Sensor data analysis and development of machine learning models for detection of glaucoma
[ 1 ] Wydział Informatyki i Telekomunikacji, Politechnika Poznańska | [ 2 ] Instytut Informatyki, Wydział Informatyki i Telekomunikacji, Politechnika Poznańska | [ SzD ] doktorant ze Szkoły Doktorskiej | [ P ] pracownik
2023
Rocznik: 2023 | Tom: vol. 86, part C
artykuł naukowy
angielski
- Machine learning
- Glaucoma
- Personalized medicine
- Sensors
- Time series
EN This paper focuses on analysis of data acquired using Triggerfish contact lens sensor and devices for continuous monitoring of cardiovascular system properties. One of the aims of our research is to build machine learning models that can be applied in clinical practice to detect glaucoma independently of currently used imaging techniques. We show that cardiac sensor data derived attributes are complementary to Triggerfish data and improve mean AUC estimates of the predictive models. We propose methods of attribute generation for raw data recorded by the devices. During data preprocessing we consider division of 24-hour monitoring period for time intervals depending on physiological circadian cycle properties. Comprehensive comparison of model predictive performance metrics is included. Our predictive model involving measurements of corneal biomechanical properties and sensor data based attributes provides AUC of 0.87. It can support glaucoma detection without direct intraocular pressure measurement. We also refer to the issues related to personalized medicine and application of machine learning techniques.
18.08.2023
105350-1 - 105350-9
Article Number: 105350
140
4,9