Screening Pulse Oximeter pada Uji Fungsi Pra-Kalibrasi Pulse Oximeter Jari dengan Analisis Response Time Menggunakan Machine Learning Model

Septia Khairunnisa, Dr. Indah Soesanti, S.T., M.T. ; Dr. Dyah Listyarifah, M.Sc.,D.Med.Sci. ; Hary Ernanto, A.Md. T.E.

2024 | Tesis | MAGISTER TEKNIK BIOMEDIS

Pengukuran waktu respons pulse oximeter dengan menggunakan simulator SpO2 belum memiliki metode standar dan ambang batas yang tepat. Oleh karena itu, diperlukan metode pengukuran alternatif untuk meminimalkan kesalahan dan meningkatkan efisiensi kalibrasi pulse oximeter, terutama karena semakin banyaknya variasi pulse oximeter yang beredar akibat pandemi COVID-19. Penelitian ini bertujuan untuk mengukur waktu respons (Response Time, RT) dari pulse oximeter jari dengan enam jenis pengkondisian berbeda guna menentukan ambang batas RT untuk pengujian dan kalibrasi alat. Kami mengevaluasi waktu respons dari 104 oksimeter denyut jari (22 jenis patient monitor, 20 jenis handheld, dan 38 jenis fingertip) menggunakan enam metode pengkondisian saturasi dan desaturasi dengan SpO2 simulator. Analisis kuantitatif dilakukan untuk menentukan nilai ambang batas awal guna memudahkan pelabelan dataset. Selanjutnya, hasil dari 104 dataset dianalisis menggunakan algoritma machine learning seperti Logistic Regression, Linear Discriminant Analysis, SVMs, K-Nearest Neighbor (KNN), Decision Trees, Random Forest, dan Neural Networks. Hasil klasifikasi dengan model algoritma Neural Networks menunjukkan nilai akurasi, presisi, recall, dan f1-score yang lebih baik dibandingkan model algoritma lainnya, dengan nilai akurasi sebesar 87%, spesifisitas 91,8%, presisi 77,5%, recall 73,5%, dan f1- score 72,3%. 

A standardized method and proper threshold to measure pulse oximeter response time using SpO2 simulator has not established. Therefore, alternative measurement approaches are necessary to minimize errors and enhance the efficiency of pulse oximeter calibration methods, particularly given the increasing variety of pulse oximeter available as a result of the COVID-19 pandemic. The purpose of this study is to measure the response time (RT) of a finger pulse oximeter with 6 different types of conditioning to determine the RT threshold for device testing and calibration. We evaluated the response time of 104 finger pulse oximeters (22 patient monitor types, 20 handheld types, and 38 fingertip types) using 6 saturation and desaturation conditioning methods with the SpO2 Simulator. Quantitative analysis was used to determine baseline thresholds to facilitate labeling of the data set. In addition, results from a total of 104 datasets were analyzed using machine learning algorithms including; Logistic Regression, Linear Discriminant Analysis, SVMs, K-Nearest Neighbor (KNN). Decision Trees, Random Forest, and Neural Networks. The results of classification using machine learning with the Neural Networks algorithm model produce better accuracy, precision, recall and f1-score values than other algorithm models, with an accuracy value of 87%, specificity 91.8%, precision 77.5%, recall 73.5% and f1-score 72.3%. 

Kata Kunci : Pulse Oximeter, Kalibrasi, Response Time, Analisis Kuantitatif, Machine Learning