In the next 5 years, machine learning will play an increasingly important role in healthcare. As a data science consultancy, SFL Scientific [https://sflscientific.com], has been on the forefront of innovation and we have seen an explosion of applications in the healthcare and pharmaceutical verticals. Whether it's aggregating new results in medical journals using Natural Language Processing, predicting diseases using Time-Series Analysis, or detecting cancer from MRIs using Machine Vision, healthcare is on the verge of a big data revolution.
The purpose of this course will be to introduce you to these topics and more. We start from the basics of machine learning and guide you through how to apply these techniques to real-world healthcare applications. Whilst this course uses healthcare use cases as examples, the techniques are general and apply to a wide range of industries and scientific fields.
What am I going to get from this course?
- Understand the underlying concepts and algorithms utilized in the Healthcare domain
- Be able to apply machine learning to real life Healthcare applications
- Be able to apply machine learning techniques to general applications in industry using the ideas, concepts, and methods discussed
Module 1: Basic Concepts, Algorithms, and Validation Methods
Some information about the background of the instructor and his team
Motivation and Goals
Why you should learn data science, and what the goals are for this course
Prerequisites and Course Overview
What the prerequisites for this course are, and an overview of what the course will cover
Machine Learning Overview
Gives an overview of different types of high-level Machine Learning methods
An introduction to what unsupervised learning is and an overview of the varieties of algorithms that are commonly used.
Introduction to Supervised Learning
An overview of Supervised and Unsupervised learning
Introduction to Semi-supervised Learning
An explanation of the bias-variance trade-off and how you need to think about it when tackling any machine learning problems.
A look at how you can validate your data to determine if you are in the high bias or variance regimes.
Determining whether or not your model is too complex or too simple is a big issue in machine learning. In this brief video, we'll discuss how you can determine where your model is.
Quantity of Data
A look into how the quantity of data is important, and how you can tell if you are data limited.
Recap of all topics considered in Module 1.
Module 2: Clustering and Dimensionality Reduction
Brief recap of Module 1 and introduction to clustering and dimensional reduction techniques.
Linear regression is one of the simplest models to fit on data.
Our first classification algorithm.
Logistic Regression - Validation
Clustering Algorithms: K-Means Clustering & Hierarchical Clustering
Kmeans Clustering - Simple clustering method using k clusters and their centres
Hierarchical Clustering - Common clustering method using a hierarchy structure.
Anomaly Detection & K-Nearest Neighbours
Methods to detect anomalous data
K-Nearest Neighbours - A simple algorithm using k nearest neighbors.
Forward-Backward Selection & Principal Component Analysis
Forward-Backward Selection - A greedy algorithm for dimensional reduction.
Principal Component Analysis - Another useful dimensional reduction technique.
Quiz covering all Module 2 material.
Module 3: Time Series Analysis on EEG Readings
What is Time Series Data?
What is time-series data and how to validate time-series data.
Decomposing time-series into seasonal components and extracting the underlying trend.
The important concept of whether a distribution is stationary and how to test for it.
ACF and PCF
Auto and Partial-Auto Correlation Functions.
Modeling time-series data with ARIMA models.
Forecasting Measles - Case Study
Our first case-study with some real-world EEG data and generating features for supervised learning methods.
Time Series Workflow
A walk-through of how to analysis time-series data.
Time Series Classification
Classifying time-series data using machine learning methods.
Additional more complicated features to improve classification accuracy.
Quiz for all material in Module 3.
Module 4: Machine Vision: Cancer Detection and Deep Learning
What does it mean for a computer to understand data from images?
Convolutional Neural Networks
A state-of-the-art method to extract data from images.
A brief overview of how neural networks work.
Putting it Together: CNNs
Combining the components to form a Convolutional Neural Network
Case Study: Diabetic Retinopathy
Applying a CNN to a real-world case in the medical field and how to validate images.
Exploration, Preprocessing and Data Augmentation
1.How to build and model and a closer look at the data.
2.Cleaning the data is very important!
3.Balancing the classes of your data for the best results.
Quiz for all material in Module 4.
Module 5: Natural Language Processing: Text Classification to Sort Patient Information
Recap & Overview
Natural Language Processing
The different aspects of natural language processing
A common step in many NLP problems is to tokenize the text data.
N-grams & Bigram
Ngram - A very simple model based on Bayes' theorem and word sequence occurrences.
Bigram - Looking at the simple N=2 gram case and building our own bigram model.
Smoothing the data for words that don't occur in the training set. This process allows the modeling of text with words/tokens not in your corpus.
Only the simplest method of regex is covered in IE here. The simplest method to extract information from text is to look for pattern matching.
More intelligent sequencing models also exist that try to model entire sentences as a sequence of word classes. These include Hidden Markov models, Conditional Random Fields etc - these are considered state of the art (and are readily available in off-the-shelf libraries such as NLTK) but difficult to construct.
Bag of Words Representation and Text Classification
Bag of Words Representation: A common representation for text in NLP problems.
Text Classification: Classifying text documents using machine learning.
Also covers preprocessing of data.
Quiz for all material in Module 5.