Role of AI in Health Practice and Research

Kamarul Imran Musa, MD PhD
drki.musa@gmail.com

School of Medical Sciences, Universiti Sains Malaysia

Outline

  1. Part 1: Introduction to AI
  2. Part 2: Predictive AI
  3. Part 3: Generative AI
  4. Part 4: Precision Medicine
  5. Part 5: AI for Medical and Health Researchers
  6. Part 6: Summary

About the Speaker

Kamarul Imran Musa or KIM

  • Professor in Epidemiology & Statistics, School of Medical Sciences, Universiti Sains Malaysia
  • MD (USM), Master of Community Medicine (Epidemiology & Biostatistics), PhD (Statistics & Epidemiology, Lancaster University, UK)
  • Public Health Medicine Specialist & Fellow of the American College of Epidemiology (FACE)
  • Top 2% Most Cited Scientists Worldwide (Stanford/Elsevier)

Expertise & Publications

Core Expertise:

  • Disease & Epidemiological Modelling
  • Statistical Modelling (GLM, mixed models, survival analysis)
  • Statistical Computing (R, Python, STATA)
  • 195 publications with 120,000+ citations (h-index: 57)

Selected Research Projects:

  • Newton-Ungku Omar Fund: Stroke caregiver solutions (RM2M)
  • FRGS: Machine learning for breast cancer detection
  • RUI: CVD risk prediction among healthcare providers
  • Cost of care for Children with Disabilities (RM1.2M)

Visit myanalytics.com.my

PART 1: Introduction to AI

What is Artificial Intelligence?

Artificial Intelligence (AI) is the study of how computers can learn to solve problems using symbolic language and computational methods.

  • Process vast amounts of patient information
  • Identify hidden patterns in large datasets
  • Make predictions about disease progression
  • Recommend optimized treatment strategies

Note

AI complements rather than replaces healthcare professionals

AI, Machine Learning & Deep Learning

Hierarchy of AI:

  • AI: Broad field of computer intelligence
  • Machine Learning (ML): Subset that learns from data
  • Deep Learning (DL): Subset using neural networks

AI enables computers to mimic human intelligence through pattern recognition and learning algorithms.

History of AI in Healthcare: Timeline

timeline
    title AI in Healthcare Timeline
    1950s : Early AI Research
          : Turing Test Proposed
    1970s : Expert Systems
          : MYCIN for diagnosis
    1980s-90s : AI Winter Period
              : Reduced funding
    2000s : Machine Learning Revival
          : Big Data emergence
    2010s : Deep Learning Revolution
          : CNN for medical imaging
    2020s : Generative AI Era
          : LLMs in healthcare

Key Healthcare AI Milestones

Year Milestone
1950 First AI contribution to medicine (shifting tests research)
1975 Early prototype study on computer applications in medicine
2007 DeepQA software marked advancement in AI-driven analysis
2015 Development of Pharmbot software
2017 First FDA-approved cloud-based deep learning application
2021-2024 Accelerated AI deployment during COVID-19 pandemic

The AI Winter & Renaissance

AI Winter (1974-1980, 1987-1993):

  • Overpromised capabilities
  • Limited computing power
  • Insufficient data
  • High costs

What ended it?

  • Big Data Revolution
  • GPU Computing
  • Algorithm Advances (Deep Learning)
  • Cloud Computing

Machine Learning (ML)

Definition: A subset of AI that enables computers to recognize patterns and acquire knowledge from data without direct programming.

Key Applications in Healthcare:

  • Disease classification
  • Patient risk stratification
  • Outcome prediction
  • Treatment response prediction

Categories of Machine Learning

Supervised Learning

  • Models trained on labeled datasets
  • Examples: Classification, regression

Unsupervised Learning

  • Identifies hidden patterns in unlabeled data
  • Examples: Clustering, dimensionality reduction

Reinforcement Learning

  • Learning through trial and error

Machine Learning Examples

Supervised Learning:

  • Cancer vs. non-cancer diagnosis
  • Predicting hospital readmission risk
  • Identifying diabetic retinopathy

Unsupervised Learning:

  • Patient subtyping in diabetes

  • Identifying unusual vital signs

  • Drug interaction discovery

Deep Learning (DL)

Definition: A specialized subset of ML employing multilayered artificial neural networks to represent complex and high-dimensional healthcare data.

Key Characteristics:

  • Multiple hidden layers
  • Captures intricate nonlinear relationships
  • Excels in image and sequence data analysis
  • Automatic feature extraction

PART 2: Predictive AI

What is Predictive AI?

Predictive Analytics (Predictive AI) applies statistical methods from:

  • Machine learning
  • Data mining
  • Forecasting modeling

Purpose: To examine historical data and forecast future events or outcomes

Important

Predictive analytics turns patterns in data into actionable insights for healthcare decisions

Categories of Predictive Models

  1. Linear Models - Linear/Logistic Regression
  2. Nonlinear Models - Polynomial regression, GAMs
  3. Distance-Based Models - K-Nearest Neighbors (KNN)
  4. Tree-Based Models - Decision Trees, Random Forest
  5. Neural Network Models - ANN, CNN, RNN
  6. Support Vector Machine (SVM) - Hyperplane separation

Tree-Based & SVM Models

Random Forest:

  • Heart failure prediction (AUC: 0.84)

  • Cancer mortality prediction

  • Handles high-dimensional data

Support Vector Machine:

  • Brain tumor classification (96.8% accuracy)

  • Breast cancer detection

  • Effective in high-dimensional spaces

Deep Learning Architectures

Convolutional Neural Networks (CNN)

  • Medical imaging (X-rays, CT, MRI)
  • Pathology slide analysis
  • Skin lesion classification

Recurrent Neural Networks (RNN)

  • Electronic health record analysis
  • Time-series vital sign monitoring
  • Heart failure prediction (AUC: 0.777)

Predictive AI Performance Examples

Application Model Performance
Breast Cancer Classification Deep Learning 94% accuracy
Diabetes Detection SVM (Radial) 98% accuracy
Heart Failure Diagnosis RNN AUC 0.777
Brain Tumor Segmentation DCNN High precision

Predictive AI: Real-World Impact

Hospital Adoption (2024):

  • 71% of hospitals use predictive AI integrated with EHR
  • Most common use: Predicting health trajectories for inpatients
  • Fastest growing: Billing simplification and scheduling

Clinical Examples:

  • Diabetic retinopathy screening
  • Sepsis early warning systems
  • Cancer recurrence prediction

PART 3: Generative AI

What is Generative AI?

Definition: AI systems that can generate new content including text, images, and data that emulates genuine information.

Key Technologies:

  • Large Language Models (LLMs)
  • Generative Adversarial Networks (GANs)
  • Variational Autoencoders (VAEs)

Note

Generative AI creates new content; Predictive AI forecasts outcomes from existing data

Predictive AI vs. Generative AI

Aspect Predictive AI Generative AI
Purpose Forecast outcomes Create new content
Output Predictions, classifications Text, images, data
Training Historical data patterns Learning to generate
Example Disease risk score Clinical note draft
Application Risk stratification Documentation assistance

Major Generative AI Providers

OpenAI - ChatGPT/GPT-4

  • Medical coding accuracy: 99% (ICD-10)
  • Patient education and engagement

Anthropic - Claude

  • Safety-focused AI assistant
  • Clinical documentation

Google - Med-PaLM

  • Healthcare-specific LLM
  • Medical question answering

GenAI Application Readiness

Near-Term:

  • Patient education and engagement

  • Information synthesis

Medium-Term:

  • Documentation and coding

  • Precision medicine support

Longer-Term:

  • Virtual health assistants

  • Disease surveillance

  • Medical education

GenAI in Electronic Medical Records

Current Applications:

  1. Clinical Note Documentation
    • 20% less time completing notes
    • 30% less after-hours work
  2. Chart Summarization
    • 45% equivalent, 36% superior to physician summaries
  3. Patient Messaging
    • 72% report reduced cognitive load

Agentic AI in Healthcare

Agentic AI: Autonomous AI systems that can perform tasks, make decisions, and take actions.

Capabilities:

  • Symptom triage and assessment

  • Appointment scheduling

  • Treatment recommendations

  • Drug interaction checking

  • Conduct symptom assessments

GenAI for Clinical Documentation

Ambient AI Scribes:

  • Listen to patient-provider conversations
  • Generate clinical notes automatically
  • Reduce documentation burden

Benefits:

  • More meaningful patient interactions
  • Reduced after-hours clerical work
  • Improved note completeness

GenAI for Medical Coding

ICD-10 Coding Performance:

  • ChatGPT 4.0: 99% accuracy in nephrology cases
  • ChatGPT 3.5: 87-91% accuracy

Benefits:

  • Faster revenue-cycle management
  • Reduced coding errors
  • Streamlined prior authorizations

GenAI in Drug Discovery

Applications:

  • Generating novel molecular structures
  • Simulating patient illness trajectories
  • Predicting drug-target interactions

Impact:

  • Accelerated drug development
  • Reduced research costs
  • Novel therapeutic discoveries

Challenges with Generative AI

  1. Accuracy and Hallucinations - May generate plausible but incorrect information
  2. Bias in Training Data - Can perpetuate healthcare disparities
  3. Privacy Concerns - Patient data protection requirements
  4. Regulatory Uncertainty - Evolving FDA guidelines
  5. Liability Questions - Responsibility for AI-generated advice

PART 4: Precision Medicine

What is Precision Medicine?

Definition: Medical treatments meticulously tailored to match the individual characteristics of each patient.

Core Concept: Moving from “one-size-fits-all” to personalized care based on:

  • Genetic makeup
  • Physical characteristics
  • Metabolic patterns
  • Lifestyle factors
  • Environmental exposures

Categories of Precision Medicine

  1. Precision Genomics - Using genomic information to guide medical decisions
  2. Precision Oncology - Tumor classification and treatment
  3. Precision Cardiology - Heart disease risk prediction
  4. Precision Public Health - Population health management
  5. Precision Pharmacology - Drug metabolism prediction

Precision Genomics & Oncology

Precision Genomics:

  • Identifying genetic mutations in rare diseases

  • Predicting drug metabolism (pharmacogenomics)

  • Hereditary cancer risk assessment

  • DNA language models predicting effects of novel genomic variants

Precision Oncology:

  • Lung cancer: AI model achieved AUC of 0.82 for predicting N2 metastasis

  • Prostate cancer: 95% AUC for predicting aggressive tendencies

Precision Cardiology & Public Health

Precision Cardiology:

  • Heart failure prediction with AUC 0.777

  • Real-time ECG analysis

  • Cardiac imaging analysis

Precision Public Health:

  • Disease outbreak prediction

  • Population health management

  • AI predicted influenza epidemics with 85% accuracy

Examples of Precision Medicine

Disease AI Application Outcome
Diabetes Real-time insulin dosage adjustment Precise glycemic control
Brain Tumors MRI + genetic marker integration 25% increase in tumor control
Rare Diseases Phenotype-based NLP diagnosis More accurate than experts
Colorectal Cancer AI-assisted endoscopy Improved polyp detection

Impact: Up to 30% enhanced patient response rates compared to conventional approaches

Challenges in Precision Medicine

  1. Data Integration - Combining genomic, clinical, and lifestyle data
  2. Interpretation - Understanding AI recommendations
  3. Access and Equity - Ensuring broad availability
  4. Cost - High initial implementation costs
  5. Privacy - Protecting genetic information

PART 5: AI for Medical Researchers

AI Advantages for Health Researchers

Key Benefits:

  1. Accelerated research timelines
  2. Enhanced data analysis capabilities
  3. Novel hypothesis generation
  4. Improved reproducibility
  5. Access to larger datasets
  6. Cost reduction

AI for Generating Data & Images

Synthetic Data Generation:

  • GANs create realistic synthetic patient data

  • Augments limited datasets

  • Preserves privacy while enabling research

Medical Image Generation:

  • Data augmentation for training

  • Missing data imputation

  • Cross-modality translation

  • Resolution enhancement

AI for Efficient Analysis

Acceleration Areas:

Task Traditional With AI
Literature Review Weeks-Months Hours-Days
Data Cleaning Days Hours
Statistical Analysis Hours Minutes
Pattern Recognition Impossible Automated

AI for Manuscript Writing

AI Assistance in:

  1. Literature Search - Finding relevant papers
  2. Writing - Draft generation, grammar improvement
  3. Citation Management - Reference formatting
  4. Visualization - Figure and table creation

AI Tools for Research Writing

Paperpall - Literature management & writing assistance

Trinka AI - Academic writing & grammar checking

LLM in IDEs - Code & analysis assistance

AI for Report Generation

Automated Reporting:

  • Clinical trial results summaries
  • Research progress reports
  • Grant application assistance
  • Conference abstract drafting

Quality Enhancement:

  • Consistency checking
  • Error detection
  • Format compliance

AI in Clinical Trial Design

Applications:

  • Patient recruitment optimization
  • Protocol design
  • Outcome prediction
  • Adverse event monitoring

Benefits:

  • Faster patient enrollment
  • Reduced trial costs
  • Better outcome measures

Responsible Use of AI in Research

Best Practices:

  1. Transparency - Disclose AI tool usage
  2. Validation - Verify AI-generated content
  3. Ethics - Maintain research integrity
  4. Documentation - Record AI methodology

PART 6: Summary

Key Takeaways: AI Fundamentals

  • AI has evolved from 1950s concepts to today’s transformative healthcare tools
  • AI Winter taught us realistic expectations and sustainable development
  • Machine Learning enables pattern recognition without explicit programming
  • Deep Learning excels in complex image and sequence analysis
  • Supervised and Unsupervised Learning serve different analytical needs

Key Takeaways: Predictive AI

  • Predictive Analytics transforms historical data into future insights
  • Multiple model types (linear, tree-based, neural networks) for different needs
  • Hospital adoption at 71% demonstrates real-world value
  • Applications span risk prediction, diagnosis, and resource optimization

Key Takeaways: Generative AI

  • LLMs revolutionizing clinical documentation and patient communication
  • Major providers (OpenAI, Anthropic, Google) advancing healthcare AI
  • Applications growing from near-term (documentation) to long-term (clinical decisions)
  • Challenges remain in accuracy, bias, and regulation

Key Takeaways: Precision Medicine

  • Personalized care based on individual genetic and clinical profiles
  • Categories include genomics, oncology, cardiology, and public health
  • AI enables complex data integration for treatment optimization
  • Results show up to 30% improvement in patient outcomes

Key Takeaways: AI for Researchers

  • Accelerates research from data collection to publication
  • Generates synthetic data while preserving privacy
  • Enhances analysis capabilities beyond human capacity
  • Supports writing and documentation tasks

The Future of AI in Healthcare

Emerging Trends:

  1. Multimodal AI combining images, text, and genomics
  2. Explainable AI for clinical transparency
  3. Federated learning for privacy-preserving research
  4. AI-robotic surgery integration
  5. Population-level disease surveillance

Ethical Considerations

Key Principles:

  • Beneficence - AI should improve patient outcomes
  • Non-maleficence - Avoid algorithmic harm and bias
  • Autonomy - Support, don’t replace, clinical judgment
  • Justice - Ensure equitable access to AI benefits

Challenges and Opportunities

Challenges Opportunities
Data privacy Improved diagnostics
Algorithmic bias Personalized treatment
Regulatory gaps Reduced healthcare costs
Implementation costs Enhanced access to care
Clinician training Research acceleration

Call to Action for Researchers

  1. Embrace AI literacy - Understand capabilities and limitations
  2. Validate AI outputs - Maintain scientific rigor
  3. Consider ethics - Address bias and equity
  4. Collaborate - Bridge AI and clinical expertise
  5. Stay current - Field evolving rapidly

Conclusion

“AI is no longer a distant prospect but an integral component of modern healthcare, transforming diagnostics, drug discovery, precision medicine, and health system operations.”

The future depends on:

  • Responsible integration
  • Evidence-driven implementation
  • Global collaboration
  • Equity-focused deployment

Thank You

Questions?

Contact Information: Email

For further discussion on AI in healthcare

Visit myanalytics.com.my

Additional Resources