Reference no: EM132363393

Project specification - Anaesthetic Data Analysis

Project Outline

In this project, you need design a Depth of Anaesthesia (DoA) index based on the training data set using the machine learning techniques you learnt in the CSC8003 course, and then assess the performance of your index based on the testing data set. During this project, you need submit three following reports on different deadlines to achieve the 50% assessment.

• Proposal report
• Index design report
• Final project report

2. Background knowledge and Data description

What is DoA index?

It is important to assess the DoA accurately since a precise assessment is helpful for avoiding various adverse reactions such as intraoperative awareness with recall (underdosage), prolonged recovery and an increased risk of postoperative complications for a patient (overdosage). Evidence shows that the depth of anaesthesia monitoring using electroencephalograph (EEG) improves patient treatment outcomes. For an accurate DoA assessment, intensive research has been conducted in finding "an ultimate index", and various monitors and DoA algorithms were developed. The main process is presented in Figure 1 (attached). The BIS index (A-2000 BIS monitor; Aspect Medical Systems Inc., Newton, MA) is a single index derived from a set of time domain and frequency domain measures of EEG data. The BIS index is presented as a numerical index ranging from 100 (awake) to 0 (isoelectric EEG /very deep anaesthetic state) and it is an important reference or benchmark for a newly developed DoA index.

Your target in this project is design a new index based on given parameter sets to assess the DoA of patients. The new index should be similar with BIS index (see attached Figure 2). The new index should also range from 100 (awake) to 0 (very deep anaesthetic state).

Data description

At the beginning of the project, 15 cases of patients' data are given as attachments. They include 9 cases training data (Train1 to Train9) and 6 cases testing data (Test1 to Test6). In each case of training data, there are 7 data sets which include one BIS index (BIS) and 6 parameter data sets(x1, x2, x3, x4, x5 and x6). In each case of testing data, there are only 6 parameter data sets(x1, x2, x3, x4, x5 and x6). The BIS data for testing data will be available after you submit your Index design report.

The BIS data is obtained from BIS monitor and the 6 parameter data sets are calculated from the raw EEG data using different feature extraction methods.
• For each case, all the data sets (BIS, x1, x2, x3, x4, x5 and x6) have the same number of data points.
• All the data sets are stored chronologically. For examples, BIS(1) is the BIS value of the first second. x2(4) is the x2 parameter value calculated from the fourth second EEG data. Each parameter value is corresponding to its BIS value in time series.
• All the parameter data sets x1 from different cases are calculated by the same feature extracting methods. So do x2, x3, x4, x5 and x6.

3. Proposal report

In this report, the following contents should be covered:
• Survey about machine learning application on DoA assessment
The survey should only focus on the machine learning application on parameter selections and DoA design parts. Not necessary for feature extraction.
• Analyse the data sets given in this project, what are their features?
• According to the data sets and survey, discuss which machine learning methods you will use in this project and show the reasons.
• Try to use the machine learning methods you selected to preprocess the data sets and briefly analysis which data sets have the stronger relationship with BIS index.
For this part, you are not required to deeply analysis the methods and results you got. Just some preliminary study results can show you have tried to analyze the data sets using machine learning methods.
• According to survey and your experiments, discuss which DoA evaluation methods you will use to compare the new index with BIS index and show your reasons. For example, R square, Pearson coefficient and so on.

4. Index design report

In this report, the following contents should be covered:
• At least two machine learning methods are discussed and used to do the parameter selections and index design. These machine learning method are not necessary to be the same as those you mentioned in proposal report. But you need show the reasons why you select these methods instead of proposed methods. In most cases, you may find better solutions because you learnt more methods than before when you prepared for the index design report.
• The parameter selection methods and results need to be present in your report clearly. You need present your methods and results with key equations, figures or tables. The programming codes and supporting figures or excel data should be presented in the appendix of the report.
• The DoA index design need to be presented in your report clearly and logically. You need present your methods and results with key equations, figures or tables. The programming codes and supporting figures or excel data should be presented in the appendix of the report. The BIS value cannot be any part of your new index. It means the new index is calculated by an equation including parameters (x1, x2, x3, x4, x5 or x6), not BIS value. The new index may be similar with:
New index = 4*x1+5*(x2)^2
If your new index design is just based on a simple liner regression without deep analysis, you cannot obtain a satisfied mark.
• The performance of new index is evaluated by comparing with BIS index based on training data sets. The results should be presented in tables or figures.
• The Pearson coefficient is required to assess your results by marker in this project. Please show your Pearson coefficient results clearly in your report. In addition, you are also encouraged to use other methods to evaluate your results.
• You can use the new index to assess the DoA of testing data sets. If the results are totally different with the training data sets. For examples, the new index value of

testing data set are higher than 100 or lower than 0 in most time. You should consider revising your new index.
• Discuss the problems you met and how you solved these problems in the process of DoA design. The content should be relevant to machine learning technique application, not time management or other management things.

5. Final project report

In this report, the following contents should be covered:
• The performance of new index is evaluated by comparing with BIS index based on testing data sets. The results are presented in tables or figures. The BIS data for testing data will be available on the StudyDesk after the deadline of index design report submission.
• Deeply analysize the performance of your new index comparing with BIS index. Discuss in which aspects your new index performs well and in which aspects it is not.
• Try to find the reasons and revise your DoA index. If it is hard to improve the index, you can do more research online about the machine learning methods you used to find out what other researchers mentioned the limitations of these methods. You need write a discussion based on your findings.
• Write a summary to this project, including the outcome from the previous reports, difficulties you met during the whole project and your solutions, your deep understanding of machine learning techniques.

6. Word limitation

• The length of Part I - Proposal report should be limited to 6 pages (excluding the title page, table of content, appendix and reference list) with not less than 1000 words.

• The length of Part II - Index design report should be limited to 8 pages (excluding the title page, table of content, appendix and reference list) with not less than 1500 words.

• The length of Part III - Final project report should be limited to 7 pages (excluding the title page, table of content, appendix and reference list) with not less than 1200 words.

Attachment:- Machine learning.rar