Nov 2026· Computer Science and Information Technology· 0 citations· 24 references
TL;DR
CIMAS HealthMate is proposed, a hybrid multilingual VHA that integrates transformer-based natural language processing (NLP) with an explainable extreme gradient boosting (XGBoost) decision model to provide accurate and transparent symptom triage.
Abstract
Limited healthcare access, shortages of healthcare professionals, and linguistic diversity continue to impede timely symptom assessment and healthcare delivery in low-resource settings such as Zimbabwe. Existing virtual health assistant (VHAs) are frequently cloud-dependent, English-centric, and lack interpretable decision-making, limiting their effectiveness in bandwidth-constrained and privacy-sensitive environments. This study proposes CIMAS HealthMate, a hybrid multilingual VHA that integrates transformer-based natural language processing (NLP) with an explainable extreme gradient boosting (XGBoost) decision model to provide accurate and transparent symptom triage. The framework employs the no language left behind (NLLB) model for offline English–Shona translation, bidirectional encoder representations from transformers (BERT)-based models for intent classification and medical entity recognition, and XGBoost for structured triage recommendation. The system was evaluated using a multilingual symptom corpus and an anonymized electronic health record-style dataset comprising approximately 23,000 patient records. Experimental results achieved translation accuracies of 76.5% for Shona-to-English and 82.2% for English-to-Shona, symptom extraction accuracy of 86.6%, and end-to-end triage accuracy of 93.3% with an F1-score of 93.3%. These findings demonstrate that the proposed hybrid architecture effectively combines multilingual language understanding, interpretable machine learning, and offline deployment to deliver reliable and privacy-preserving triage support. The proposed approach provides a scalable and practical solution for improving equitable digital healthcare services in multilingual, resource-constrained environments.
In recent years, natural language processing has become an important tool in healthcare for extracting useful information from unstructured clinical text such as electronic health records, physician notes, and medical literature. Deep learning has significantly improved the performance of NLP systems, enabling stronger results in tasks such as disease prediction, clinical decision support, and patient risk assessment. However, healthcare NLP still faces major challenges in real-world deployment. Clinical text is often noisy, fragmented, and inconsistent, which can reduce model reliability. In addition, deep learning models lack transparency, which limits their adoption by clinicians who require explainable outputs for clinical decision-making. Privacy and security also remain major barriers because patient data is highly sensitive and subject to strict legal and ethical requirements. Bias in training data can further lead to uneven performance across patient populations. This paper combines a literature review with a healthcare-oriented case study to examine these issues in real-world settings. The findings show that although deep learning offers strong potential for healthcare analytics, progress depends on solving problems related to data quality, interpretability, privacy, and domain adaptation.
Madhurima Kommuru, Swathi Thatraju, Appala Nooka Kumar Doodala· International Journal of Mac...· 0 citations
Rural communities in India face ongoing challenges in accessing primary healthcare. These challenges include a lack of doctors, geographical isolation, and high consultation costs. This paper discusses an AI-assisted healthcare kiosk, a web-based, fully offline system that provides automated disease prediction, medication recommendations, BMI assessment, cardiovascular risk evaluation, and doctor referrals. It does all this without needing a permanent physician or internet connection. The system uses a Decision Tree classifier along with several machine learning models trained on a structured dataset of 4,920 samples, 132 symptom features, and 41 disease categories. It achieved 100% classification accuracy on the test set. The Flask-based web application features offline speech recognition through Kiosk, text-to-speech output using pyttsx3, TF-IDF vectorisation for natural language processing, and a MySQL backend for managing patient sessions. All ten functional test cases were successful, with an end-to-end response time of about one second. The proposed system proves it is possible to implement AI-powered primary healthcare tools in rural and resource-limited areas.
Thanu Shree M N, Vijayalakshmi M M· International Research Journ...· 0 citations
Access to healthcare remains a critical challenge across rural and semi-urban regions of India, where patients encounter difficulties such as limited availability of specialist doctors, communication barriers due to regional language diversity, absence of early screening tools, and fragmented systems for booking medical consultations. These issues often lead to delayed treatment, worsened health outcomes, and unnecessary travel to distant urban hospitals. Swasth AI addresses these challenges through an intelligent healthcare platform that enables users to describe symptoms via text or voice in seven Indian languages, receive AI-powered preliminary health assessments, identify suitable doctors through a multi-factor ranking system, and book appointments with automated email confirmations. Additionally, the platform incorporates a deep learning module for image-based skin condition classification using a Convolutional Neural Network trained on clinical dermatology data. The system architecture employs React 19 with Next.js 15 for the user interface, a Python-based REST API with PostgreSQL for data management, TensorFlow.js for in-browser model inference, the Web Speech API for multilingual voice recognition, and EmailJS for patient notifications. A geolocation-powered hospital mapping feature helps users identify nearby healthcare facilities without reliance on external map services. By combining NLP-driven symptom analysis, transfer learning for medical image classification, intelligent scheduling with patient registration, and an emergency alert mechanism, SwasthAI delivers an accessible, technology-driven healthcare experience designed to serve communities with limited medical infrastructure.
Pavithra Madipeddi, G. Narasimham· International Journal of Inn...· 0 citations
Healthcare systems increasingly depend on the structured exchange of information between hospitals, laboratories, and digital platforms. The HL7 v2.x standard provides the backbone for this communication but remains challenging for machine interpretation because of its variable syntax and optional segments. To address this limitation, a hybrid artificial intelligence model was developed for automated processing and classification of HL7 messages, integrating both structural learning and semantic validation. The experimental workflow included the generation of a synthetic dataset of 3,000 patient lifecycles with more than 7,000 ADT messages, followed by parsing, feature engineering, and supervised training. Logistic Regression, Random Forest, and Gradient Boosting were evaluated as baseline classifiers, while a semantic layer combining Named Entity Recognition and Regular Expressions introduced context-aware features such as physician names, medical facilities, and diagnosis indicators. After retraining, ensemble models demonstrated measurable improvement, with Random Forest achieving an increase of +9.3 % in accuracy and +7.0 % in F1-score. The results confirm that the addition of semantic cues enhances model interpretability and overall robustness, bridging the gap between structured message parsing and naturallanguage understanding. The proposed hybrid pipeline may serve as a foundation for intelligent interoperability solutions and future FHIR-compatible healthcare data systems.
T. M. Abilmazhinova, A. Kuatbayeva· Herald of Kazakh-British tec...· 0 citations
This paper presents the Integrated Predictive Intelligence Tool for Pediatric Health (IPITPH), a mobile-enabled AI-driven decision support system for addressing complex pediatric healthcare challenges in resource-limited, multilingual settings. The system comprises four integrated modules: a hybrid LSTM-DNN predictive analytics module incorporating culturally specific dietary patterns and temporal growth signals for pediatric risk prediction, an LLM-based nutrition optimization module generating personalized meal plans aligned with clinical guidelines and caregiver behavior, a multimodal emergency response module for real-time triage and teleconsultation, and a Retrieval-Augmented Generation-based multilingual conversational AI enabling voice-first caregiver interaction in Sinhala, Tamil, and English. Evaluated on pediatric data from 69 Sri Lankan families supplemented by the PIC clinical database, the predictive module achieved 3.18% MAPE for height prediction and 0.975 AUC for multi-domain risk classification. The nutrition module demonstrated high caloric precision and behavioral adaptation, the emergency module achieved robust triage accuracy with sub-100 ms edge latency, and the conversational module showed significant improvements in clinical accuracy and WHO guideline alignment across all three languages. These results establishing IPITPH as a promising unified AI framework for community-based pediatric healthcare in resource-limited regional contexts, pending broader prospective validation.
Munasing A. R. Tharana, Tharindu E. Nawarathne, Githadi S. Wijayarathna et al.· 2026 IEEE International Conf...· 0 citations
An engineering-oriented, end-to-end roadmap that structures the full lifecycle of clinical language model systems—from model design and domain adaptation to optimization and real-world evaluation is introduced.
A new method for surgically removing training examples from a model reveals that as datasets grow, the link between what a model learns and what it produces dissolves.
