A brief introduction to Machine Learning

Increasing computational power over the past decade has enabled the rapid development of a variety of new machine learning methods that primarily aim to analyse medical data, in order to offer extended insights about the inner workings of humans. Evidently, understanding the human body better can potentially aid the earlier diagnosis and treatment of certain diseases that have been troubling humans for many years now.

An inherent problem of medical data is their increased complexity and high dimensionality, which is essentially caused by the complicated design of the human body but is also a result of noise coming from the acquisition hardware. In this talk, we will demonstrate some recent machine learning methodologies for removing noise and extracting inference from medical data using linear and non-linear dimensionality reduction techniques along with a set of unsupervised and supervised classification algorithms. Moreover, we show how these techniques can be best applied in order to aid the diagnosis of different types of cancer and also help in understanding how the human brain works when we are performing various actions in our everyday life.

Transcript

1. A brief introduction to Machine Learning A summary of basic feature selection and classification methods on medical data Zena Hira1 - Loizos Makrides1 1Imperial College London, Department of Computing January 30, 2014 Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 1 / 25
2. Outline 1 Dimensionality Reduction Overview Methods Linear Discriminant Analysis Principal Components Analysis Independent Components Analysis Manifold Learning 2 Classification Overview Methods Support Vector Machines k - Nearest Neighbours Decision Trees Neural Networks 3 Applications Medical Applications Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 2 / 25
3. Outline 1 Dimensionality Reduction Overview Methods Linear Discriminant Analysis Principal Components Analysis Independent Components Analysis Manifold Learning 2 Classification Overview Methods Support Vector Machines k - Nearest Neighbours Decision Trees Neural Networks 3 Applications Medical Applications Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 3 / 25
4. Dimensionality Reduction Overview Dimensionality reduction is the process of reducing the number of random variables under consideration and can be divided into two steps: Feature extraction transforms the data in the high-dimensional space to a space of fewer dimensions. Feature selection process of selecting a subset of relevant features for use in model construction. Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 4 / 25
5. Dimensionality Reduction Why is it needed? The curse of dimensionality (Small number of samples - Large number of features) Cancer data involve tens of thousands of gene expressions and thousands of different patients in which they are expressed. Functional MRI data model the brain with almost 400.000 volumetric pixels for each time point. Real-life data contain a lot of noise Noise coming from the hardware that is used to capture the measurements Noise coming from other effects that are happening alongside the effect we are studying Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 5 / 25
6. Outline 1 Dimensionality Reduction Overview Methods Linear Discriminant Analysis Principal Components Analysis Independent Components Analysis Manifold Learning 2 Classification Overview Methods Support Vector Machines k - Nearest Neighbours Decision Trees Neural Networks 3 Applications Medical Applications Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 6 / 25
7. Linear Discriminant Analysis Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 7 / 25
8. Principal Components Analysis PCA transforms a number of possibly correlated variables into a (smaller) number of uncorrelated variables called principal components. Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 8 / 25
9. Independent Components Analysis ICA transforms a multivariate signal into a number of additive subcomponents which are statistically independent. The assumption that the input signals are not normally distributed must hold. Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 9 / 25
10. Manifold Learning / Isomap ISOMAP High Dimensional Space mapped to a lower one by perceiving topological characteristics. Works by determining the geodesic distances of the neighbourhood graph and creating the multidimensional scaling. Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 10 / 25
11. Outline 1 Dimensionality Reduction Overview Methods Linear Discriminant Analysis Principal Components Analysis Independent Components Analysis Manifold Learning 2 Classification Overview Methods Support Vector Machines k - Nearest Neighbours Decision Trees Neural Networks 3 Applications Medical Applications Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 11 / 25
12. Classification Classification is the problem of identifying to which of a set of categories (sub-populations) a new observation belongs, on the basis of a training set of data containing observations (or instances) whose category membership is known. Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 12 / 25
13. Outline 1 Dimensionality Reduction Overview Methods Linear Discriminant Analysis Principal Components Analysis Independent Components Analysis Manifold Learning 2 Classification Overview Methods Support Vector Machines k - Nearest Neighbours Decision Trees Neural Networks 3 Applications Medical Applications Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 13 / 25
14. Support Vector Machines SVM represents all the observations as points in space, mapped so that observations of separate categories are divided by a clear gap that is as wide as possible. New observations are then mapped into that same space and predicted to belong to a category based on which side of the gap they fall on. Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 14 / 25
15. Support Vector Machines In addition to performing linear classification, SVMs can efficiently perform a non-linear classification using what is called the kernel trick, implicitly mapping their inputs into high-dimensional feature spaces. Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 15 / 25
16. k - Nearest Neighbours In k-NN classification an object is classified by a majority vote of its neighbours, with the object being assigned to the class most common among its k nearest neighbours. Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 16 / 25
17. Decision Trees Creates a model that predicts the value of a target variable Learns simple decision rules inferred from the data features. Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 17 / 25
18. Neural Networks Inspired by the way biological nervous systems Starting with random weights on each arc tries to find the solution that minimizes the overall solution error by adjusting them in each cycle Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 18 / 25
19. Outline 1 Dimensionality Reduction Overview Methods Linear Discriminant Analysis Principal Components Analysis Independent Components Analysis Manifold Learning 2 Classification Overview Methods Support Vector Machines k - Nearest Neighbours Decision Trees Neural Networks 3 Applications Medical Applications Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 19 / 25
20. Microarrays A Quick Description Consist of a small membrane or glass slide containing samples of many genes arranged in a regular pattern. Used for Gene Discovery, Disease Diagnosis, Drug Discovery, Toxicological Research Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 20 / 25
21. Microarrays Illustrated Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 21 / 25
22. Leukaemia Diagnosis Thanks To George Trigeorgis Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 22 / 25
23. Functional MRI Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 23 / 25
24. Brain Decoding Predicting what task the subject was performing in the MRI scanner Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 24 / 25
25. Zena Hira - Loizos Makrides (Imperial College London) A brief introduction to Machine Learning January 30, 2014 25 / 25