A Spectroscopy-Based Sensor for the AI-Based Classification of Lipemic and Hematic Parameters

Dyslipidemias are chronic conditions characterized by elevated levels of LDL cholesterol and/or triglycerides in the bloodstream. Guidelines from the European Society of Cardiology recommended lipoprotein apheresis (LA) as the last-resort therapy for severe cases of familial hypercholesterolemia. However, LA devices equipped with a blood leak detector (BLD) sensor often generate false alarms due to turbidity caused by serum lipoproteins. This letter presents a protocol for creating hyperlipemic plasma samples and proposes the use of a microspectrometer-based BLD sensor to collect the spectra. Furthermore, to address the false alarm problem (blood rate ≥0.7%), we developed a first approach using thresholding and linear support vector machine (SVM). In addition, to improve data classification performance, three artificial neural networks (ANNs) were developed with different classification tasks, then applied to four different databases. ANNs aim to solve the problem of false alarms and accurately classify samples based on blood and phospholipids concentration, achieving 99% accuracy for data classification based on the combination of phospholipids and blood percentage.