eHealth

Question 1

What is Medical Data?

Answer 1

Data that is important for delivering medical care.

Question 2

How is Medical Data useful?

Answer 2

Used to Categorise problems, to understand what a disease is, to assist the decision of treatment.

Question 3

Name some types of medical data (5)

Answer 3

Narrative, Textual, Numerical, Recorded Signals, Images

Question 4

What is Moore’s Law?

Answer 4

CPU power doubles every 18 months while the price halves

Question 5

What is Metcalfe’s Law?

Answer 5

As a network grows, the value of being connected grows exponentially, while the cost of adding a new user remains the same or decreases

Question 6

What is Gilder’s Law

Answer 6

Bandwidth/Communications power doubles every 6 months

Question 7

What is Zuckerberg’s Law

Answer 7

The amount of information you share on the web doubles every year.

Question 8

With respect to Infrastructure, what is the most expensive part of operating a Data Centre?

Answer 8

Servers - 57%, Power and Cooling - 18%

Question 9

Name some ways to reduce the cost of operating Data Centres.

Answer 9

• Build where energy costs less
• Raise Temperature up to 27C - less cooling costs
• Reduce conversion costs with bespoke parts.
• Cooling towers

Question 10

With respect to End User Devices, how do we reduce power consumption?

Answer 10

• Take measurements at sensible intervals, no 1000 measures per second
• On device processing vs Remote processing (One may be better than another)

Question 11

With respect to heart rate monitoring, what is ECG?

Answer 11

Electrodes on the skin pick up the heart’s electrical signals

Question 12

With respect to heart rate monitoring, what is PPG?

Answer 12

Led Light source shining green light into the skin. Green light is absorbed by blood, so the amount of light reflected back is a data point that can be processed by machine learning algorithms to determine the amount of blood present.

Question 13

Name advantages of Green Light PPG

Answer 13

Good signal to noise ratio, resistant to motion

Question 14

Name disadvantages of Green Light PPG

Answer 14

Skin absorbs light dependant on melanin, creating disparity is use between different patients.

Green light cannot reach deeper tissue to take different data points.

Question 15

Name advantages of Infrared Light PPG

Answer 15

Penetrates Deeper into the skin, can obtain several biometric data points.

Question 16

Name disadvantages of Infrared Light PPG

Answer 16

Very susceptible to motion, but ignores skin color plus tattoos.

Question 17

How does the PWV blood pressure monitor work?

Answer 17

Two PPG sensors at different points on the finger, estimates BP from Pulse Wave Velocity (Difference in phase between the waveforms)
PPG waves are passed to machine learning algorithms to estimate BP from Pulse Wave Velocity.

Question 18

What are some advantages of the Smart Ring

Answer 18

Miniturisable, low power consumption, provides varied biometrics (Pulse, Blood Oxygenation, Pulse Rate Variability)

Question 19

What can the smart ring offer that sensors cannot on their own?

Answer 19

The smart ring can do low level signal processing, data filtering and two way communication with a phone

Question 20

What is a common pitfall of wearable PPG based devices?

Answer 20

They are all susceptible to motion as they are worn - Need to remove motion artifacts to get accurate information from biometric data harvested.

Question 21

With respect to Stress Level Monitoring, what is Heart Rate Variability?

Answer 21

Measure of variation in time intervals between successive heartbeats.

Question 22

Stress Level Monitoring Metrics : What is AVNN?

Answer 22

Average IBI (Interbeat Interval) of normal heart beats

Question 23

Stress Level Monitoring Metrics : What is SDNN?

Answer 23

Standard Deviation of the IBI of normal heartbeats

Question 24

Stress Level Monitoring Metrics : What is RMSSD?

Answer 24

Root mean square of sucessive difference of IBI of normal heart beats

Question 25

Stress Level Monitoring Metrics : What is HRV?

Answer 25

Heart Rate Variability

Question 26

In a Stress State, what metric tends to be lower?

Answer 26

In a stress state, the variation between successive beats tends to be lower.

Question 27

In a relaxed State, what metric tends to be higher?

Answer 27

In a relaxed state, the variation between successive beats tends to be higher.

Question 28

With respect to the knowledge discovery process, name with 4 steps that take us to knowledge.

Answer 28

Data - Preprocessed Data - Patterns - Knowledge

Question 29

With respect to the knowledge discovery process, name the 3 inbetween steps of the process.

Answer 29

Attribute Selection, Data Mining Algorithms, Evaluation - Interpetation.

Question 30

With respect to classification, what are the predictor atttibutes?

Answer 30

All attributes other than the class attribute. They determine whether the class attribute is true or false through classification

Question 31

Describe Ten Fold Validation Process up until Accuracy Calculation.

Answer 31

Divide class members into 10 folds/slices. Take out one fold as a test set. Other Nine are the Training Set. Use the training set to determine the classifier. Use the test set to calculate its predictive accuracy.

Question 32

Describe the formula to calculate accuracy in Ten Fold Validation

Answer 32

Accuracy = Number of Correctly classified examples in the test set Divided By Total number of examples in the test set. Then we repeat the ten fold validation processes. Predicitive accuracy of the whole model = mean of accuracy of all folds.

Question 33

How do we move from Data - Information - Knowledge Process

Answer 33

Data is raw - Information is derived from data by interpreting the data , - Knowledge is a course of action or suggestion inductively deduced through the information

Question 34

What is data mining for eHealth?

Answer 34

Extracting meaningful information and patterns from data

Question 35

Give an Example of a Descriptive Data Mining Task

Answer 35

Clustering - Finding Groups in Data

Question 36

Give an Example of a Prescriptive Data Mining Task

Answer 36

Regression - Using current or past data to predict trends in the future

Question 37

Explain Classification as a Prescriptive Task

Answer 37

Given the Column that is the classifier, inductively determine the relationship between the predictor attributes and what value the classifier will have.

Question 38

Explain loosely what an Aritifical Neural Network is

Answer 38

a machine learning algorithm inspired by the structure and function of the human brain. It consists of interconnected nodes, called neurons, that process and transmit information.

Question 39

Explain loosely what an Support Vector Machine

Answer 39

support vector machines (SVMs, also support vector networks[1]) are supervised max-margin models with associated learning algorithms that analyze data for classification and regression analysis.

Question 40

Explain loosely Decision Tree Induction

Answer 40

Scalability and Decision Tree Induction in Data Mining ... The goal of decision tree induction is to build a model that can accurately predict the outcome of a given event, based on the values of the attributes in the dataset.

Question 41

How do we compute accuracy with a Confusion Matrix

Answer 41

All True Positives + True Negatives Divided By Total Test Population

Question 42

What is sensitivity and how do we compute it?

Answer 42

Sensitivity is how good our model is at Determing positives True Positives / All Positives (True Positives + False Negatives)

Question 43

What is specificity and how do we compute it?

Answer 43

Specificity is how good our model is at Determing negatives True negatives / All Negatives (True Negatvies + False Positives)

Question 44

How do we approach decision making under uncertainty in eHealth?

Answer 44

We solve this problem by using the Bayes Theorem. Bayes Theorem computes the probability of event A when we know event B has taken place.

Question 45

ReWrite Bayes using sensitivity and specificity P(disease is present | positive test)

Answer 45

P(A | B) = True Positive Rate x P(A) Divided By True Positive Rate x P(A) - (1 - True Negative Rate) x P(A^c) (negation of A is the remainder of the percentage chance that A will occur)