Raj Gururajan

Background: Documentation in oncologist Electronic Health Records (EHRs) plays a critical role in communication, shared decision-making, and the detection of adverse effects all of which influence treatment concordance and adherence. However, narrative content is often incomplete, delayed, or written in formal styles that obscure patient priorities. Methods: Following PRISMA-ScR guidelines, we conducted a scoping review of studies published between 2013 and early 2024 that used natural language processing (NLP) or text mining on oncologist notes, or qualitatively examined EHR use in oncology. Data were charted by topic (e.g., adverse effects, note style, stigma, workflow burden) and synthesized using discourse analysis. Results: Twenty-three studies met inclusion criteria. Four clinician-side themes emerged: (1) compliance-oriented EHR design; (2) incomplete or delayed documentation of adverse effects, pain, and social determinants of health (SDOH); (3) formal or stigmatizing language; and (4) time and cognitive burden limiting person-centred narratives. These factors hinder concordance documentation and communication transparency. Conclusions: Improvements in EHR design, including person-centred prompts, plain-language templates, and audit-log-informed workflow changes, may enhance adherence by supporting better documentation and communication. Nursing and service leaders can implement these changes to promote trust, engagement, and continuity of care. Registration: Not registered.

Background: Treatment nonadherence in oncology is common. Surveys often miss why patients do not follow recommendations. We synthesised Natural Language Processing (NLP) studies, mainly sentiment analysis, of patient-generated content (social media, forums, blogs, review sites, and survey free text) to identify communication and relationship factors linked to nonadherence and concordance. Methods: We conducted a scoping review (PRISMA-ScR). Searches of PubMed, CINAHL, and Scopus from 2013 to 15 June 2024 identified eligible studies. We included 25 studies. Data were charted by source, cancer type, NLP technique, and adherence/concordance indicators, then synthesised via discourse analysis and narrative synthesis. Results: Four themes emerged: (1) unmet emotional needs; (2) suboptimal information and communication; (3) unclear concordance within person-centred care; and (4) misinformation dynamics and perceived clinician bias. Sentiment analysis helped identify emotions and information gaps that surveys often miss. Conclusions: Patient-voice data suggest practical actions for nursing, including routine distress screening, teach-back, misinformation countermeasures, and explicit concordance checks to improve adherence and shared decision making. Registration: Not registered.

Depression is a highly prevalent mental illness that poses significant challenges to global public health. Although various treatment options exist, their effectiveness is often limited by patient heterogeneity and the lack of individualized treatment strategies. Repetitive transcranial magnetic stimulation (rTMS) has emerged as a safe and noninvasive intervention for depression. However, current clinical practices typically rely on subjective or empirical decision-making and often lack personalized rTMS protocol selection, leading to suboptimal outcomes. To address this issue, we propose an artificial intelligence (AI)-powered model for personalized rTMS treatment recommendation. The model leverages multisource patient profiles, including demographic information, clinical depression data, and rTMS treatment parameters. Unlike conventional approaches, our method uses contrastive learning (CL) to enhance patient representation learning, enabling the model to better capture similarities between patient profiles. Treatment recommendations are generated by identifying the Top-$k$ most similar patients in the learned embedding space. Additionally, a similarity-based grouping strategy is introduced to divide the dataset into two subgroups, one receiving personalized rTMS protocols and the other receiving standard protocols. Experimental results show that BERT-based models achieve over 77.3% accuracy in Acc@Top-5. Furthermore, patients who received AI-suggested personalized rTMS protocols exhibited greater improvements in psychometric scale scores and larger effect sizes, demonstrating the effectiveness of our approach. Finally, we conduct an ablation study to examine the role of CL and systematically analyze the impact of different neural encoders and Top-$k$ values. These findings offer insights into model design and performance optimization.

Background: Treatment nonadherence in oncology is prevalent and often assessed by surveys that miss the qualitative reasons underpinning nonadherence or treatment discordance. We aimed to synthesize evidence from Natural Language Processing (NLP) studies, primarily sentiment analysis of patient generated content (social media, forums, blogs, review platforms, and survey free text), to identify communication-related and relational factors linked to nonadherence/concordance. Methods: We conducted a scoping review following PRISMA-ScR. Searches of PubMed, CINAHL, and Scopus (2013 early 2024) targeted NLP studies of the cancer patient experience; conventional reports were included where they clarified communication/adherence constructs. Data were charted against source, cancer type, NLP technique, and inferred adherence/concordance factors, then synthesized using discourse analysis and narrative synthesis. Results: Four patient side themes consistently emerged: (1) unmet emotional needs; (2) suboptimal information and communication; (3) unclear treatment concordance within patient/person-centred care; and (4) online misinformation dynamics (and perceptions of clinician bias). Sentiment analysis detected fine grained emotions (fear, disgust, sadness, surprise), informational gaps and trust issues that are often less visible in structured surveys. Conclusions: Patient voice data offer actionable insights for nursing practice: routine distress screening, teach back strategies, misinformation countermeasures, and explicit concordance checks. Integrating these into person-centred workflows may improve adherence and shared decision making. Registration: Not registered.

As more machine learning models are used in sensitive fields like healthcare, finance, and smart infrastructure, protecting structured tabular data from privacy attacks is a key research challenge. Although several privacy-preserving methods have been proposed for tabular data, a comprehensive comparison of their performance and trade-offs has yet to be conducted. We introduce and empirically assess a combined defense system that integrates differential privacy, federated learning, adaptive noise injection, hybrid cryptographic encryption, and ensemble-based obfuscation. The given strategies are analyzed on the benchmark tabular datasets (ADULT, GSS, FTE), showing that the suggested methods can mitigate up to 50 percent of model inversion attacks in relation to baseline models without decreasing the model utility (F1 scores are higher than 0.85). Moreover, on these datasets, our results match or exceed the latest state-of-the-art (SOTA) in terms of privacy. We also transform each defense into essential data privacy laws worldwide (GDPR and HIPAA), suggesting the best applicable guidelines for the ethical and regulation-sensitive deployment of privacy-preserving machine learning models in sensitive spaces.

Breast cancer is one of the leading causes of cancer-related morbidity worldwide, underscoring the need for advanced diagnostic tools to improve early detection and treatment outcomes. This study introduces MammoSegNet, a novel convolutional neural network architecture optimized for precisely segmenting mammographic images. The proposed MammoSegNet incorporates Inception-ResNet blocks, Squeeze-and-Excitation (SE) modules, and dilated convolutions to enable multi-scale feature extraction and efficient attention refinement while maintaining low computational complexity. MammoSegNet performance was rigorously evaluated on BCDR-D01 and INbreast datasets to examine its robustness and generalization. Using stratified fivefold cross-validation, the model was trained on BCDR-D01 and tested on the unseen INbreast dataset through Monte Carlo cross-validation. Preprocessing techniques, including Region of Interest (ROI) Isolation to concentrate on relevant areas, Normalization to standardized pixel intensities, and Data Augmentation to expand the dataset and enhance the model’s robustness, were employed. Additionally, a specialized image enhancement method called peak feature intensity transformation (PFIT) was designed to amplify diagnostic features while preserving structural integrity. Comparative evaluations confirmed MammoSegNet’s superior performance across metrics, achieving 97% accuracy on BCDR-D01 and 95% on INbreast. Statistical t-tests validated these improvements, and visual heatmaps demonstrated the model’s effectiveness in isolating tumor regions. These findings establish MammoSegNet as a promising tool for enhancing breast cancer diagnostic accuracy and reliability in medical applications.

The swift progression of technology has increased the complexity of cyber fraud, posing an escalating challenge for the banking sector to reliably and efficiently identify fraudulent credit card transactions. Conventional detection approaches fail to adapt to the advancing strategies of fraudsters, resulting in heightened false positives and inefficiency within fraud detection systems. This study overcomes these restrictions by creating an innovative stacking hybrid machine learning (ML) approach that combines decision trees (DT), random forests (RF), support vector machines (SVM), XGBoost, CatBoost, and logistic regression (LR) within a stacking ensemble framework. This method uses stacking to integrate diverse ML models, enhancing predictive performance, with a meta-model consolidating base model predictions, resulting in superior detection accuracy compared to any single model. The methodology utilizes sophisticated data preprocessing techniques, such as correlation-based feature selection and principal component analysis (PCA), to enhance computing efficiency while preserving essential information. Experimental assessments of a credit card transaction dataset reveal that the stacking ensemble model exhibits higher performance, achieving an F1-score of 88.14%, thereby efficiently balancing precision and recall. This outcome highlights the significance of ensemble methods such as stacking in attaining strong and dependable cyber fraud detection, emphasizing its capacity to markedly enhance the security of financial transactions.

This paper proposes a Human-AI collaborative framework for cybersecurity consulting tailored to the needs of small businesses, designed and implemented within a Master of Cybersecurity capstone program. The framework outlines a structured four-stage development model that integrates students into real-world consulting tasks while aligning with academic and industry objectives. Human–AI collaboration is embedded throughout the process, combining generative AI tools and domain-specific AI agents with human expertise to support the design, delivery, and refinement of consulting resources. The four stages include (1) AI agent development; (2) cybersecurity roadmap creation; (3) resource development; and (4) industry application. Each stage supports both development-oriented outputs—such as templates, training materials, and client deliverables—and research-oriented projects that explore design practices, collaboration models, and consulting strategies. This dual-track structure enables iterative learning and improvement while addressing educational standards and the evolving cybersecurity landscape for small businesses. This framework provides a scalable foundation for capstone-based consulting initiatives that bridge academic learning and industry impact through Human–AI collaboration.

The journal retracts and remove the article Triaging Medical Referrals Based on Clinical Prioritisation Criteria Using Machine Learning Techniques