Source-linked AI summary
Natural Language Processing for Smart Healthcare
Binggui Zhou, Guanghua Yang, Zheng Shi, Shaodan Ma
TL;DR
NLP for smart healthcare lacks a systematic synthesis across techniques and applications. This paper reviews NLP approaches, pipelines, healthcare scenarios, COVID-19, mental health, limitations, and future directions, concluding that NLP supports diverse smart-healthcare analyses while motivating more integrated systems.
Problem
Existing NLP-for-smart-healthcare research has lacked a well-organized systematic summary spanning technical approaches and healthcare applications.
Method
The paper reviews NLP approaches and pipelines from technical perspectives and surveys applications across clinical practice, management, personal care, public health, drug development, COVID-19, and mental health.
Results
The review identifies broad NLP use in smart healthcare, including COVID-19 forecasting and information needs, drug development, and mental-health screening and prediction.
Takeaways & Limitations
NLP-driven smart healthcare can support analysis and understanding of healthcare language across diverse scenarios and medical issues.
Takeaways & Limitations
Current applications often handle complex healthcare tasks step by step rather than integrating NLP end to end, leaving intermediate-component biases and pipeline complexity.
Abstract
from arXiv · showhide
Smart healthcare has achieved significant progress in recent years. Emerging artificial intelligence (AI) technologies enable various smart applications across various healthcare scenarios. As an essential technology powered by AI, natural language processing (NLP) plays a key role in smart healthcare due to its capability of analysing and understanding human language. In this work, we review existing studies that concern NLP for smart healthcare from the perspectives of technique and application. We first elaborate on different NLP approaches and the NLP pipeline for smart healthcare from the technical point of view. Then, in the context of smart healthcare applications employing NLP techniques, we introduce representative smart healthcare scenarios, including clinical practice, hospital management, personal care, public health, and drug development. We further discuss two specific medical issues, i.e., the coronavirus disease 2019 (COVID-19) pandemic and mental health, in which NLP-driven smart healthcare plays an important role. Finally, we discuss the limitations of current works and identify the directions for future works.
I. INTRODUCTION
Smart healthcare uses emerging technologies to connect healthcare participants and improve care, while NLP enables machines to analyze and understand human language across healthcare data and scenarios. The paper reviews NLP for smart healthcare systematically from technical and application perspectives, including COVID-19, mental health, limitations, and future directions.
- Smart healthcare context: Smart healthcare combines emerging technologies to connect the public, healthcare service providers, and third-party healthcare participants.Representative technologies include AI, blockchain, big data, cloud or edge computing, and the internet of things.
- NLP in healthcare: NLP concerns the automatic analysis, representation, and understanding of human language, making it essential for smart healthcare.Human language appears mainly as text and speech, including clinical records and other healthcare text data.
- NLP in healthcare: NLP for smart healthcare processes clinical text and other text data from scenarios such as electronic health records, population screening, and evidence-based reference.Clinical text includes medical notes, diagnostic reports, and electronic prescriptions.
- Review scope: The paper addresses a lack of well-organized systematic synthesis by reviewing NLP for smart healthcare from technical and application perspectives.It also discusses COVID-19, mental health, limitations of existing work, and future directions.
II. NLP FOR SMART HEALTHCARE FROM TECHNICAL PERSPECTIVE
The paper organizes NLP methods into rule-based, statistical, and neural approaches, comparing their accuracy, robustness, interpretability, data needs, and computational demands. It introduces representative algorithms while highlighting trade-offs across the three categories.
- Approach categories: NLP approaches are classified into rule-based, statistical, and neural NLP, each with distinct characteristics and representative algorithms.The paper compares their advantages and disadvantages in Table I.
- Rule-based NLP: Rule-based NLP can be accurate and interpretable in specific cases but requires expert-designed rules that may not cover complex language patterns.Its limited flexibility hinders large-scale applications.
- Statistical NLP: Statistical NLP generally improves performance and robustness over rule-based NLP but depends on handcrafted features and intensive feature engineering.It remains efficient and interpretable because it does not require large datasets or extensive computational power.
- Statistical NLP: Statistical NLP remains active in healthcare despite deep learning’s recent dominance, while n-gram methods trade greater sequential information for exponentially increasing computational complexity with n.TF-IDF improves word-importance measurement but does not model sequential order.
- Neural NLP: Neural NLP often performs better when abundant data are available, but it has lower interpretability, expensive computing requirements, and weaker performance with limited data.Some neural models are also large and expensive to pre-train, constraining broad healthcare application.
B. NLP pipeline for smart healthcare
The smart-healthcare NLP pipeline transforms text or speech into task outputs through preprocessing, feature extraction, and modelling. Preprocessing prepares inputs, feature extraction supplies representations, and task-specific models produce outputs.
- NLP pipeline: The NLP pipeline contains three stages: preprocessing, feature extraction, and modelling.The stages connect text or speech inputs to outputs for specific NLP tasks.
- NLP pipeline: Preprocessing handles input variation and quality to facilitate feature extraction and modelling.Feature extraction is described as the most important step in the pipeline.
- NLP pipeline: Task-specific models are built from extracted features to yield the corresponding outputs.The final modelling stage follows preprocessing and feature extraction.
1) Preprocessing:
Preprocessing normalizes language so it becomes machine-readable and suitable for later NLP stages, while feature extraction increasingly uses automated and pretrained representations. The paper presents these operations as foundations for efficient downstream modelling.
- Preprocessing: Preprocessing operations such as tokenization, stemming, lemmatization, and stopword removal normalize language for machine processing.Normalized input supports postprocessing and helps NLP tasks handle morphological variation in medical terms.
- Feature extraction: Feature extraction has driven NLP development by improving how systems represent information for downstream tasks.Neural methods automate feature engineering through supervised, unsupervised, or reinforcement learning.
- Feature extraction: Pretraining extracts features from large unlabelled corpora through self-supervised learning before downstream tasks begin.The resulting contextual or non-contextual embeddings can encode lexical, syntactic, semantic, and pragmatic features.
- Technical comparison: Table I compares the three NLP approach categories and their representative algorithms.The comparison provides a technical overview of rule-based, statistical, and neural approaches.
3) Modelling:
NLP modelling for smart healthcare spans diverse tasks, from extracting and retrieving information to understanding language and generating responses. These tasks use different models and capabilities depending on the application.
- NLP tasks: NLP models support classification, information extraction, translation, text generation, information retrieval, question answering, and dialogue systems.These tasks process text or speech for outputs ranging from labels and extracted entities to answers and generated text.
- NLP tasks: Information extraction identifies diseases, drugs, events, and related information from medical text through entity, relation, and event extraction.
- NLP tasks: Information retrieval finds documents matching queries across health information systems, digital documents, and online resources.Methods aim to improve retrieval relevance and diversity or reduce query time.
- NLP tasks: Question answering depends on natural language understanding and information retrieval to provide answers to human questions in natural language.Healthcare QA has developed from information-retrieval and knowledge-based systems toward hybrid systems.
- NLP tasks: Knowledge engineering links medical information extraction with knowledge acquisition, representation, and knowledge-based systems such as knowledge graphs and KBQA.
- NLP tasks: Natural language understanding addresses comprehension of unstructured text or speech, but diversity, ambiguity, and dependence make it difficult.Slot filling and intent detection remain mainstream approaches, supporting intelligent agents and medical-query understanding.
III. APPLICATIONS OF NLP FOR SMART HEALTHCARE
NLP-driven smart healthcare applications connect users and intelligent systems through interfaces and backend processing. They span clinical practice, hospital management, personal care, public health, and drug development.
- Application architecture: A typical NLP-driven application has a user interface for receiving inputs and returning results, plus a backend that processes them with NLP models.The backend may also use knowledge bases when the task depends on stored knowledge.
- Application architecture: The user interface supports information exchange through speech and text, with speech recognition and natural language understanding enabling accessible interaction.
- Application categories: Smart healthcare applications using NLP are classified into clinical practice, hospital management, personal care, public health, and drug development.
A. Clinical practice
In clinical practice, NLP supports clinical communication, decision support, resource allocation, data management, and service quality control. These applications use language technologies to collect, interpret, organize, and act on clinical information.
- Clinical communication and data collection: Speech recognition can capture free-text clinical notes, while machine translation can assist communication with patients who speak different languages or have limited health literacy.These tools support clinical data collection and can reduce staff time spent on documentation.
- Clinical decision support: Clinical decision support systems provide physicians with diagnosis and treatment suggestions, including through question-answering systems.
- Medical resource allocation: Patient triage systems prioritize critical cases, helping allocate limited hospital spaces, personnel, and materials more efficiently.
- Data management: Text classification, information extraction, summarization, semantic search, and question answering help manage medical documentation and retrieve healthcare information.
- Service quality control: Sentiment analysis of patient feedback helps hospitals improve service quality and patient experience while reducing personnel demands.
C. Personal care
NLP supports personal care and related healthcare activities by improving access, assistance, education, screening, and drug-development workflows. Its uses extend from personal health services to research and safety monitoring.
- Personal care: Personal health assistants provide access to medical information and healthcare services without requiring visits to healthcare institutions.They may combine medical information access and remote healthcare subsystems.
- Personal care: Voice-controlled systems and communicating robots can assist elderly and disabled individuals with daily activities, social interaction, and companionship.
- Public health: Knowledge engineering supports health knowledge popularization and medical education by establishing medical knowledge bases.These resources aim to improve health literacy and encourage healthy living habits.
- Public health: NLP supports population screening by identifying higher-risk populations and analyzing healthcare questionnaires, including open-ended responses.
- Drug development: In drug discovery and preclinical research, NLP represents biochemical entities, maps relationships, supports virtual screening, and predicts adverse drug reactions.
- Drug development: NLP can support clinical trial design, patient recruitment, trial analytics, adverse drug-event discovery, and drug safety monitoring.
IV. NLP-DRIVEN SMART HEALTHCARE FOR SPECIFIC MEDICAL ISSUES
NLP-driven smart healthcare addresses the COVID-19 pandemic across prevention, diagnosis, and drug development. Reviewed studies describe forecasting, information support, clinical text applications, and drug-discovery uses.
- COVID-19 pandemic: NLP-driven smart healthcare supports COVID-19 prevention through case forecasting and pandemic knowledge popularization.An improved susceptible–infected model incorporated text inputs through an NLP module, while NLP-aided retrieval, literature-based discovery, and question answering addressed information needs.
- COVID-19 pandemic: Clinical NLP can identify COVID-19 patients, assess imaging reports, identify severe-complication risk, and provide testing advice.These applications are described as useful for accelerating diagnosis, mitigating severe effects, and reducing pandemic-response costs.
- COVID-19 pandemic: NLP supports COVID-19 drug development by identifying therapeutic associations, predicting protein interactions, modelling molecular reactions, and informing clinical and regulatory processes.The reviewed applications are described as opportunities to reduce drug-development time and cost.
B. Mental health
NLP-driven smart healthcare is applied to mental-health prediction, screening, and intervention. Reviewed studies cover multiple disorders and risk behaviours, while also describing online therapeutic support and socially oriented digital assistance.
- Mental health: NLP techniques can predict or screen for psychiatric illness, late-life depression, and severe mental illness.Severe mental illness includes schizophrenia, schizoaffective disorder, and bipolar disorder.
- Mental health: NLP can predict risk-taking behaviours such as suicide with good discrimination, supporting early interventions.Analyses may use social-media posts, screening surveys, electronic health records, and narrative-interview speech.
- Mental health: NLP may provide web-based psychoeducation and online counseling that augment therapist-based interventions.Other described uses include counselor training, real-time counseling-quality monitoring, answer suggestions, and socially oriented digital assistants for older adults.
V. LIMITATIONS AND OUTLOOK
The review identifies persistent limitations in NLP-driven smart healthcare and outlines directions for improving data efficiency, integration, interpretability, and practical reliability.
- Limitations: Human-language ambiguity, especially in biomedical texts, can produce inaccurate actions, useless information, and wrong decisions in healthcare systems.The authors associate these misunderstandings with economic loss, wasted time, and potentially more serious consequences.
- Limitations: Neural NLP applications often lack interpretability, limiting confidence in their use for clinical decision support.The review states that clinical decision support can remain auxiliary until interpretability is adequately addressed.
- Limitations: Implementation remains constrained by computing requirements, training costs, system reliability concerns, patient privacy, and medical ethics.These constraints complicate the deployment of large neural networks in smart healthcare.
- Outlook: Combining multiple NLP techniques could improve data-limited applications through augmentation, automated questionnaire generation, translation, and knowledge integration.The review gives examples involving question answering, population screening, multilingual healthcare information, and clinical decision support.
- Outlook: End-to-end applications could directly map inputs to outputs, simplify complex pipelines, reduce intermediate-component bias, and improve healthcare performance.The review highlights population screening and interventions for reducing readmission as example applications.
- Outlook: Few-shot learning, domain knowledge, and multimodal longitudinal data are proposed to address rare diseases, limited high-quality labels, and incomplete healthcare analysis.The review anticipates combining NLP with computer vision and other AI methods across text, images, recordings, sensors, laboratory results, and genetic information.
VI. CONCLUSION
The paper reviews NLP for smart healthcare from both technical and application perspectives, covering approaches, pipelines, tasks, scenarios, and specific medical issues.
- Technical perspective: The review examines NLP approaches and the NLP pipeline for smart healthcare from a technical perspective.It compares approaches and discusses text-oriented and speech-oriented NLP tasks.
- Healthcare applications: It surveys NLP applications across clinical practice, hospital management, personal care, public health, and drug development.These scenarios illustrate the potential of NLP techniques for delivering smart healthcare.
- Specific medical issues: The review further discusses NLP-driven smart healthcare in the COVID-19 pandemic and mental health.
- Outlook: It organizes representative applications with their related NLP techniques and identifies limitations and directions for future work.