Rafael Pinheiro, researcher at ciTechCare in the area of systems engineering, integrated the XXXI Meeting of the Portuguese Association for Classification and Data Analysis (JOCLAD 2024), which took place from 18 to 20 April, at the School of Technology and Management, at the Polytechnic of Leiria, through the presentation of two scientific projects.
‘Diabetes diagnosis support via CBmeter data classification algorithm’ was one of the presented works as an oral communication, in which Maria Guarino and Rui Fonseca-Pinto are researchers as well. CBmeter is a system under development that provides biosignals in parallel with stimuli to the carotid bodies (CB) for the characterisation of metabolic diseases. So, the objective of this work was to present an algorithm for classifying CBmeter biosignals to support the diabetes diagnosis via Support Vector Machine (SVM) technique – a machine learning algorithm. Among the types of SVM, the one that showed the best results was the polynomial with an average accuracy of 82.4%.
The biosignals submitted to the algorithm proposed in this work were heart rate (HR) and respiratory rate (RR) over a period of 84 minutes.
The CBmeter project will move on to a level of new data collection to establish a larger database for better training of the algorithms. This work aimed to reach technology readiness level standards in the future for dissemination in the industry and the market for diagnostic medical devices.
The researcher also presented the poster entitled ‘Bubble Segmentation Algorithm via Watershed technique with OTSO method and mean shift filtering’, in order to introduce an algorithm for bubble segmentation, in which Rui Fonseca-Pinto and Fadi Dohna also investigated.
This work was carried out in collaboration with the Vorarlberg University of Applied Sciences, in Austria – institution of the RUN-EU network, with whom it was possible to use watershed image segmentation associated with other computer vision technicians, normally used in the health area, to develop an algorithm in an industrial context for energy efficiency, pollution reduction and increased equipment lifetime. The proposed algorithm achieved a success rate of 96.69% in identifying closely spaced bubbles.
This work was funded by Portuguese national funds provided by Fundação para a Ciência e Tecnologia, I.P. FCT/UI/05704/2020