Title: |
ArogyaNetra Mission: An AI & IoT-Based Smart Health Monitoring System for Proactive Population Health Surveillance Integrating Multimodal AI and IoT for National Health Equity |
Authors: |
Mayuri Thakkar, Anjali Jha |
Source: |
International Journal of Latest Engineering Research and Applications, pp 01 - 06, Vol 11 - No. 03, 2026 |
Abstract: |
Healthcare monitoring in India remains predominantly reactive rather than proactive, with significant portions of the population particularly senior citizens and rural communities lacking access to timely health surveillance or emergency alert systems. Hospitals and health authorities frequently operate without centralized real-time health data monitoring capabilities for population-level analytics. This paper presents ArogyaNetra Mission (सर्वजन आरोग्य ननरीक्षण आनण नीनन बुननमत्ता), an AI and IoT-based intelligent health surveillance system designed as a digital "eye" monitoring national health indicators in real time. Unlike conventional hospital monitoring systems that confine patients to clinical settings, ArogyaNetra employs a decentralized architecture integrating IoT sensors, cloud communication, and AI-driven analytics into a continuous, privacy-preserving health intelligence platform. The system introduces four transformative innovations: a Multimodal IoT Sensor Fusion Framework that integrates physiological parameters into unified health representations; an AI-Powered Health Monitoring Engine that analyzes temporal data patterns for early warning detection; a Multi-Tier Alert System delivering automated notifications via SMS/email to patients, family members, and health authorities; and a Conversational AI Health Assistant powered by large language models that provides empathetic, culturally competent guidance in multiple Indian languages. Evaluation using prototype testing with 150 participants across urban and rural Maharashtra demonstrates that ArogyaNetra achieves 94.2% sensor data transmission reliability, detects critical health deviations 3.8 hours faster than traditional monitoring, and generates automated alerts with 96.7% accuracy. We address implementation challenges and policy recommendations for scaling AI-driven preventive healthcare across India's diverse communities under the Digital India Health Mission framework. |
Kaywords: |
AI, IoT, preventive healthcare, health equity, large language models, national health surveillance, real-time monitoring. |
 |
Download Full Article |
DOI: |
10.56581/IJLERA.11.03.01-06 |
Title: |
HEAL-TX: A Generative AI-Powered Context-Aware Preventive Healthcare Integrating Multimodal AI for Proactive Health Equity |
Authors: |
Nusrat Khan, Anjali Shah |
Source: |
International Journal of Latest Engineering Research and Applications, pp 07 - 21, Vol 11 - No. 03, 2026 |
Abstract: |
Preventive healthcare faces a fundamental paradox: while 80% of chronic diseases are preventable, existing systems remain reactive, fragmented, and disconnected from individuals lived realities. This paper presents HEAL-TX (Holistic Ecosystem for Adaptive, Longitudinal, and Transparent Health), a generative AI-powered context-aware preventive healthcare system designed for statewide deployment in Texas. Unlike existing approaches that apply narrow AI models to isolated data streams, HEAL-TX employs a unified multimodal AI architecture that integrates physiological, behavioral, environmental, social, and healthcare system data into a continuous, privacy-preserving health intelligence platform. The system introduces four transformative AI innovations: (1) a Multimodal Fusion Transformer that learns cross-modal representations from heterogeneous temporal data without manual feature engineering; (2) a Generative Digital Twin Framework that creates personalized health simulations enabling counterfactual reasoning and "what-if" intervention planning; (3) a Federated Learning with Differential Privacy architecture that enables population-level model improvement without centralizing sensitive data; and (4) a Conversational AI Health Coach powered by fine-tuned large language models that delivers empathetic, culturally competent guidance in multiple languages. Evaluation using retrospective data from 50,000 patients and a prospective six-month pilot with 1,200 participants across three Texas counties demonstrates that HEAL-TX reduces preventable hospitalizations by 31.7% (surpassing our previous 28.3%), improves medication adherence by 44.2%, and achieves 91.3% user trust ratings. The paper details how AI enables: early warning systems that detect deterioration 5.2 days before clinical presentation, personalized intervention generation that adapts to individual preferences and constraints, and population health analytics that identify emerging outbreaks while preserving individual privacy. We address implementation challenges, partnership models, and policy recommendations for scaling AI-driven preventive healthcare across diverse communities. |
Kaywords: |
Generative AI, multimodal transformers, federated learning, preventive healthcare, context-aware systems, digital twins, health equity, large language models, state-level implementation. |
|
Download Full Article |
DOI: |
10.56581/IJLERA.11.03.07-21 |
Title: |
Weakly Supervised Concrete Crack Localization via Multi-Scale Feature Refinement |
Authors: |
Salman Shehzad, Aman Ullah, Chunsheng Yang, Muhammad Safi Ullah |
Source: |
International Journal of Latest Engineering Research and Applications, pp 22 - 29, Vol 11 - No. 03, 2026 |
Abstract: |
Concrete crack detection is essential for maintaining infrastructure integrity; however, manual inspection is time-consuming, subjective, and costly. Automated vision-based methods often rely on pixel-level annotations, which are labor intensive. Weakly supervised learning (WSL) addresses this limitation by using only image-level labels for defect localization. This paper proposes a novel WSL framework for concrete crack localization, addressing limitations in multi-scale feature representation and heatmap precision. The framework integrates: (1) a Multi-Scale Residual Feature Extractor (MS-RFE) to capture fine- and coarse-grained crack patterns across multiple scales; and (2) an Adaptive Heatmap Refinement (AHR) module to suppress background noise and improve localization accuracy. Experiments are conducted on a real-world dataset of 40,000 images using a ResNet-18 backbone. The proposed method achieves strong performance with an AUROC of 0.9882±0.0015, accuracy of 0.9847±0.0012, precision of 0.9835±0.0011, recall of 0.9871±0.0013, and F1-score of 0.9853±0.0010, outperforming baseline methods. Qualitative results demonstrate precise and low-noise crack localization, even for thin and low-contrast defects. The framework is lightweight and suitable for real-time deployment in infrastructure inspection. |
Kaywords: |
Concrete crack detection, weakly supervised learning, multi-scale feature extraction, heatmap refinement, defect localization. |
|
Download Full Article |
DOI: |
10.56581/IJLERA.11.03.22-29 |