Hasil untuk "Computer applications to medicine. Medical informatics"

Cristian Valero-Abundio, Emilio Sansano-Sansano, Raúl Montoliu et al.

Handling geometric transformations, particularly rotations, remains a challenge in deep learning for computer vision. Standard neural networks lack inherent rotation invariance and typically rely on data augmentation or architectural modifications to improve robustness. Although effective, these approaches increase computational demands, require specialised implementations, or alter network structures, limiting their applicability. This paper introduces General Intensity Direction (GID), a preprocessing method that improves rotation robustness without modifying the network architecture. The method estimates a global orientation for each image and aligns it to a canonical reference frame, allowing standard models to process inputs more consistently across different rotations. Unlike moment-based approaches that extract invariant descriptors, this method directly transforms the image while preserving spatial structure, making it compatible with convolutional networks. Experimental evaluation on the rotated MNIST dataset shows that the proposed method achieves higher accuracy than state-of-the-art rotation-invariant architectures. Additional experiments on the CIFAR-10 dataset, confirm that the method remains effective under more complex conditions.

Ruijie Tang, Peiqi Zhu, Ruxue Yan et al.

Objectives Urinary tract infections (UTIs) frequently affect individuals of all ages, necessitating antibiotic treatment and medical care, which can impair quality of life and cause psychological strain. Online Health Consultation (OHC) platforms serve as a widely used communication tool, offering integrated support for medical guidance and disease management. By examining OHC interactions, this study explores the concerns and difficulties experienced by UTI patients to better understand their perspectives. Methods Data from 20,000 anonymized UTI-related records (2020–2024) were obtained from a major Chinese online healthcare platform, Good Doctor Online. Analysis occurred in two stages: BERTopic extracted key themes and keywords from text data, followed by sentiment analysis of these findings using a generative AI language model. All data was publicly accessible and de-identified. Results Analysis of 18,479 cleaned records using BERTopic identified six key themes: “Polite Expressions for Consultation,” “Symptom and Management Challenges,” “Differential Diagnosis of Cystitis,” “Etiology Related to Sexual Activity,” “Nocturnal Symptoms and Fever,” and “Perinatal Considerations.” Sentiment analysis showed predominantly negative emotions, reflecting the condition's substantial physical and mental toll. The “Etiology Related to Sexual Activity” theme had the highest negativity (97%), while “Polite Expressions for Consultation” showed the most positivity (9%). Conclusion These research results highlight the important role of online communities in providing support and information to patients, and the insights derived from this study can provide valuable reference for social media developers, medical service providers, and policymakers.

Na Li, Jane Courtney, Robert Ross

The dataset was collected from 28 participants (17 female, 9 male, and 1 non-binary) for a study aimed at modelling and detecting user social behaviours with different confusion states in task-oriented situated human-robot interaction (HRI). The dataset consists of user facial body video recordings synchronised with user speech across three designed experiment scenarios (Tasks 1 - 3). Each experiment lasted approximately one hour per participant. The videos are segmented into individual clips corresponding to specific experimental conversations under predefined conditions: general confusion and non-confusion for Task 1 and 3; and productive confusion, unproductive confusion, and non-confusion for Task 2.In total, the dataset contains 789 video clips (body: 392, face: 397). Each video is recorded in high-definition RGB format, capturing user facial expressions or body language along with their speech. These multimodal data provide a valuable resource for studying user cognitive and mental states in human-robot interaction and human-computer interaction.The data collected for Task 2 was used in [9]. In compliance with GDPR (General Data Protection Regulation) and DPIA (data protection impact assessment) guidelines, the dataset is freely available upon request at https://sites.google.com/view/hridatarequst/home.

Jianhong Luo, Chaoqi Jin

Abstract Social bots pose a significant threat to online platforms, demanding robust methods to detect their increasingly complex behaviors. This paper introduces MM-HGT-Bot, a multi-modal framework that advances the field by operationalizing social network theory in a new way. Our core contribution is the deconstruction of social ties into two distinct, theoretically-grounded dimensions: information source selection (the following network) and potential influence (the follower network). Our architecture employs a Heterogeneous Graph Transformer (HGT) to learn the unique patterns emerging from these different relationship types. It then synergistically fuses these relational insights with context-aware representations of user-generated content. Extensive experiments on the widely-used Cresci-15 and Twibot-20 datasets demonstrate that our approach consistently outperforms state-of-the-art baselines. These findings highlight that a more fine-grained and theoretically-informed modeling of social relationships is crucial for building effective and robust bot detection systems.

Yitao Zhu, Yuan Yin, Jiaming Li et al.

The adoption of visual foundation models has become a common practice in computer-aided diagnosis (CAD). While these foundation models provide a viable solution for creating generalist medical AI, privacy concerns make it difficult to pre-train or continuously update such models across multiple domains and datasets, leading many studies to focus on specialist models. To address this challenge, we propose Med-LEGO, a training-free framework that enables the seamless integration or updating of a generalist CAD model by combining multiple specialist models, similar to assembling LEGO bricks. Med-LEGO enhances LoRA (low-rank adaptation) by incorporating singular value decomposition (SVD) to efficiently capture the domain expertise of each specialist model with minimal additional parameters. By combining these adapted weights through simple operations, Med-LEGO allows for the easy integration or modification of specific diagnostic capabilities without the need for original data or retraining. Finally, the combined model can be further adapted to new diagnostic tasks, making it a versatile generalist model. Our extensive experiments demonstrate that Med-LEGO outperforms existing methods in both cross-domain and in-domain medical tasks while using only 0.18% of full model parameters. These merged models show better convergence and generalization to new tasks, providing an effective path toward generalist medical AI.

Lin Wang, Jiaming Su, Zhongjie Liu et al.

Abstract Background Diabetic nephropathy (DN) is a major microvascular complication of diabetes and has become the leading cause of end-stage renal disease worldwide. A considerable number of DN patients have experienced irreversible end-stage renal disease progression due to the inability to diagnose the disease early. Therefore, reliable biomarkers that are helpful for early diagnosis and treatment are identified. The migration of immune cells to the kidney is considered to be a key step in the progression of DN-related vascular injury. Therefore, finding markers in this process may be more helpful for the early diagnosis and progression prediction of DN. Methods The gene chip data were retrieved from the GEO database using the search term ' diabetic nephropathy ‘. The ' limma ' software package was used to identify differentially expressed genes (DEGs) between DN and control samples. Gene set enrichment analysis (GSEA) was performed on genes obtained from the molecular characteristic database (MSigDB. The R package ‘WGCNA’ was used to identify gene modules associated with tubulointerstitial injury in DN, and it was crossed with immune-related DEGs to identify target genes. Gene ontology (GO) enrichment analysis and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis were performed on differentially expressed genes using the ‘ClusterProfiler’ software package in R. Three methods, least absolute shrinkage and selection operator (LASSO), support vector machine recursive feature elimination (SVM-RFE) and random forest (RF), were used to select immune-related biomarkers for diagnosis. We retrieved the tubulointerstitial dataset from the Nephroseq database to construct an external validation dataset. Unsupervised clustering analysis of the expression levels of immune-related biomarkers was performed using the ‘ConsensusClusterPlus ‘R software package. The urine of patients who visited Dongzhimen Hospital of Beijing University of Chinese Medicine from September 2021 to March 2023 was collected, and Elisa was used to detect the mRNA expression level of immune-related biomarkers in urine. Pearson correlation analysis was used to detect the effect of immune-related biomarker expression on renal function in DN patients. Results Four microarray datasets from the GEO database are included in the analysis : GSE30122, GSE47185, GSE99340 and GSE104954. These datasets included 63 DN patients and 55 healthy controls. A total of 9415 genes were detected in the data set. We found 153 differentially expressed immune-related genes, of which 112 genes were up-regulated, 41 genes were down-regulated, and 119 overlapping genes were identified. GO analysis showed that they were involved in various biological processes including leukocyte-mediated immunity. KEGG analysis showed that these target genes were mainly involved in the formation of phagosomes in Staphylococcus aureus infection. Among these 119 overlapping genes, machine learning results identified AGR2, CCR2, CEBPD, CISH, CX3CR1, DEFB1 and FSTL1 as potential tubulointerstitial immune-related biomarkers. External validation suggested that the above markers showed diagnostic efficacy in distinguishing DN patients from healthy controls. Clinical studies have shown that the expression of AGR2, CX3CR1 and FSTL1 in urine samples of DN patients is negatively correlated with GFR, the expression of CX3CR1 and FSTL1 in urine samples of DN is positively correlated with serum creatinine, while the expression of DEFB1 in urine samples of DN is negatively correlated with serum creatinine. In addition, the expression of CX3CR1 in DN urine samples was positively correlated with proteinuria, while the expression of DEFB1 in DN urine samples was negatively correlated with proteinuria. Finally, according to the level of proteinuria, DN patients were divided into nephrotic proteinuria group (n = 24) and subrenal proteinuria group. There were significant differences in urinary AGR2, CCR2 and DEFB1 between the two groups by unpaired t test (P < 0.05). Conclusions Our study provides new insights into the role of immune-related biomarkers in DN tubulointerstitial injury and provides potential targets for early diagnosis and treatment of DN patients. Seven different genes ( AGR2, CCR2, CEBPD, CISH, CX3CR1, DEFB1, FSTL1 ), as promising sensitive biomarkers, may affect the progression of DN by regulating immune inflammatory response. However, further comprehensive studies are needed to fully understand their exact molecular mechanisms and functional pathways in DN.

Prasanth Tej Kumar Jagannadham, Thirugnanavel Anbalagan, Sonia Balyan et al.

Mandarin orange (Citrus reticulata Blanco) is the most common citrus fruit, covering nearly 42 % of the total citrus cultivation area in India. The main varieties of mandarin oranges cultivated in India include Nagpur Mandarin, Khasi Mandarin, Coorg Mandarin and Sikkim Mandarin. Globally, genomic data is being used to unravel the complexities and mysteries of citrus taxonomy. However, despite India being a primary centre of citrus origin, these valuable genomic resources remain underutilized. Here, we conducted whole genome resequencing of four mandarin genotypes viz., Nagpur Mandarin (22,861,254 bp raw reads), Sikkim Mandarin (24,160,847 bp raw reads), Coorg Mandarin (27,974,860 bp raw reads), and Khasi Mandarin (40,532,383 bp raw reads) using Illumina Novaseq 6000 sequencing platform with 28x sequencing coverage. These genomic sequences will provide valuable insights into the taxonomic complexities and evolutionary history of mandarin oranges. The identified SNPs can further be used to study the evolution of flowering patterns in citrus, especially under tropical and subtropical conditions. The NGS data obtained (FASTQ format) for all four mandarin genotypes have been deposited in the Indian Biological Data Centre (https://ibdc.dbtindia.gov.in/inda/submittedStudyHome) under INDA study Id INRP000149. The sample accession numbers are INS0004744 (Sikkim Mandarin), INS0004745 (Nagpur Mandarin), INS0004746 (Coorg Mandarin), INS0004747 (Khasi Mandarin).

Ziru Fu, Yu Cheng Hsu, Christian S Chan et al.

BackgroundSentiment analysis is a significant yet difficult task in natural language processing. The linguistic peculiarities of Cantonese, including its high similarity with Standard Chinese, its grammatical and lexical uniqueness, and its colloquialism and multilingualism, make it different from other languages and pose additional challenges to sentiment analysis. Recent advances in models such as ChatGPT offer potential viable solutions. ObjectiveThis study investigated the efficacy of GPT-3.5 and GPT-4 in Cantonese sentiment analysis in the context of web-based counseling and compared their performance with other mainstream methods, including lexicon-based methods and machine learning approaches. MethodsWe analyzed transcripts from a web-based, text-based counseling service in Hong Kong, including a total of 131 individual counseling sessions and 6169 messages between counselors and help-seekers. First, a codebook was developed for human annotation. A simple prompt (“Is the sentiment of this Cantonese text positive, neutral, or negative? Respond with the sentiment label only.”) was then given to GPT-3.5 and GPT-4 to label each message’s sentiment. GPT-3.5 and GPT-4’s performance was compared with a lexicon-based method and 3 state-of-the-art models, including linear regression, support vector machines, and long short-term memory neural networks. ResultsOur findings revealed ChatGPT’s remarkable accuracy in sentiment classification, with GPT-3.5 and GPT-4, respectively, achieving 92.1% (5682/6169) and 95.3% (5880/6169) accuracy in identifying positive, neutral, and negative sentiment, thereby outperforming the traditional lexicon-based method, which had an accuracy of 37.2% (2295/6169), and the 3 machine learning models, which had accuracies ranging from 66% (4072/6169) to 70.9% (4374/6169). ConclusionsAmong many text analysis techniques, ChatGPT demonstrates superior accuracy and emerges as a promising tool for Cantonese sentiment analysis. This study also highlights ChatGPT’s applicability in real-world scenarios, such as monitoring the quality of text-based counseling services and detecting message-level sentiments in vivo. The insights derived from this study pave the way for further exploration into the capabilities of ChatGPT in the context of underresourced languages and specialized domains like psychotherapy and natural language processing.

Tim J. M. Jaspers, Ronald L. P. D. de Jong, Yasmina Al Khalil et al.

Over the past decade, computer vision applications in minimally invasive surgery have rapidly increased. Despite this growth, the impact of surgical computer vision remains limited compared to other medical fields like pathology and radiology, primarily due to the scarcity of representative annotated data. Whereas transfer learning from large annotated datasets such as ImageNet has been conventionally the norm to achieve high-performing models, recent advancements in self-supervised learning (SSL) have demonstrated superior performance. In medical image analysis, in-domain SSL pretraining has already been shown to outperform ImageNet-based initialization. Although unlabeled data in the field of surgical computer vision is abundant, the diversity within this data is limited. This study investigates the role of dataset diversity in SSL for surgical computer vision, comparing procedure-specific datasets against a more heterogeneous general surgical dataset across three different downstream surgical applications. The obtained results show that using solely procedure-specific data can lead to substantial improvements of 13.8%, 9.5%, and 36.8% compared to ImageNet pretraining. However, extending this data with more heterogeneous surgical data further increases performance by an additional 5.0%, 5.2%, and 2.5%, suggesting that increasing diversity within SSL data is beneficial for model performance. The code and pretrained model weights are made publicly available at https://github.com/TimJaspers0801/SurgeNet.

Dingkang Yang, Jinjie Wei, Dongling Xiao et al.

Developing intelligent pediatric consultation systems offers promising prospects for improving diagnostic efficiency, especially in China, where healthcare resources are scarce. Despite recent advances in Large Language Models (LLMs) for Chinese medicine, their performance is sub-optimal in pediatric applications due to inadequate instruction data and vulnerable training procedures. To address the above issues, this paper builds PedCorpus, a high-quality dataset of over 300,000 multi-task instructions from pediatric textbooks, guidelines, and knowledge graph resources to fulfil diverse diagnostic demands. Upon well-designed PedCorpus, we propose PediatricsGPT, the first Chinese pediatric LLM assistant built on a systematic and robust training pipeline. In the continuous pre-training phase, we introduce a hybrid instruction pre-training mechanism to mitigate the internal-injected knowledge inconsistency of LLMs for medical domain adaptation. Immediately, the full-parameter Supervised Fine-Tuning (SFT) is utilized to incorporate the general medical knowledge schema into the models. After that, we devise a direct following preference optimization to enhance the generation of pediatrician-like humanistic responses. In the parameter-efficient secondary SFT phase, a mixture of universal-specific experts strategy is presented to resolve the competency conflict between medical generalist and pediatric expertise mastery. Extensive results based on the metrics, GPT-4, and doctor evaluations on distinct doctor downstream tasks show that PediatricsGPT consistently outperforms previous Chinese medical LLMs. Our model and dataset will be open-source for community development.

Enzo Ferrante, Rodrigo Echeveste

Recently, the research community of computerized medical imaging has started to discuss and address potential fairness issues that may emerge when developing and deploying AI systems for medical image analysis. This chapter covers some of the pressing challenges encountered when doing research in this area, and it is intended to raise questions and provide food for thought for those aiming to enter this research field. The chapter first discusses various sources of bias, including data collection, model training, and clinical deployment, and their impact on the fairness of machine learning algorithms in medical image computing. We then turn to discussing open challenges that we believe require attention from researchers and practitioners, as well as potential pitfalls of naive application of common methods in the field. We cover a variety of topics including the impact of biased metrics when auditing for fairness, the leveling down effect, task difficulty variations among subgroups, discovering biases in unseen populations, and explaining biases beyond standard demographic attributes.

Zahra Galavi, Reza Khajouei

Guillermo A. Farias-Basulto, Maximilian Riedel, Mark Khenkin et al.

This article provides datasets containing three years worth of solar spectra for the optimum installation angle of 35° and the building-integrated-photovoltaics relevant vertical angle of 90°. These datasets were obtained by measuring the spectrally resolved solar spectra using a five minute interval, where two sets of spectrometers, which measure different ranges of the solar spectrum, were employed. In addition, a merged dataset of these two spectral measurements, related to every specific five minute interval measurement, is provided. An analysis and interpretation of the data using only year the 2020 is provided in “Measurement and analysis of annual solar spectra at different installation angles in central Europe” [1].

Md Manjurul Ahsan, Shivakumar Raman, Zahed Siddique

Over the years, the Invariant Scattering Transform (IST) technique has become popular for medical image analysis, including using wavelet transform computation using Convolutional Neural Networks (CNN) to capture patterns' scale and orientation in the input signal. IST aims to be invariant to transformations that are common in medical images, such as translation, rotation, scaling, and deformation, used to improve the performance in medical imaging applications such as segmentation, classification, and registration, which can be integrated into machine learning algorithms for disease detection, diagnosis, and treatment planning. Additionally, combining IST with deep learning approaches has the potential to leverage their strengths and enhance medical image analysis outcomes. This study provides an overview of IST in medical imaging by considering the types of IST, their application, limitations, and potential scopes for future researchers and practitioners.

N. D. Tran, C. Leung, Evan W. R. Madill et al.

Healthcare informatics is an interdisciplinary area where computer science, data science, cognitive science, informatics principles, and information technology meet to address problems and support healthcare, medicine, public health, and/or everyday wellness. In many medical and healthcare applications, having models that can learn from historical healthcare data or instances to make predictions on future instances is helpful. However, partially due to privacy issues, the availability of healthcare data to be learned may be limited. Hence, in this paper, we present a deep learning based predictive model for healthcare analytics. In particular, our model consists of an autoencoder (comprising an encoder and a decoder) and a predictor to make accurate predictions. It can learn from a few shots of historical healthcare data to make either binary or multi-label predictions. Evaluation results on real-life datasets demonstrates the effectiveness of our deep learning-based predictive model in supporting healthcare analytics.

C. Leung, Evan W. R. Madill, Joglas Souza et al.

Healthcare informatics is an interdisciplinary area where computer science, data science, cognitive science, informatics principles, and information technology meet to address problems and support healthcare, medicine, public health, and/or everyday wellness. In many healthcare and medical applications, it is helpful to have models that can learn from historical healthcare data or instances to make predictions on future instances. For human to trust these models or to perceive these models to be trustworthy, it is equally important to build a trustworthy artificial intelligence (AI) solution. Hence, in this paper, towards trustworthy AI in healthcare, we present an explainable AI (XAI) solution that makes accurate predictions and explains the predictions. Evaluation results on real-life datasets demonstrates the effectiveness of our XAI solution towards trustworthy AI in healthcare.

Mary Yang, Brian D. Athey, H. Arabnia et al.

C. Leung, Thanh Huy Daniel Mai, N. D. Tran

Health informatics is an interdisciplinary area where computer science and related disciplines meet to address problems and support healthcare and medicine. In particular, computer has played an important role in medicine. Many existing computer-based systems (e.g., machine learning models) for healthcare applications produce binary prediction (e.g., whether a patient catches a disease or not). However, there are situations in which a non-binary prediction (e.g., what is hospitalization status of a patient) is needed. As a concrete example, over the past two years, people around the world have been affected by the coronavirus disease 2019 (COVID-19) pandemic. There have been works on binary prediction to determine whether a patient is COVID-19 positive or not. With availability of alternative methods (e.g., rapid test), such a binary prediction has become less important. Moreover, with the evolution of the disease (e.g., recent development of COVID-19 Omicron variant), multi-label prediction of the hospitalization status has become more important when compared with binary prediction on the confirmation of cases. Hence, in this paper, we present a multi-label prediction system for computer-based medical applications. Our system makes use of autoencoders (consisting of encoders and decoders) and few-shot learning to predict the hospitalization status (e.g., ICU, semi-ICU, regular wards, or no hospitalization). The prediction is important for allocation of medical resources (e.g., hospital facilities and medical staff), which in turn affect patient lives. Experimental results on real-life open datasets show that, when training with only a few data, our multilabel prediction system gave a high F1-score when predicting hospitalization status of COVID-19 cases.

Yuliya Shapovalova, Tom Heskes, Tjeerd Dijkstra

Abstract Background Understanding the synergetic and antagonistic effects of combinations of drugs and toxins is vital for many applications, including treatment of multifactorial diseases and ecotoxicological monitoring. Synergy is usually assessed by comparing the response of drug combinations to a predicted non-interactive response from reference (null) models. Possible choices of null models are Loewe additivity, Bliss independence and the recently rediscovered Hand model. A different approach is taken by the MuSyC model, which directly fits a generalization of the Hill model to the data. All of these models, however, fit the dose–response relationship with a parametric model. Results We propose the Hand-GP model, a non-parametric model based on the combination of the Hand model with Gaussian processes. We introduce a new logarithmic squared exponential kernel for the Gaussian process which captures the logarithmic dependence of response on dose. From the monotherapeutic response and the Hand principle, we construct a null reference response and synergy is assessed from the difference between this null reference and the Gaussian process fitted response. Statistical significance of the difference is assessed from the confidence intervals of the Gaussian process fits. We evaluate performance of our model on a simulated data set from Greco, two simulated data sets of our own design and two benchmark data sets from Chou and Talalay. We compare the Hand-GP model to standard synergy models and show that our model performs better on these data sets. We also compare our model to the MuSyC model as an example of a recent method on these five data sets and on two-drug combination screens: Mott et al. anti-malarial screen and O’Neil et al. anti-cancer screen. We identify cases in which the HandGP model is preferred and cases in which the MuSyC model is preferred. Conclusion The Hand-GP model is a flexible model to capture synergy. Its non-parametric and probabilistic nature allows it to model a wide variety of response patterns.

Marina Klanjčić, Laetitia Gauvin, Michele Tizzoni et al.

Abstract One of the targets of the UN Sustainable Development Goals is to substantially reduce the number of global deaths and injuries from road traffic collisions. To this aim, European cities adopted various urban mobility policies, which has led to a heterogeneous number of injuries across Europe. Monitoring the discrepancies in injuries and understanding the most efficient policies are keys to achieve the objectives of Vision Zero, a multi-national road traffic safety project that aims at zero fatalities or serious injuries linked to road traffic. Here, we identify urban features that are determinants of vulnerable road user safety through the analysis of inter-mode collision data across European cities. We first build up a data set of urban road crashes and their participants from 24 cities in 5 European countries, using the widely recommended KSI indicator (killed or seriously injured individuals) as a safety performance metric. Modelling the casualty matrices including road infrastructure characteristics and modal share distribution of the different cities, we observe that cities with the highest rates of walking and cycling modal shares are the safest for the most vulnerable users. Instead, a higher presence of low-speed limited roads seems to only significantly reduce the number of injuries of car occupants. Our results suggest that policies aimed at increasing the modal share of walking and cycling are key to improve road safety for all road users.