Projects / Programmes
Machine learning in coronary artery disease stepwise diagnostic process
| Code |
Science |
Field |
Subfield |
| 3.06.00 |
Medical sciences |
Cardiovascular system |
|
| Code |
Science |
Field |
| B145 |
Biomedical sciences |
Nuclear medicine, radiobiology |
| B530 |
Biomedical sciences |
Cardiovascular system |
coronary artery disease, stepwise diagnostic process, artificial intelligence, machine learning, Naive Bayesian's classifier, probability approach
Researchers (6)
| no. |
Code |
Name and surname |
Research area |
Role |
Period |
No. of publicationsNo. of publications |
| 1. |
17679 |
Jožica Arko |
|
Researcher |
2002 - 2004 |
0 |
| 2. |
09790 |
PhD Jurij Fettich |
Cardiovascular system |
Researcher |
2002 - 2004 |
297 |
| 3. |
09791 |
PhD Valentin Fidler |
Cardiovascular system |
Researcher |
2001 - 2002 |
118 |
| 4. |
06769 |
PhD Ciril Grošelj |
Cardiovascular system |
Head |
2002 - 2004 |
85 |
| 5. |
16192 |
MSc Tomaž Milanez |
Oncology |
Researcher |
2002 - 2004 |
70 |
| 6. |
05366 |
Milan Prepadnik |
Cardiovascular system |
Researcher |
2001 - 2002 |
51 |
Organisations (1)
Abstract
Coronary artery disease (CAD) in our time is the biggest cause of early mortality and morbidity. Although the diagnostic modalities and therapy for the disease are well known there is a problem of discordance in disposable health-money and medical needs, so the rationalization of the diagnostic process would be helpful.
Machine learning (ML) - rapidly growing artificial intelligence subfield - has in last decade already proven to be useful tool in many fields of decisions making, also in some fields of medicine. Usually its decision accuracy exceeds the human one.
In our preliminary, already finished research, we proved the applicability of ML in CAD diagnostic process by comparing its diagnostic decision power to standard (human) one. ML method named ''Naive Bayesian's classifier'' was used. A group of 327 patients in diagnostic process for CAD has performed all steps of standard diagnostics for the disease as: history/physical/laboratory examination, exercise ECG, stress perfusion myocardial scintigraphy and coronary angiography. By diagnostic process the disease was confirmed or excluded. Comparing the results of each diagnostic step with the results of angiography we expressed the diagnostic accuracy of each step in standard way and repeated the calculation with ML. The decision power of ML we tested by comparing the results of both methods. Additional we tested the probability approach to CAD diagnostic decision in standard and ML way.
The average diagnostic accuracy of the standard approach was 69 %, by ML it rises to 80 %. The much lower accuracy of probability approach compared with standard one rises by ML significantly.
According to this promising results we plan to prove our results in a bigger group - approximately 1000 patients. In case of similar results we will suggest performance of a similar study in a few independent centers with a final goal of suggesting the introduction of the method in regular practice.
According to our results, 11 % of the correct diagnoses by ML would be put in the lower diagnostic level, what would be an important saving in diagnostic process.
The work with the ML method in any diagnostic center would need only a medium capacity, appropriate programmed personal computer and on-line input of particular diagnostic data for each patient.
