Milestone Transactions on Medical Technometrics

Stress Monitoring With Heart Rate Variability Using Deep Learning

B K Preethika Raj — 2025-04-23

Prolonged stress can lead to mental health issues like anxiety and sleep disorders. Heart Rate Variability (HRV) serves as a key physiological marker for stress detection. Unlike heart rate, HRV measures the variation in time intervals between heartbeats (RR intervals). This study proposes a CNN-based model for classifying stress into no stress, interruption stress, and time pressure stress using HRV features. Evaluated on the SWELL-KW dataset, the model achieves 99.9% accuracy, outperforming existing methods. Feature extraction techniques, such as ANOVA, further validate the significance of HRV features in stress detection.

Securing Patient Data With Blockchain Enabled Federated Learning For Medical Diagnostics

S Nadiya — 2025-04-23

The proposed framework introduces a groundbreaking solution in healthcare by integrating Federated Learning (FL) with blockchain technology for diagnosing lung diseases. Traditional machine learning models often rely on centralized data collection, raising concerns about patient data privacy and potential misuse of sensitive information. FL tackles this issue by allowing hospitals to collaborate in training machine learning models without sharing raw patient data. Each hospital processes its local data independently, sending encrypted model updates instead, ensuring that data privacy is preserved while fostering collective innovation. To further strengthen this approach, blockchain technology is employed to securely encrypt and immutably store the shared model updates, creating a transparent and tamper-proof system. This combination not only addresses privacy concerns but also builds trust and accountability among participating hospitals. Significantly, the framework empowers smaller hospitals and under-resourced medical centers to contribute to and benefit from advanced diagnostic capabilities. By pooling resources through FL and ensuring equitable access via blockchain, it reduces disparities between healthcare providers, enabling improved diagnostics in distant or underserved areas. To the best of your knowledge, this represents the first practical implementation of blockchain-empowered FL on such diverse medical data, making a substantial contribution to the integration of artificial intelligence, blockchain, and healthcare. With its potential to revolutionize collaborative diagnostics while prioritizing privacy and security, this framework sets a new standard for technological innovation in medicine

Early Detection of Autism Spectrum Disorder using Transfer Learning on Brain Imaging Data

K Sai Sree — 2025-04-23

This study focuses on classifying and representing learning tasks using powerful deep learning models, including Convolutional Neural Networks (CNN) and Transfer Learning algorithms. The analysis utilizes data from the Autism Brain Imaging Data Exchange (ABIDE I and ABIDE II) datasets. It explores the application of deep learning techniques to enhance the detection of Autism Spectrum Disorder (ASD). Functional magnetic resonance imaging (fMRI) data has shown potential in identifying brain irregularities associated with ASD. We utilize Convolutional Neural Networks (CNNs) combined with transfer learning to analyse fMRI data from the Autism Brain Imaging Data Exchange (ABIDE) dataset. Our optimized CNN model achieves an accuracy of 81%, surpassing conventional classification models. This research establishes deep learning as a promising tool for ASD diagnosis, evaluated based on accuracy, precision, and recall

Leveraging Stacked Machine Learning Models to Advance Diagnostic Precision and Predictive Insights in Chronic Kidney Disease

Vishal Kumar Jaiswal — 2025-06-26

A considerable number of people worldwide suffer from chronic kidney disease (CKD), a progressive illness that often goes undiagnosed until it has advanced to a severe degree. Timely action and better patient outcomes rely on early diagnosis. However, traditional diagnostic methods are time-consuming and may lack consistency, especially in resource-constrained settings. To address this, our study explores the application of machine learning (ML) for early-stage CKD prediction. We used the publicly available CKD dataset from the UCI Machine Learning Repository, which includes 400 patient records across 14 clinical features. After performing thorough preprocessing—including handling missing values and converting categorical data—we applied multiple ML classifiers: Naïve Bayes, Decision Tree, Gradient Boosting, AdaBoost, and XGBoost. Each model was evaluated using 10-fold cross-validation to ensure reliability. The core of our approach lies in a stacking ensemble model, which combines predictions from three base learners—Naïve Bayes, Decision Tree, and Gradient Boosting—and passes them to a meta-learner based on AdaBoost. This layered learning framework was specifically designed to mitigate overfitting, which is evident in some individual models during evaluation. The proposed stacking model demonstrated the best performance among all tested models, achieving a training accuracy of 99.2% and a testing accuracy of 98.93%. Furthermore, it yielded a precision of 98.70%, a recall of 98.10%, and an F1-score of 98.40%, outperforming other standalone algorithms in accuracy and stability.

A Hybrid ANN and XGBoost Approach to Urban Air Quality Classification

Posina Anusha — 2025-08-18

Public health is severely threatened by urban air pollution, particularly in densely populated and rapidly industrialising cities. For risk management, environmental monitoring, and targeted policymaking, urban zones must be accurately classified according to pollution levels. This paper proposes a classification framework that integrates the ensemble-based decision-making power of Gradient Boosting With the nonlinear feature extraction capabilities of neural networks, a publicly available dataset containing over 52,000 daily air quality records from six major cities was used. The model was designed to distinguish between industrial and residential urban areas based on six major pollutants: PM2.5, PM10, CO, NO₂, SO₂, and O₃. The proposed two-stage architecture first transforms input features through an ANN to capture complex pollutant interactions, then feeds the learned representations into an XGBoost classifier for final prediction. The performance of this hybrid model was compared to that of several well-known classifiers, including standalone ANN, standalone XGBoost, Support Vector Machine, and Logistic Regression. With an accuracy of 99.98%, the suggested ANN–XGBoost model outperformed all baseline techniques. At 99.92%, 99.96%, and 99.98%, respectively, precision, recall, and F1-score were likewise exceptionally high, demonstrating exceptional classification performance and generalization ability.

Improving Electronic Health Records with NLP and LLM-RAG: A Scalable AI Method for Processing Medical Data

Syeda Amena — 2025-08-18

The rapid adoption of Artificial Intelligence (AI) has transformed Electronic Health Records (EHRs) for clinical decision-making, yet traditional systems suffer from poor contextual awareness, slow retrieval, and limited adaptability to real-time medical updates. To overcome these challenges, this study proposes an AI-powered healthcare assistant leveraging Retrieval-Augmented Generation (RAG) in Large Language Models (LLMs). Unlike existing chatbots that face issues with factual consistency, outdated data, and inefficient information retrieval, the proposed system integrates Groq LLaMA 3.1 (LLM), Qdrant (vector database), Hugging Face E5-large-v2 (embeddings), Tavily API (real-time search), and Supabase (authentication & storage) to provide a comprehensive solution. Through semantic search, AI-driven summarization, and dynamic access to reliable sources, the assistant significantly improves response accuracy, document search efficiency, and adaptability to evolving medical guidelines. Experimental results highlight enhanced decision support, automation, and patient care, underscoring the potential of AI-driven EHR systems to improve interactivity, intelligence, and accessibility in healthcare while enabling better real-time clinical outcomes.

Machine Learning Hybrid Models for Early Cervical Cancer Detection – A Comparative Study

Geeta C Mara — 2025-05-30

Cervical cancer arises as a result of the uncontrolled growth of abnormal cells in the cervix, usually caused by chronic infection with high-risk types of Human PapillomaVirus (HPV). Early detection and prevention can be facilitated through regular screening and HPV vaccination. This study proposes an enhanced ML Hybrid Model focused on the early spotting of Cervical Cancer using the best available Machine Learning techniques. To solve a primary challenge in diagnosing accuracy and precision in Cervical Cancer, the model utilizes AdaBoost, XGBoost, Stacking Classifiers, and Logistic Regression. The Hybrid Model uses ensemble methods such as AdaBoost and XGBoost, which improves productivity by properly integrating poor learners. Stacking Classifiers increases accuracy even more by incorporating the predictions of several models while Logistic Regression adds interpretability as well as reliability to the results. Collectively, these approaches form a model that produces low false positive rates alongside low false negative rates for early detection of the disease. This study focuses on the effects that Machine Learning can have in treating advanced stages of healthcare issues especially relating to early detection of Cervical cancer. The combination of sophisticated computational methods with clinical data sets presents an effective, globally relevant solution to urgent health concerns.

Precision Medicine: Innovations and Challenges

Jitender Bisht — 2025-06-30

Precision medicine is an emerging approach for disease prevention and treatment that will take into account individual heterogeneity in genes, environment, and lifestyle for each person. The paper takes stock of the state of precision medicine as it is applied in different specialties, such as oncology, neurology, and managing chronic conditions. The first aim is to find major technological and methodological innovations for improved personalized treatment approaches and address ethical issues of accessibility and fairness. We conduct a large-scale literature review to explore the latest research literature focused on the use of artificial intelligence and genomic data for clinical practice. Based on the analysis conducted, the results revealed that although there has been substantial progress in utilizing these technologies to enhance patient outcomes advances, equitable access to precision medicine solutions remains a challenge. However, it highlights the necessity of further research to explore ethical architectures that will serve the equitable realization of precision medicine practices.