Example 2

Demonstration example of myocardial infarction risk assessment

---

Problem description

The application of methods and algorithms of machine learning and artificial intelligence in medical diagnostics is rapidly developing.

This example demonstrates some simple procedures for estimating the risk of myocardial infarction for a particular group of people.

---

Example Description

This example demonstrates an element of the final part of a multi-stage study on developing a decision support system in medical diagnosis. The critical stages and features of this study are:

1. Working with a limited amount of data (limited sampling). In addition, even the available limited sample size, especially in medicine, often requires careful data checking for correctness and qualitative data preparation for analysis.
2. Selection of significant features. What factors, parameters and variables significantly affect the output indicators? How does the significance of specific attributes change depending on the space of features or other factors (sample size, particular characteristics of specific factors...)?
3. A two-level approach to integral risk assessment. In this case, the expert knowledge of physicians has shown that it is first appropriate and necessary to learn how to correctly estimate an integral risk by modelling several output indicators. In this example, the first level of modelling assesses qualitatively (note - not quantitatively) 10 indicators (e.g. JgA, HBcAg, etc.) on a scale from 1 to 5), where 1 - a white background of the indicator and the absence of a "+" sign, indicating a very insignificant level of the indicator. 2 - a green background of the indicator and the presence of one "+" sign, indicating a small level of the indicator, 3 - yellow background of the indicator and the presence of two "+" signs, indicating an average level of the indicator, 4 - orange background of the indicator and the presence of three "+" signs, indicating a moderate level of the indicator, 5 - red background of the indicator and the presence of four "+" signs, indicating a high level of the indicator.
4. In order to work effectively with limited data and to select significant parameters for modelling each of the ten indicators, the inductive method of self-organisation of models was used, the founder of which is the outstanding scientist A. G. Ivakhnenko. The total sample size selected for analysis was 187, of which 20 as a control sample, and 167 divided into a training sample (117), and a test sample (50). The regularity criterion was used as an external criterion. Based on the joint work of doctors with experts in machine learning, a set of mathematical models was developed to assess indicators correctly.

---

What is shown in the example?

1. A fragment of the system of integral assessment of myocardial infarction risk depending on 13 attributes is demonstrated by preliminary modelling of the level of values of 10 indicators in qualitative form. It is necessary to note that these 13 attributes are selected automatically by the inductive method of self-organisation of models from 37 potential candidate attributes. Moreover, two characteristics among the 13 selected are categorical.
2. Another peculiarity of the demonstration example is that this fragment presents the possibility of evaluation when entering the values of attributes in a particular range, which means that the optimal models are selected according to the ranges of variable values.
3. The example demonstrates the results of the optimal structure models for estimating indicator values.
4. The risk of myocardial infarction is estimated as an integral indicator based on the obtained results of modelling by indicators.

---

Note

1. The example is for demonstration purpose only, and the results obtained in this example are not for practical application.
2. The value of the example is in demonstrating the potential of machine learning methods in medical diagnostics if applied correctly.

To get information about our cases, please register

Sign in

Email

Password

Sign in

Create account? Forgot password?