Comparing Machine Learning Algorithms for Medical Time-Series Data

Alex Helmersson; Faton Hoti; Sebastian Levander; Aliasgar Shereef; Emil Svensson; Ali El-Merhi; Richard Vithal; Jaquette Liljencrantz; Linda Block; Helena Odenstedt Hergès; Miroslaw Staron

doi:10.1007/978-3-031-49266-2_14

Comparing Machine Learning Algorithms for Medical Time-Series Data
Paper in proceeding, 2024

Medical software becomes increasingly advanced and more mission-critical. Machine learning is one of the methods which is used in medical software to tackle a diversity of patient data, problems with data quality and providing the ability to process increasingly large amounts of data from medical procedures. However, one of the challenges is the lack of comparisons of algorithms in-situ, during medical procedures. This paper explores the potential of performing real-time comparisons of algorithms for early stroke detection during carotid endarterectomy. SimSAX, DTW (dynamic time warping), and Pearson correlation were compared based on the real-time data against medical specialists in clinical evaluations. The analysis confirmed the general feasibility of the app- roach, though the algorithms were inadequate in extracting significant information from specific signals. Interviews with physicians revealed a positive outlook toward the system's potential, advocating for further investigation. Despite their limitations, the algorithms and the prototype application provides a promising foundation for future development of new methods for detecting stroke.

Machine learning

SimSAX

stroke

dynamic time warping

Author

Alex Helmersson

Student at Chalmers

Faton Hoti

Student at Chalmers

Sebastian Levander

Student at Chalmers

Aliasgar Shereef

Student at Chalmers