AST and 97.43% for PSLA. Furthermore, we qualitatively evaluated the ability of the models to capture the acoustic features of Parkinson's through various CAM (class activation map)-based XAI (eXplainable AI) models such as GradCAM and EigenCAM. Based on PSLA, we found that the model focuses well on the muffled frequency band of Parkinson's speech, and the heatmap analysis of false positives and false negatives shows that the speech features are also visually represented when the model actually makes incorrect predictions. The contribution of this paper is that we not only found a suitable model for diagnosing Parkinson's through speech using two different types of models but also validated the predictions of the model in practice.

Leveraging artificial intelligence in vaccine development: A narrative review

Vaccine development stands as a cornerstone of public health efforts, pivotal in curbing infectious diseases and reducing global morbidity and mortality. However, traditional vaccine development methods are often time-consuming, costly, and inefficient. The advent of artificial intelligence (AI) has ushered in a new era in vaccine design, offering unprecedented opportunities to expedite the process. This narrative review explores the role of AI in vaccine development, focusing on antigen selection, epitope prediction, adjuvant identification, and optimization strategies. AI algorithms, including machine learning and deep learning, leverage genomic data, protein structures, and immune system interactions to predict antigenic epitopes, assess immunogenicity, and prioritize antigens for experimentation. Furthermore, AI-driven approaches facilitate the rational design of immunogens and the identification of novel adjuvant candidates with optimal safety and efficacy profiles. Challenges such as data heterogeneity, model interpretability, and regulatory considerations must be addressed to realize the full potential of AI in vaccine development. Integrating emerging technologies, such as single-cell omics and synthetic biology, promises to enhance vaccine design precision and scalability. This review underscores the transformative impact of AI on vaccine development and highlights the need for interdisciplinary collaborations and regulatory harmonization to accelerate the delivery of safe and effective vaccines against infectious diseases.

MFS-DBF: A trustworthy multichannel feature sieve and decision boundary formulation system for Obstructive Sleep Apnea detection

The fine identification of sleep apnea events is instrumental in Obstructive Sleep Apnea (OSA) diagnosis. The development of sleep apnea event detection algorithms based on polysomnography is becoming a research hotspot in medical signal processing. In this paper, we propose an Inverse-Projection based Visualization System (IPVS) for sleep apnea event detection algorithms. The IPVS consists of a feature dimensionality reduction module and a feature reconstruction module. First, features of blood oxygen saturation and nasal airflow are extracted and used as input data for event analysis. Then, visual analysis is conducted on the feature distribution for apnea events. Next, dimensionality reduction and reconstruction methods are combined to achieve the dynamic visualization of sleep apnea event feature sets and the visual analysis of classifier decision boundaries. Moreover, the decision-making consistency is explored for various sleep apnea event detection classifiers, which provides researchers and users with an intuitive understanding of the detection algorithm. We applied the IPVS to an OSA detection algorithm with an accuracy of 84% and a diagnostic accuracy of 92% on a publicly available dataset. The experimental results show that the consistency between our visualization results and prior medical knowledge provides strong evidence for the practicality of the proposed system. For clinical practice, the IPVS can guide users to focus on samples with higher uncertainty presented by the OSA detection algorithm, reducing the workload and improving the efficiency of clinical diagnosis, which in turn increases the value of trust.

Approaches for the use of Artificial Intelligence in workplace health promotion and prevention: A systematic scoping review

BACKGROUND

Artificial intelligence (AI) is an umbrella term for various algorithms and rapidly emerging technologies with huge potential for workplace health promotion and prevention (WHPP). WHPP interventions aim to improve people's health and well-being through behavioral and organizational measures or by minimizing the burden of workplace-related diseases and associated risk factors. While AI has been the focus of research in other health-related fields, such as Public Health or biomedicine, the transition of AI into WHPP research has yet to be systematically investigated.

OBJECTIVE

The systematic scoping review aims to comprehensively assess an overview of the current use of AI in WHPP. The results will be then used to point to future research directions. The following research questions were derived: (1) what are the study characteristics of studies on AI algorithms and technologies in the context of WHPP, (2) what specific WHPP fields (prevention, behavioral, and organizational approaches) were addressed by the AI algorithms and technologies, and (3) what kind of interventions lead to which outcomes?

METHODS

A systematic scoping literature review (PRISMA-ScR) was conducted in the three academic databases PubMed, IEEE, and ACM in July 2023, searching for articles published between January 2000 and December 2023. Studies needed to be 1) peer-reviewed, 2) written in English, and 3) focused on any AI-based algorithm or technology that (4) were conducted in the context of WHPP or (5) an associated field. Information on study design, AI algorithms and technologies, WHPP fields, and the PICO framework were extracted blindly with Rayyan and summarized.

RESULTS

A total of ten studies were included. Risk prevention and modeling were the most identified WHPP fields (n=6), followed by behavioral health promotion (n=4) and organizational health promotion (n=1). Four studies focused on mental health. Most AI algorithms were machine learning-based, and three studies used combined deep learning algorithms. AI algorithms and technologies were primarily implemented in smartphone applications (eg, in the form of a Chatbot) or used the smartphone as a data source (eg, GPS). Behavioral approaches ranged from 8 to 12 weeks and were compared to control groups. Three studies evaluated the robustness and accuracy of an AI model or framework.

CONCLUSIONS

Although AI has caught increasing attention in health-related research, the review reveals that AI in WHPP is marginally investigated. Our results indicate that AI is promising for individualization and risk prediction in WHPP, but current research does not cover the scope of WHPP. Beyond that, future research will profit from an extended range of research in all fields of WHPP, longitudinal data, and reporting guidelines.

CLINICALTRIAL

Registered on 5th July 2023 at Open Science Framework [1].

Prediction of early-phase cytomegalovirus pneumonia in post-stem cell transplantation using a deep learning model

BACKGROUND

Diagnostic challenges exist for CMV pneumonia in post-hematopoietic stem cell transplantation (post-HSCT) patients, despite early-phase radiographic changes.

OBJECTIVE

The study aims to employ a deep learning model distinguishing CMV pneumonia from COVID-19 pneumonia, community-acquired pneumonia, and normal lungs post-HSCT.

METHODS

Initially, 6 neural network models were pre-trained with COVID-19 pneumonia, community-acquired pneumonia, and normal lung CT images from Kaggle's COVID multiclass dataset (Dataset A), then Dataset A was combined with the CMV pneumonia images from our center, forming Dataset B. We use a few-shot transfer learning strategy to fine-tune the pre-trained models and evaluate model performance in Dataset B.

RESULTS

34 cases of CMV pneumonia were found between January 2018 and December 2022 post-HSCT. Dataset A contained 1681 images of each subgroup from Kaggle. Combined with Dataset A, Dataset B was initially formed by 98 images of CMV pneumonia and normal lung. The optimal model (Xception) achieved an accuracy of 0.9034. Precision, recall, and F1-score all reached 0.9091, with an AUC of 0.9668 in the test set of Dataset B.

CONCLUSIONS

This framework demonstrates the deep learning model's ability to distinguish rare pneumonia types utilizing a small volume of CT images, facilitating early detection of CMV pneumonia post-HSCT.

Conceptualizing future groundwater models through a ternary framework of multisource data, human expertise, and machine intelligence

Groundwater models are essential for understanding aquifer systems behavior and effective water resources spatio-temporal distributions, yet they are often hindered by challenges related to model assumptions, parametrization, uncertainty, and computational efficiency. Machine intelligence, especially deep learning, promises a paradigm shift in overcoming these challenges. A critical examination of existing machine-driven methods reveals the inherent limitations, particularly in terms of the interpretability and the ability to generalize findings. To overcome these challenges, we develop a ternary framework that synergizes the valuable insights from multisource data, human expertise, and machine intelligence. This framework capitalizes on the distinct strengths of each element: the value and relevance of multisource data, the innovative capacity of human expertise, and the analytical efficiency of machine intelligence. Our goal is to conceptualize sustainable water management practices and enhance our understanding and predictive capabilities of groundwater systems. Unlike approaches that rely solely on abundant data, our framework emphasizes the quality and strategic use of available data, combined with human intellect and advanced computing, to overcome current limitations and pave the way for more realistic groundwater simulations.

Privacy-Preserving Synthetic Continual Semantic Segmentation for Robotic Surgery

Deep Neural Networks (DNNs) based semantic segmentation of the robotic instruments and tissues can enhance the precision of surgical activities in robot-assisted surgery. However, in biological learning, DNNs cannot learn incremental tasks over time and exhibit catastrophic forgetting, which refers to the sharp decline in performance on previously learned tasks after learning a new one. Specifically, when data scarcity is the issue, the model shows a rapid drop in performance on previously learned instruments after learning new data with new instruments. The problem becomes worse when it limits releasing the dataset of the old instruments for the old model due to privacy concerns and the unavailability of the data for the new or updated version of the instruments for the continual learning model. For this purpose, we develop a privacy-preserving synthetic continual semantic segmentation framework by blending and harmonizing (i) open-source old instruments foreground to the synthesized background without revealing real patient data in public and (ii) new instruments foreground to extensively augmented real background. To boost the balanced logit distillation from the old model to the continual learning model, we design overlapping class-aware temperature normalization (CAT) by controlling model learning utility. We also introduce multi-scale shifted-feature distillation (SD) to maintain long and short-range spatial relationships among the semantic objects where conventional short-range spatial features with limited information reduce the power of feature distillation. We demonstrate the effectiveness of our framework on the EndoVis 2017 and 2018 instrument segmentation dataset with a generalized continual learning setting. Code is available at https://github.com/XuMengyaAmy/Synthetic_CAT_SD.

Weapon Violence Dataset 2.0: A synthetic dataset for violence detection

In the current era, satisfying the appetite of data hungry models is becoming an increasingly challenging task. This challenge is particularly magnified in research areas characterised by sensitivity, where the quest for genuine data proves to be elusive. The study of violence serves as a poignant example, entailing ethical considerations and compounded by the scarcity of authentic, real-world data that is predominantly accessible only to law enforcement agencies. Existing datasets in this field often resort to using content from movies or open-source video platforms like YouTube, further emphasising the scarcity of authentic data. To address this, our dataset aims to pioneer a new approach by creating the first synthetic virtual dataset for violence detection, named the Weapon Violence Dataset (WVD). The dataset is generated by creating virtual violence scenarios inside the photo-realistic video game namely: Grand Theft Auto-V (GTA-V). This dataset includes carefully selected video clips of person-to-person fights captured from a frontal view, featuring various weapons-both hot and cold across different times of the day. Specifically, WVD contains three categories: Hot violence and Cold violence (representing the violence category) as well as No violence (constituting the control class). The dataset is designed and created in a way that will enable the research community to train deep models on such synthetic data with the ability to increase the data corpus if the needs arise. The dataset is publicly available on Kaggle and comprises normal RGB and optic flow videos.

Enhancing psychiatric rehabilitation outcomes through a multimodal multitask learning model based on BERT and TabNet: An approach for personalized treatment and improved decision-making

Evaluating the rehabilitation status of individuals with serious mental illnesses (SMI) necessitates a comprehensive analysis of multimodal data, including unstructured text records and structured diagnostic data. However, progress in the effective assessment of rehabilitation status remains limited. Our study develops a deep learning model integrating Bidirectional Encoder Representations from Transformers (BERT) and TabNet through a late fusion strategy to enhance rehabilitation prediction, including referral risk, dangerous behaviors, self-awareness, and medication adherence, in patients with SMI. BERT processes unstructured textual data, such as doctor's notes, whereas TabNet manages structured diagnostic information. The model's interpretability function serves to assist healthcare professionals in understanding the model's predictive decisions, improving patient care. Our model exhibited excellent predictive performance for all four tasks, with an accuracy exceeding 0.78 and an area under the curve of 0.70. In addition, a series of tests proved the model's robustness, fairness, and interpretability. This study combines multimodal and multitask learning strategies into a model and applies it to rehabilitation assessment tasks, offering a promising new tool that can be seamlessly integrated with the clinical workflow to support the provision of optimized patient care.

An artificial intelligence tool predicts blastocyst development from static images of fresh mature oocytes

RESEARCH QUESTION

Can a deep learning image analysis model be developed to assess oocyte quality by predicting blastocyst development from images of denuded mature oocytes?

DESIGN

A deep learning model was developed utilizing 37,133 static oocyte images with associated laboratory outcomes from eight fertility clinics (six countries). A subset of data (n = 7807) was allocated to test model performance. External model validation was conducted to assess generalizability and robustness on new data (n = 12,357) from two fertility clinics (two countries). Performance was assessed by calculating area under the curve (AUC), balanced accuracy, specificity and sensitivity. Subgroup analyses were performed on the test dataset for age group, male factor and geographical location of the clinic. Model probabilities of the external dataset were converted to a 0-10 scoring scale to facilitate analysis of correlation with blastocyst development and quality.

RESULTS

The deep learning model demonstrated AUC of 0.64, balanced accuracy of 0.60, specificity of 0.55 and sensitivity of 0.65 on the test dataset. Subgroup analyses displayed the highest performance for age group 38-39 years (AUC 0.68), a negligible impact of male factor, and good model generalizability across geographical locations. Model performance was confirmed on external data: AUC of 0.63, balanced accuracy of 0.58, specificity of 0.57 and sensitivity of 0.59. Analysis of the scoring scale revealed that higher scoring oocytes correlated with higher likelihood of blastocyst development and good-quality blastocyst formation.

CONCLUSION

The deep learning model showed a favourable performance for the evaluation of oocytes in terms of competence to develop into a blastocyst, and when the predictions were converted into scores, they correlated with blastocyst quality. This represents a significant first step in oocyte evaluation for scientific and clinical applications.

Enhancement of OCTen faceimages by unsupervised deep learning

Objective. The quality of optical coherence tomography (OCT)en faceimages is crucial for clinical visualization of early disease. As a three dimensional and coherent imaging, defocus and speckle noise are inevitable, which seriously affect evaluation of microstructure of bio-samples in OCT images. The deep learning has demonstrated great potential in OCT refocusing and denoising, but it is limited by the difficulty of sufficient paired training data. This work aims to develop an unsupervised method to enhance the quality of OCTen faceimages.Approach. We proposed an unsupervised deep learning-based pipeline. The unregistered defocused conventional OCT images and focused speckle-free OCT images were collected by a home-made speckle modulating OCT system to construct the dataset. The image enhancement model was trained with the cycle training strategy. Finally, the speckle noise and defocus were both effectively improved.Main results. The experimental results on complex bio-samples indicated that the proposed method is effective and generalized in enhancing the quality of OCTen faceimages.Significance. The proposed unsupervised deep learning method helps to reduce the complexity of data construction, which is conducive to practical applications in OCT bio-sample imaging.

Data-driven evolution of water quality models: An in-depth investigation of innovative outlier detection approaches-A case study of Irish Water Quality Index (IEWQI) model

Recently, there has been a significant advancement in the water quality index (WQI) models utilizing data-driven approaches, especially those integrating machine learning and artificial intelligence (ML/AI) technology. Although, several recent studies have revealed that the data-driven model has produced inconsistent results due to the data outliers, which significantly impact model reliability and accuracy. The present study was carried out to assess the impact of data outliers on a recently developed Irish Water Quality Index (IEWQI) model, which relies on data-driven techniques. To the author's best knowledge, there has been no systematic framework for evaluating the influence of data outliers on such models. For the purposes of assessing the outlier impact of the data outliers on the water quality (WQ) model, this was the first initiative in research to introduce a comprehensive approach that combines machine learning with advanced statistical techniques. The proposed framework was implemented in Cork Harbour, Ireland, to evaluate the IEWQI model's sensitivity to outliers in input indicators to assess the water quality. In order to detect the data outlier, the study utilized two widely used ML techniques, including Isolation Forest (IF) and Kernel Density Estimation (KDE) within the dataset, for predicting WQ with and without these outliers. For validating the ML results, the study used five commonly used statistical measures. The performance metric (R2) indicates that the model performance improved slightly (R2 increased from 0.92 to 0.95) in predicting WQ after removing the data outlier from the input. But the IEWQI scores revealed that there were no statistically significant differences among the actual values, predictions with outliers, and predictions without outliers, with a 95 % confidence interval at p < 0.05. The results of model uncertainty also revealed that the model contributed <1 % uncertainty to the final assessment results for using both datasets (with and without outliers). In addition, all statistical measures indicated that the ML techniques provided reliable results that can be utilized for detecting outliers and their impacts on the IEWQI model. The findings of the research reveal that although the data outliers had no significant impact on the IEWQI model architecture, they had moderate impacts on the rating schemes' of the model. This finding indicated that detecting the data outliers could improve the accuracy of the IEWQI model in rating WQ as well as be helpful in mitigating the model eclipsing problem. In addition, the results of the research provide evidence of how the data outliers influenced the data-driven model in predicting WQ and reliability, particularly since the study confirmed that the IEWQI model's could be effective for accurately rating WQ despite the presence of the data outliers in the input. It could occur due to the spatio-temporal variability inherent in WQ indicators. However, the research assesses the influence of data input outliers on the IEWQI model and underscores important areas for future investigation. These areas include expanding temporal analysis using multi-year data, examining spatial outlier patterns, and evaluating detection methods. Moreover, it is essential to explore the real-world impacts of revised rating categories, involve stakeholders in outlier management, and fine-tune model parameters. Analysing model performance across varying temporal and spatial resolutions and incorporating additional environmental data can significantly enhance the accuracy of WQ assessment. Consequently, this study offers valuable insights to strengthen the IEWQI model's robustness and provides avenues for enhancing its utility in broader WQ assessment applications. Moreover, the study successfully adopted the framework for evaluating how data input outliers affect the data-driven model, such as the IEWQI model. The current study has been carried out in Cork Harbour for only a single year of WQ data. The framework should be tested across various domains for evaluating the response of the IEWQI model's in terms of the spatio-temporal resolution of the domain. Nevertheless, the study recommended that future research should be conducted to adjust or revise the IEWQI model's rating schemes and investigate the practical effects of data outliers on updated rating categories. However, the study provides potential recommendations for enhancing the IEWQI model's adaptability and reveals its effectiveness in expanding its applicability in more general WQ assessment scenarios.

Unveiling value patterns via deep reinforcement learning in heterogeneous data analytics

Artificial intelligence has substantially improved the efficiency of data utilization across various sectors. However, the insufficient filtering of low-quality data poses challenges to uncertainty management, threatening system stability. In this study, we introduce a data-valuation approach employing deep reinforcement learning to elucidate the value patterns in data-driven tasks. By strategically optimizing with iterative sampling and feedback, our method is effective in diverse scenarios and consistently outperforms the classic methods in both accuracy and efficiency. In China's wind-power prediction, excluding 25% of the overall dataset deemed low-value led to a 10.5% improvement in accuracy. Utilizing just 42.8% of the dataset, the model discerned 80% of linear patterns, showcasing the data's intrinsic and transferable value. A nationwide analysis identified a data-value-sensitive geographic belt across 10 provinces, leading to robust policy recommendations informed by variances in power outputs and data values, as well as geographic climate factors.

Synthetic PET from CT improves diagnosis and prognosis for lung cancer: Proof of concept

[18F]Fluorodeoxyglucose positron emission tomography (FDG-PET) and computed tomography (CT) are indispensable components in modern medicine. Although PET can provide additional diagnostic value, it is costly and not universally accessible, particularly in low-income countries. To bridge this gap, we have developed a conditional generative adversarial network pipeline that can produce FDG-PET from diagnostic CT scans based on multi-center multi-modal lung cancer datasets (n = 1,478). Synthetic PET images are validated across imaging, biological, and clinical aspects. Radiologists confirm comparable imaging quality and tumor contrast between synthetic and actual PET scans. Radiogenomics analysis further proves that the dysregulated cancer hallmark pathways of synthetic PET are consistent with actual PET. We also demonstrate the clinical values of synthetic PET in improving lung cancer diagnosis, staging, risk prediction, and prognosis. Taken together, this proof-of-concept study testifies to the feasibility of applying deep learning to obtain high-fidelity PET translated from CT.

Trustworthy deep learning framework for the detection of abnormalities in X-ray shoulder images

Musculoskeletal conditions affect an estimated 1.7 billion people worldwide, causing intense pain and disability. These conditions lead to 30 million emergency room visits yearly, and the numbers are only increasing. However, diagnosing musculoskeletal issues can be challenging, especially in emergencies where quick decisions are necessary. Deep learning (DL) has shown promise in various medical applications. However, previous methods had poor performance and a lack of transparency in detecting shoulder abnormalities on X-ray images due to a lack of training data and better representation of features. This often resulted in overfitting, poor generalisation, and potential bias in decision-making. To address these issues, a new trustworthy DL framework has been proposed to detect shoulder abnormalities (such as fractures, deformities, and arthritis) using X-ray images. The framework consists of two parts: same-domain transfer learning (TL) to mitigate imageNet mismatch and feature fusion to reduce error rates and improve trust in the final result. Same-domain TL involves training pre-trained models on a large number of labelled X-ray images from various body parts and fine-tuning them on the target dataset of shoulder X-ray images. Feature fusion combines the extracted features with seven DL models to train several ML classifiers. The proposed framework achieved an excellent accuracy rate of 99.2%, F1Score of 99.2%, and Cohen's kappa of 98.5%. Furthermore, the accuracy of the results was validated using three visualisation tools, including gradient-based class activation heat map (Grad CAM), activation visualisation, and locally interpretable model-independent explanations (LIME). The proposed framework outperformed previous DL methods and three orthopaedic surgeons invited to classify the test set, who obtained an average accuracy of 79.1%. The proposed framework has proven effective and robust, improving generalisation and increasing trust in the final results.

Deep reinforcement learning enables better bias control in benchmark for virtual screening

Virtual screening (VS) has been incorporated into the paradigm of modern drug discovery. This field is now undergoing a new wave of revolution driven by artificial intelligence and more specifically, machine learning (ML). In terms of those out-of-the-box datasets for model training or benchmarking, their data volume and applicability domain are limited. They are suffering from the biases constantly reported in the ML application. To address these issues, we present a novel benchmark named MUBDsyn. The utilization of synthetic decoys (i.e., presumed inactives) is the main feature of MUBDsyn, where deep reinforcement learning was leveraged for bias control during decoy generation. Then, we carried out extensive validations on this new benchmark. First, we confirmed that MUBDsyn was superior to the classical benchmarks in control of domain bias, artificial enrichment bias and analogue bias. Moreover, we found that the assessment of ML models based on MUBDsyn was less biased as revealed by the analysis of asymmetric validation embedding bias. In addition, MUBDsyn showed better setting of benchmarking challenge for deep learning models compared with NRLiSt-BDB. Overall, we have proven that MUBDsyn is the close-to-ideal benchmark for VS. The computational tool is publicly available for the easy extension of MUBDsyn.

Evaluating the Utilities of Foundation Models in Single-cell Data Analysis

Foundation Models (FMs) have made significant strides in both industrial and scientific domains. In this paper, we evaluate the performance of FMs in single-cell sequencing data analysis through comprehensive experiments across eight downstream tasks pertinent to single-cell data. By comparing ten different single-cell FMs with task-specific methods, we found that single-cell FMs may not consistently excel in all tasks than task-specific methods. However, the emergent abilities and the successful applications of cross-species/cross-modality transfer learning of FMs are promising. In addition, we present a systematic evaluation of the effects of hyper-parameters, initial settings, and stability for training single-cell FMs based on a proposed scEval framework, and provide guidelines for pre-training and fine-tuning. Our work summarizes the current state of single-cell FMs and points to their constraints and avenues for future development.

Suboptimal capability of individual machine learning algorithms in modeling small-scale imbalanced clinical data of local hospital

In recent years, artificial intelligence (AI) has shown promising applications in various scientific domains, including biochemical analysis research. However, the effectiveness of AI in modeling small-scale, imbalanced datasets remains an open question in such fields. This study explores the capabilities of eight basic AI algorithms, including ridge regression, logistic regression, random forest regression, and others, in modeling a small, imbalanced clinical dataset (total n = 387, class 0 = 27, class 1 = 360) related to the records of the biochemical blood tests from the patients with multiple wasp stings (MWS). Through rigorous evaluation using k-fold cross-validation and comprehensive scoring, we found that none of the models could effectively model the data. Even after fine-tuning the hyperparameters of the best-performing models, the results remained below acceptable thresholds. The study highlights the challenges of applying AI to small-scale datasets with imbalanced groups in biochemical or clinical research and emphasizes the need for novel algorithms tailored to small-scale data. The findings also call for further exploration into techniques such as transfer learning and data augmentation, and they underline the importance of understanding the minimum dataset scale required for effective AI modeling in biochemical contexts.

A deep-learning-based framework for identifying and localizing multiple abnormalities and assessing cardiomegaly in chest X-ray

Accurate identification and localization of multiple abnormalities are crucial steps in the interpretation of chest X-rays (CXRs); however, the lack of a large CXR dataset with bounding boxes severely constrains accurate localization research based on deep learning. We created a large CXR dataset named CXR-AL14, containing 165,988 CXRs and 253,844 bounding boxes. On the basis of this dataset, a deep-learning-based framework was developed to identify and localize 14 common abnormalities and calculate the cardiothoracic ratio (CTR) simultaneously. The mean average precision values obtained by the model for 14 abnormalities reached 0.572-0.631 with an intersection-over-union threshold of 0.5, and the intraclass correlation coefficient of the CTR algorithm exceeded 0.95 on the held-out, multicentre and prospective test datasets. This framework shows an excellent performance, good generalization ability and strong clinical applicability, which is superior to senior radiologists and suitable for routine clinical settings.

Artificial Intelligence in Kidney Disease: A Comprehensive Study and Directions for Future Research

Artificial intelligence (AI) has emerged as a promising tool in the field of healthcare, with an increasing number of research articles evaluating its applications in the domain of kidney disease. To comprehend the evolving landscape of AI research in kidney disease, a bibliometric analysis is essential. The purposes of this study are to systematically analyze and quantify the scientific output, research trends, and collaborative networks in the application of AI to kidney disease. This study collected AI-related articles published between 2012 and 20 November 2023 from the Web of Science. Descriptive analyses of research trends in the application of AI in kidney disease were used to determine the growth rate of publications by authors, journals, institutions, and countries. Visualization network maps of country collaborations and author-provided keyword co-occurrences were generated to show the hotspots and research trends in AI research on kidney disease. The initial search yielded 673 articles, of which 631 were included in the analyses. Our findings reveal a noteworthy exponential growth trend in the annual publications of AI applications in kidney disease. Nephrology Dialysis Transplantation emerged as the leading publisher, accounting for 4.12% (26 out of 631 papers), followed by the American Journal of Transplantation at 3.01% (19/631) and Scientific Reports at 2.69% (17/631). The primary contributors were predominantly from the United States (n = 164, 25.99%), followed by China (n = 156, 24.72%) and India (n = 62, 9.83%). In terms of institutions, Mayo Clinic led with 27 contributions (4.27%), while Harvard University (n = 19, 3.01%) and Sun Yat-Sen University (n = 16, 2.53%) secured the second and third positions, respectively. This study summarized AI research trends in the field of kidney disease through statistical analysis and network visualization. The findings show that the field of AI in kidney disease is dynamic and rapidly progressing and provides valuable information for recognizing emerging patterns, technological shifts, and interdisciplinary collaborations that contribute to the advancement of knowledge in this critical domain.

Pathogenomics for accurate diagnosis, treatment, prognosis of oncology: a cutting edge overview

The capability to gather heterogeneous data, alongside the increasing power of artificial intelligence to examine it, leading a revolution in harnessing multimodal data in the life sciences. However, most approaches are limited to unimodal data, leaving integrated approaches across modalities relatively underdeveloped in computational pathology. Pathogenomics, as an invasive method to integrate advanced molecular diagnostics from genomic data, morphological information from histopathological imaging, and codified clinical data enable the discovery of new multimodal cancer biomarkers to propel the field of precision oncology in the coming decade. In this perspective, we offer our opinions on synthesizing complementary modalities of data with emerging multimodal artificial intelligence methods in pathogenomics. It includes correlation between the pathological and genomic profile of cancer, fusion of histology, and genomics profile of cancer. We also present challenges, opportunities, and avenues for future work.

Enhancing interpretability and generalizability of deep learning-based emulator in three-dimensional lake hydrodynamics using Koopman operator and transfer learning: Demonstrated on the example of lake Zurich

Three-dimensional lake hydrodynamic model is a powerful tool widely used to assess hydrological condition changes of lake. However, its computational cost becomes problematic when forecasting the state of large lakes or using high-resolution simulation in small-to-medium size lakes. One possible solution is to employ a data-driven emulator, such as a deep learning (DL) based emulator, to replace the original model for fast computing. However, existing DL-based emulators are often black-box and data-dependent models, causing poor interpretability and generalizability in practical applications. In this study, a data-driven emulator is established using deep neural network (DNN) to replace the original model for fast computing of three-dimensional lake hydrodynamics. Then, the Koopman operator and transfer learning (TL) are employed to enhance the interpretability and generalizability of the emulator. Finally, the generalizability of DL-based emulators is comprehensively analyzed through linear regression and correlation analysis. These methods are tested against an existing hydrodynamic model of Lake Zurich (Switzerland) whose data was provided by an open-source web-based platform called Meteolakes/Alplakes. According to the results, (1) The DLEDMD offers better interpretability than DNN because its Koopman operator reveals the linear structure behind the hydrodynamics; (2) The generalization of the DL-based emulators in three-dimensional lake hydrodynamics are influenced by the similarity between the training and testing data; (3) TL effectively improves the generalizability of the DL-based emulators.

Sketch-based semantic retrieval of medical images

The volume of medical images stored in hospitals is rapidly increasing; however, the utilization of these accumulated medical images remains limited. Existing content-based medical image retrieval (CBMIR) systems typically require example images, leading to practical limitations, such as the lack of customizable, fine-grained image retrieval, the inability to search without example images, and difficulty in retrieving rare cases. In this paper, we introduce a sketch-based medical image retrieval (SBMIR) system that enables users to find images of interest without the need for example images. The key concept is feature decomposition of medical images, which allows the entire feature of a medical image to be decomposed into and reconstructed from normal and abnormal features. Building on this concept, our SBMIR system provides an easy-to-use two-step graphical user interface: users first select a template image to specify a normal feature and then draw a semantic sketch of the disease on the template image to represent an abnormal feature. The system integrates both types of input to construct a query vector and retrieves reference images. For evaluation, ten healthcare professionals participated in a user test using two datasets. Consequently, our SBMIR system enabled users to overcome previous challenges, including image retrieval based on fine-grained image characteristics, image retrieval without example images, and image retrieval for rare cases. Our SBMIR system provides on-demand, customizable medical image retrieval, thereby expanding the utility of medical image databases.

Neural networks memorise personal information from one sample

Deep neural networks (DNNs) have achieved high accuracy in diagnosing multiple diseases/conditions at a large scale. However, a number of concerns have been raised about safeguarding data privacy and algorithmic bias of the neural network models. We demonstrate that unique features (UFs), such as names, IDs, or other patient information can be memorised (and eventually leaked) by neural networks even when it occurs on a single training data sample within the dataset. We explain this memorisation phenomenon by showing that it is more likely to occur when UFs are an instance of a rare concept. We propose methods to identify whether a given model does or does not memorise a given (known) feature. Importantly, our method does not require access to the training data and therefore can be deployed by an external entity. We conclude that memorisation does have implications on model robustness, but it can also pose a risk to the privacy of patients who consent to the use of their data for training models.

A framework for integrating artificial intelligence for clinical care with continuous therapeutic monitoring

The complex relationships between continuously monitored health signals and therapeutic regimens can be modelled via machine learning. However, the clinical implementation of the models will require changes to clinical workflows. Here we outline ClinAIOps ('clinical artificial-intelligence operations'), a framework that integrates continuous therapeutic monitoring and the development of artificial intelligence (AI) for clinical care. ClinAIOps leverages three feedback loops to enable the patient to make treatment adjustments using AI outputs, the clinician to oversee patient progress with AI assistance, and the AI developer to receive continuous feedback from both the patient and the clinician. We lay out the central challenges and opportunities in the deployment of ClinAIOps by means of examples of its application in the management of blood pressure, diabetes and Parkinson's disease. By enabling more frequent and accurate measurements of a patient's health and more timely adjustments to their treatment, ClinAIOps may substantially improve patient outcomes.

Only the Best of the Bunch-Sperm Preparation Is Not Just about Numbers

In this Seminar, we present an overview of the current and emerging methods and technologies for optimizing the man and the sperm sample for fertility treatment. We argue that sperms are the secret to success, and that there are many avenues for improving both treatment and basic understanding of their role in outcomes. These outcomes encompass not just whether treatment is successful or not, but the wider intergenerational health of the offspring. We discuss outstanding challenges and opportunities of new technologies such as microfluidics and artificial intelligence, including potential pitfalls and advantages. This article aims to provide a comprehensive overview of the importance of sperm in fertility treatment and suggests future directions for research and innovation.

Boundary delineation in transrectal ultrasound images for region of interest of prostate

Accurate and robust prostate segmentation in transrectal ultrasound (TRUS) images is of great interest for ultrasound-guided brachytherapy for prostate cancer. However, the current practice of manual segmentation is difficult, time-consuming, and prone to errors. To overcome these challenges, we developed an accurate prostate segmentation framework (A-ProSeg) for TRUS images. The proposed segmentation method includes three innovation steps: (1) acquiring the sequence of vertices by using an improved polygonal segment-based method with a small number of radiologist-defined seed points as prior points; (2) establishing an optimal machine learning-based method by using the improved evolutionary neural network; and (3) obtaining smooth contours of the prostate region of interest using the optimized machine learning-based method. The proposed method was evaluated on 266 patients who underwent prostate cancer brachytherapy. The proposed method achieved a high performance against the ground truth with a Dice similarity coefficient of 96.2% ± 2.4%, a Jaccard similarity coefficient of 94.4% ± 3.3%, and an accuracy of 95.7% ± 2.7%; these values are all higher than those obtained using state-of-the-art methods. A sensitivity evaluation on different noise levels demonstrated that our method achieved high robustness against changes in image quality. Meanwhile, an ablation study was performed, and the significance of all the key components of the proposed method was demonstrated.

Harnessing human and machine intelligence for planetary-level climate action

The ongoing global race for bigger and better artificial intelligence (AI) systems is expected to have a profound societal and environmental impact by altering job markets, disrupting business models, and enabling new governance and societal welfare structures that can affect global consensus for climate action pathways. However, the current AI systems are trained on biased datasets that could destabilize political agencies impacting climate change mitigation and adaptation decisions and compromise social stability, potentially leading to societal tipping events. Thus, the appropriate design of a less biased AI system that reflects both direct and indirect effects on societies and planetary challenges is a question of paramount importance. In this paper, we tackle the question of data-centric knowledge generation for climate action in ways that minimize biased AI. We argue for the need to co-align a less biased AI with an epistemic web on planetary health challenges for more trustworthy decision-making. A human-in-the-loop AI can be designed to align with three goals. First, it can contribute to a planetary epistemic web that supports climate action. Second, it can directly enable mitigation and adaptation interventions through knowledge of social tipping elements. Finally, it can reduce the data injustices associated with AI pretraining datasets.

Deep Transfer Learning with Enhanced Feature Fusion for Detection of Abnormalities in X-ray Images

Medical image classification poses significant challenges in real-world scenarios. One major obstacle is the scarcity of labelled training data, which hampers the performance of image-classification algorithms and generalisation. Gathering sufficient labelled data is often difficult and time-consuming in the medical domain, but deep learning (DL) has shown remarkable performance, although it typically requires a large amount of labelled data to achieve optimal results. Transfer learning (TL) has played a pivotal role in reducing the time, cost, and need for a large number of labelled images. This paper presents a novel TL approach that aims to overcome the limitations and disadvantages of TL that are characteristic of an ImageNet dataset, which belongs to a different domain. Our proposed TL approach involves training DL models on numerous medical images that are similar to the target dataset. These models were then fine-tuned using a small set of annotated medical images to leverage the knowledge gained from the pre-training phase. We specifically focused on medical X-ray imaging scenarios that involve the humerus and wrist from the musculoskeletal radiographs (MURA) dataset. Both of these tasks face significant challenges regarding accurate classification. The models trained with the proposed TL were used to extract features and were subsequently fused to train several machine learning (ML) classifiers. We combined these diverse features to represent various relevant characteristics in a comprehensive way. Through extensive evaluation, our proposed TL and feature-fusion approach using ML classifiers achieved remarkable results. For the classification of the humerus, we achieved an accuracy of 87.85%, an F1-score of 87.63%, and a Cohen's Kappa coefficient of 75.69%. For wrist classification, our approach achieved an accuracy of 85.58%, an F1-score of 82.70%, and a Cohen's Kappa coefficient of 70.46%. The results demonstrated that the models trained using our proposed TL approach outperformed those trained with ImageNet TL. We employed visualisation techniques to further validate these findings, including a gradient-based class activation heat map (Grad-CAM) and locally interpretable model-independent explanations (LIME). These visualisation tools provided additional evidence to support the superior accuracy of models trained with our proposed TL approach compared to those trained with ImageNet TL. Furthermore, our proposed TL approach exhibited greater robustness in various experiments compared to ImageNet TL. Importantly, the proposed TL approach and the feature-fusion technique are not limited to specific tasks. They can be applied to various medical image applications, thus extending their utility and potential impact. To demonstrate the concept of reusability, a computed tomography (CT) case was adopted. The results obtained from the proposed method showed improvements.

A Robust and Explainable Structure-Based Algorithm for Detecting the Organ Boundary From Ultrasound Multi-Datasets

Detecting the organ boundary in an ultrasound image is challenging because of the poor contrast of ultrasound images and the existence of imaging artifacts. In this study, we developed a coarse-to-refinement architecture for multi-organ ultrasound segmentation. First, we integrated the principal curve-based projection stage into an improved neutrosophic mean shift-based algorithm to acquire the data sequence, for which we utilized a limited amount of prior seed point information as the approximate initialization. Second, a distribution-based evolution technique was designed to aid in the identification of a suitable learning network. Then, utilizing the data sequence as the input of the learning network, we achieved the optimal learning network after learning network training. Finally, a scaled exponential linear unit-based interpretable mathematical model of the organ boundary was expressed via the parameters of a fraction-based learning network. The experimental outcomes indicated that our algorithm 1) achieved more satisfactory segmentation outcomes than state-of-the-art algorithms, with a Dice score coefficient value of 96.68 ± 2.2%, a Jaccard index value of 95.65 ± 2.16%, and an accuracy of 96.54 ± 1.82% and 2) discovered missing or blurry areas.

Implications of predicting race variables from medical images

AI-predicted race variables pose risks and opportunities for studying health disparities.

Converting Nanotoxicity Data to Information Using Artificial Intelligence and Simulation

Decades of nanotoxicology research have generated extensive and diverse data sets. However, data is not equal to information. The question is how to extract critical information buried in vast data streams. Here we show that artificial intelligence (AI) and molecular simulation play key roles in transforming nanotoxicity data into critical information, i.e., constructing the quantitative nanostructure (physicochemical properties)-toxicity relationships, and elucidating the toxicity-related molecular mechanisms. For AI and molecular simulation to realize their full impacts in this mission, several obstacles must be overcome. These include the paucity of high-quality nanomaterials (NMs) and standardized nanotoxicity data, the lack of model-friendly databases, the scarcity of specific and universal nanodescriptors, and the inability to simulate NMs at realistic spatial and temporal scales. This review provides a comprehensive and representative, but not exhaustive, summary of the current capability gaps and tools required to fill these formidable gaps. Specifically, we discuss the applications of AI and molecular simulation, which can address the large-scale data challenge for nanotoxicology research. The need for model-friendly nanotoxicity databases, powerful nanodescriptors, new modeling approaches, molecular mechanism analysis, and design of the next-generation NMs are also critically discussed. Finally, we provide a perspective on future trends and challenges.

Multimodal data fusion for cancer biomarker discovery with deep learning

Technological advances now make it possible to study a patient from multiple angles with high-dimensional, high-throughput multi-scale biomedical data. In oncology, massive amounts of data are being generated ranging from molecular, histopathology, radiology to clinical records. The introduction of deep learning has significantly advanced the analysis of biomedical data. However, most approaches focus on single data modalities leading to slow progress in methods to integrate complementary data types. Development of effective multimodal fusion approaches is becoming increasingly important as a single modality might not be consistent and sufficient to capture the heterogeneity of complex diseases to tailor medical care and improve personalised medicine. Many initiatives now focus on integrating these disparate modalities to unravel the biological processes involved in multifactorial diseases such as cancer. However, many obstacles remain, including lack of usable data as well as methods for clinical validation and interpretation. Here, we cover these current challenges and reflect on opportunities through deep learning to tackle data sparsity and scarcity, multimodal interpretability, and standardisation of datasets.

A multi-institutional study using artificial intelligence to provide reliable and fair feedback to surgeons

BACKGROUND

Surgeons who receive reliable feedback on their performance quickly master the skills necessary for surgery. Such performance-based feedback can be provided by a recently-developed artificial intelligence (AI) system that assesses a surgeon's skills based on a surgical video while simultaneously highlighting aspects of the video most pertinent to the assessment. However, it remains an open question whether these highlights, or explanations, are equally reliable for all surgeons.

METHODS

Here, we systematically quantify the reliability of AI-based explanations on surgical videos from three hospitals across two continents by comparing them to explanations generated by humans experts. To improve the reliability of AI-based explanations, we propose the strategy of training with explanations -TWIX -which uses human explanations as supervision to explicitly teach an AI system to highlight important video frames.

RESULTS

We show that while AI-based explanations often align with human explanations, they are not equally reliable for different sub-cohorts of surgeons (e.g., novices vs. experts), a phenomenon we refer to as an explanation bias. We also show that TWIX enhances the reliability of AI-based explanations, mitigates the explanation bias, and improves the performance of AI systems across hospitals. These findings extend to a training environment where medical students can be provided with feedback today.

CONCLUSIONS

Our study informs the impending implementation of AI-augmented surgical training and surgeon credentialing programs, and contributes to the safe and fair democratization of surgery.

Analysis: Flawed Datasets of Monkeypox Skin Images

The self-proclaimed first publicly available dataset of Monkeypox skin images consists of medically irrelevant images extracted from Google and photography repositories through a process denominated web-scrapping. Yet, this did not stop other researchers from employing it to build Machine Learning (ML) solutions aimed at computer-aided diagnosis of Monkeypox and other viral infections presenting skin lesions. Neither did it stop the reviewers or editors from publishing these subsequent works in peer-reviewed journals. Several of these works claimed extraordinary performance in the classification of Monkeypox, Chickenpox and Measles, employing ML and the aforementioned dataset. In this work, we analyse the initiator work that has catalysed the development of several ML solutions, and whose popularity is continuing to grow. Further, we provide a rebuttal experiment that showcases the risks of such methodologies, proving that the ML solutions do not necessarily obtain their performance from the features relevant to the diseases at issue.

Non-coding deep learning models for tomato biotic and abiotic stress classification using microscopic images

Plant disease classification is quite complex and, in most cases, requires trained plant pathologists and sophisticated labs to accurately determine the cause. Our group for the first time used microscopic images (×30) of tomato plant diseases, for which representative plant samples were diagnostically validated to classify disease symptoms using non-coding deep learning platforms (NCDL). The mean F1 scores (SD) of the NCDL platforms were 98.5 (1.6) for Amazon Rekognition Custom Label, 93.9 (2.5) for Clarifai, 91.6 (3.9) for Teachable Machine, 95.0 (1.9) for Google AutoML Vision, and 97.5 (2.7) for Microsoft Azure Custom Vision. The accuracy of the NCDL platform for Amazon Rekognition Custom Label was 99.8% (0.2), for Clarifai 98.7% (0.5), for Teachable Machine 98.3% (0.4), for Google AutoML Vision 98.9% (0.6), and for Apple CreateML 87.3 (4.3). Upon external validation, the model's accuracy of the tested NCDL platforms dropped no more than 7%. The potential future use for these models includes the development of mobile- and web-based applications for the classification of plant diseases and integration with a disease management advisory system. The NCDL models also have the potential to improve the early triage of symptomatic plant samples into classes that may save time in diagnostic lab sample processing.

Open Data and transparency in artificial intelligence and machine learning: A new era of research

Artificial Intelligence (AI) and machine learning are the current forefront of computer science and technology. AI and related sub-disciplines, including machine learning, are essential technologies which have enabled the widespread use of smart technology, such as smart phones, smart home appliances and even electric toothbrushes. It is AI that allows the devices used day-to-day across people's personal lives, working lives and in industry to better anticipate and respond to our needs. However, the use of AI technology comes with a range of ethical questions - including issues around privacy, security, reliability, copyright/plagiarism and whether AI is capable of independent, conscious thought. We have seen several issues related to racial and sexual bias in AI in the recent times, putting the reliability of AI in question. Many of these issues have been brought to the forefront of cultural awareness in late 2022, early 2023, with the rise of AI art programs (and the copyright issues arising from the deep-learning methods employed to train this AI), and the popularity of ChatGPT alongside its ability to be used to mimic human output, particularly in regard to academic work. In critical areas like healthcare, the errors of AI can be fatal. With the incorporation of AI in almost every sector of our everyday life, we need to keep asking ourselves- can we trust AI, and how much? This Editorial outlines the importance of openness and transparency in the development and applications of AI to allow all users to fully understand both the benefits and risks of this ubiquitous technology, and outlines how the Artificial Intelligence and Machine Learning  Gateway on F1000Research meets these needs.

Simulation of a machine learning enabled learning health system for risk prediction using synthetic patient data

When enabled by machine learning (ML), Learning Health Systems (LHS) hold promise for improving the effectiveness of healthcare delivery to patients. One major barrier to LHS research and development is the lack of access to EHR patient data. To overcome this challenge, this study demonstrated the feasibility of developing a simulated ML-enabled LHS using synthetic patient data. The ML-enabled LHS was initialized using a dataset of 30,000 synthetic Synthea patients and a risk prediction XGBoost base model for lung cancer. 4 additional datasets of 30,000 patients were generated and added to the previous updated dataset sequentially to simulate addition of new patients, resulting in datasets of 60,000, 90,000, 120,000 and 150,000 patients. New XGBoost models were built in each instance, and performance improved with data size increase, attaining 0.936 recall and 0.962 AUC (area under curve) in the 150,000 patients dataset. The effectiveness of the new ML-enabled LHS process was verified by implementing XGBoost models for stroke risk prediction on the same Synthea patient populations. By making the ML code and synthetic patient data publicly available for testing and training, this first synthetic LHS process paves the way for more researchers to start developing LHS with real patient data.

Threats to Terrestrial Plants from Emerging Nanoplastics

Nanoplastics are ubiquitous in ecosystems and impact planetary health. However, our current understanding on the impacts of nanoplastics upon terrestrial plants is fragmented. The lack of systematic approaches to evaluating these impacts limits our ability to generalize from existing studies and perpetuates regulatory barriers. Here, we undertook a meta-analysis to quantify the overall strength of nanoplastic impacts upon terrestrial plants and developed a machine learning approach to predict adverse impacts and identify contributing features. We show that adverse impacts are primarily associated with toxicity metrics, followed by plant species, nanoplastic mass concentration and size, and exposure time and medium. These results highlight that the threats of nanoplastics depend on a diversity of reactions across molecular to ecosystem scales. These reactions are rooted in both the spatial and functional complexities of nanoplastics and, as such, are specific to both the plastic characteristics and environmental conditions. These findings demonstrate the utility of interrogating the diversity of toxicity data in the literature to update both risk assessments and evidence-based policy actions.