Diagnosing respiratory diseases with machine learning

This study examines the Clinical Decision Support Systems (CDSS) in healthcare, in particular about the prevention, diagnosis and treatment of respiratory diseases, such as Asthma and Chronic Obstructive Pulmonary Disease (COPD). The empirical pulmonology study of a representative sample (n=132), attempts to identify the major factors that contribute to the diagnosis of these diseases. Our machine learning results show that in COPD’s case, SVM outperform other techniques with 96.7% precision, while the most prominent attributes for diagnosis are: smoking, FEV1, age and FVC. In asthma’s case, the best precision, 84.8%, is achieved with the Random Forest classifier, while the most prominent attributes are: MEF2575, smoking, age and wheeze. Finally, the prediction model is integrated in a mobile-web app ready to be used by doctors and patients

Transcript

1. Diagnosing Asthma and Chronic Obstructive Pulmonary Disease with machine learning THESIS PROJECT by DIMITRIS SPATHIS ADVISOR: PANAYIOTIS VLAMOS BIHELAB IONIAN UNIVERSITY
2. Why Asthma chronic disease of the airways symptoms appear after exposure to stimuli 300 million patients diagnosed yearly 250,000 deaths COPD chronic bronchitis and emphysema main factor of the progressive airway obstruction is smoking 330 million patients diagnosed yearly 3 million deaths
3. Workflow Data collection Doctor Visits 132 unique patients who visited a private doctor in Thessaloniki during 2014- 2015. Each patient record describes 22 different features: demographic profile, medical and special lung measurements, habits and associated symptoms. Machine Learning Statistics and Learning Descriptive statistics of the sample. Machine learning with algorithms such as NNs, SVM and DTs. Feature significance analysis. Mobile app Building a mobile app Extracting the machine learning model into a mobile- web app ready to use.
4. Tools used WEKA JAVA SPSS STATISTICS SKLEARN PYTHON BACKBONE.JS JQUERY.MOBILE UNDERSCORE.JS JAVASCRIPT
5. Variables 1/3
6. Variables 2/3
7. Variables 3/3
8. Devices Spirometer
9. Devices Oxymeter
10. Statistics
11. Crosstab visualization: asthma
12. Crosstab visualization: COPD
13. Machine Learning
14. Training Accuracy
15. Feature Selection Infogain vs Correlations InfoGain (Class, Attribute) = H (Class) - H (Class | Attribute)
16. Asthma Decision Tree
17. COPD Decision Tree
18. Extracting the training model into a mobile app
22. Code and thesis available bihelab.di.ionio.gr/?page_id=209