You are on page 1of 47

Machine Learning in

Medical and Healthcare

Risman Adnan
Outline

• Machine Learning Problem


• ML Model and Techniques
• Machine Learning in Healthcare
• Our First Predictor
• Tools and Frameworks
Automatic discovery of regularities in data through
computer algorithms and the use of those regularities to
take actions such as classifying the data into categories.

A pattern exists

The Essence of Machine We cannot pin it down


mathematically (no analytical
Learning: solution)

The Learning We have data on it

Problem Initial: Problems that hard for human – but easy for
computer.

Current: Problems that hard for computer– but easy for


human.
Machine Learning Framework
 We have data UNKNOWN TARGET FUNCTION
PROBABILITY
𝑓: 𝛸  𝛶
 𝑥 can be anything DISTRIBUTION
 𝑦 is label of data P on X

 There is Pattern TRAINING DATASET


𝑥1 , 𝑦1 , 𝑥2 , 𝑦2 , … … . (𝑥𝑁 , 𝑦𝑁 )
𝑥1 , 𝑥2 , … 𝑥𝑁
ERROR
 Unknown target function MEASURE
 No analytic solution e(…)
𝑔 𝐱 ≈ 𝑓(𝐱)
 Make hypothesis LEARNING FINAL HYPOTHESIS
ALGORITHM
 Find best hypothesis 𝜜
𝑔: 𝛸  𝛶

 Use Algorithm
 Model = Hypothesis +
Algorithm HYPOTHESIS SET

Health Insurance Sample
Approve

Metaphor: Health Insurance Approval (Classification) Insurance


New
Cust Approval System

• Input: 𝒙 (customer application) Not Approve

• Output: 𝑦 (good/bad customer)


• Unknown Target Function f
• Data (historical data)
• Hypothesis Set & Final Hypothesis

Learning Model = Hypothesis Set + Learning Algorithm


Outline

• Machine Learning Problem


• ML Model and Techniques
• Machine Learning in Healthcare
• Our First Predictor
• Tools and Frameworks
The Map of Machine • Machine learning become a jungle of models, methods!
Learning

TECHNIQUES
THEORY PARADIGMS
MODELS METHODS
VC Linear Regularization Supervised
Bias-Variance Neural Networks Validation Unsupervised

Complexity SVM Aggregation Reinforcement


Bayesian Nearest Neighbors Input Processing Active
RBF Online
Gaussian Processes

SVD
Types of Machine Learning
 Supervised Learning: Learning by labeled example
 E.g. Breast Cancer Detection
 We have (input, correct output), and we can predict (new input, predicted output)
 Amazingly effective if you have lots of data
 Unsupervised Learning: Discovering Patterns
 E.g. Data Clustering for Covid Patients
 Instead of (input, correct output), we get (input, ?)
 Difficult in practices but useful if we lack labeled data
 Reinforcement Learning: Feedback & Error
 E.g. Automated Medical Diagnostic
 Instead of (input, correct output), we get (input, only some output, grade of this output)
 Works well in some domains, becoming more important

In this course we will focus on Supervised Learning paradigm


Machine vs Deep Learning

Classical Machine Learning


 Unknown Function Approximator
 Various Learning model
 Hand-Crafted Features
 Cheap Computation and “Less Data”

Modern Deep Learning


 NN Based Unknown Function Approximator
 Deep Neural Nets: DNN, CNN, & RNN
 Automatic Features Extraction
 Heavy Computation and “Huge Data”
Outline

• Machine Learning Problem


• ML Model and Techniques
• Machine Learning in Healthcare
• Our First Predictor
• Tools and Frameworks
1970’s:
MYCIN
Expert
System
1980’s:
INTERNIST-
1/QMR
model
1980’s:
Automating
Medical
Discovery
1990’s:
Neural
Networks in
Medicine
Neural
Network in
Medical
Decision
Making
Machine Learning • Secure and Robust Machine Learning for Healthcare: A Survey

Based Healthcare
Typology in Medical Imaging
Data Is Every Where
Diversity on Data
Data Standardization
Breakthrough
AI
Achievement
Industry Supports
Startups
Ecosytem
Outline

• Machine Learning Problem


• ML Model and Techniques
• Machine Learning in Healthcare
• Our First Predictor
• Tools and Frameworks
Google AutoML
Main Capabilities on GCP (Google Cloud Platform)
AutoML is automatic process of model algorithm selection

Build your own model AutoML Call AutoML Trained model as APIs
your data + your model your data + autoML model AutoML data + autoML model

Cloud AutoML
Cloud ML Engine
Google AutoML
How it works

Your Dataset

Cloud AutoML

Sneakers Cap Handbag Generate prediction


with a REST API

Training Deploy Serve


Google AutoML
Cloud Vision API

OCR Crop Hint Face Detection

Multiple Object
Detection

Landmark Logo Detection


Detection
Google AutoML
Cloud Natural Language

Extract Detect Analyze Classify


Entities Sentiment Syntax Content
Microsoft Innereye Project
• InnerEye is a research project from Microsoft Research
Cambridge that uses state of the art machine learning
technology
• Build innovative tools for the automatic, quantitative
analysis of three-dimensional medical images
• The goal of Project InnerEye is to democratize AI for
medical image analysis
• Empower
• researchers,
• hospitals,
• life science organizations, and
• healthcare.
Microsoft Innereye Project

Open source code


https://github.com/microsoft/InnerEye-DeepLearning
Outline

• Machine Learning Problem


• ML Model and Techniques
• Machine Learning in Healthcare
• Our First Predictor
• Tools and Frameworks
Overview of Orange Software

• Open source
• Component based
• Data visualization
• Analysis for novice and expert
• Data mining through visual programming or python scripting
• Add-ons for bioinformatics and text mining
• Packed with features for data analytics
Overview of Orange Software

Widgets
• Data entry and processing
• Data visualization
• Classification
• Data Evaluation
• Unsupervised Learning
Scripting in Orange Software

Scripting Widgets
Overview of Orange Software

Examples
• Story Arcs
• Twitter Data Analysis
• Text Classification
• Text Clustering
• Text Processing
• Cross Validation
• Etc.
Overview of Azure ML

• Machine Learning on Azure


• Custom AI
• Compute Targets (DSVMs and Managed Compute)
• DevOps for Machine Learning
• Azure Machine Learning Pipelines
• Flexible and Support for Open Source Frameworks
• Deployment
• Tool Agnostic Python SDK
Building Block for Data Science Project
Machine Learning on Azure

Domain Specific Pretrained Models
To reduce time to market Vision Speech Language Search

Familiar Data Science Tools


To simplify model development PyCharm Jupyter Visual Studio Code Command line

Popular Frameworks
To build machine learning and deep learning solutions PyTorch TensorFlow Scikit-Learn ONNX

Productive Services
To empower data science and development teams Azure
Databricks
Azure Machine
Learning
Machine
Learning VMs

Powerful Hardware
To accelerate deep learning CPU GPU FPGA

From the Intelligent Cloud to the Intelligent Edge


Model Management in Azure ML
Azure Machine Learning Pipelines
Prepare Data Build & Train Model Deploy & Predict

Data Data Preparation Model Building & Training Model Deployment


Ingestion
Normalization Hyper-parameter tuning Deployment

Transformation Automatic model Batch Scoring


Selection
Validation
Model Testing
Data Storage Featurization
Locations
Model Validation
Popular Frameworks
Model Types in Azure ML

Regression Classification Anomaly Detection

Clustering Recommendation
System
Azure Machine Learning Tools

Azure Machine Learning is a ML Studio you can build


cloud predictive analytics service Artificial Intelligence (AI)
that makes it possible to quickly applications that intelligently
create and deploy predictive sense, process, and act on
models as analytics solutions.
information - augmenting
human capabilities, increasing
speed and efficiency, and
helping organizations achieve
more.
Azure Machine Learning Studio

• Collaborative, drag-and-drop tool you can use to build, test, and deploy predictive
analytics solutions on your data.
• ML Studio publishes models as web services that can easily be consumed by
custom apps or BI tools such as Excel.
Azure Machine Learning Studio

• A fully-managed cloud service that enables you to easily build, deploy, and share
predictive analytics solutions.
Azure Machine Learning Studio
5 steps to create an experiment
• Before you start
Open Machine Learning Studio: https://studio.azureml.net and Sign In.
• If you’ve signed into Machine Learning Studio before, click Sign In.
• Otherwise, click Sign up here and choose between free and paid options.
• Create model
• Step 1 Get data
• Step 2 Prepare the data
• Step 3 Define features
• Train the model
• Step 4 Choose & apply a learning algorithm
• Score and test the model
• Step 5 Predict new incoming data
Q&A

You might also like