Electronic Theses and Dissertation

Diabetes adalah penyakit metabolik kronis yang ditandai dengan peningkatan kadar glukosa darah, menyebabkan kerusakan serius pada berbagai organ. Penelitian ini menawarkan solusi monitoring glukosa darah secara non-invasif dengan mendeteksi sinyal photoplethysmography (PPG) melalui sensor di permukaan kulit, yang lebih praktis dan nyaman dibandingkan metode invasif tradisional. Sinyal PPG di windowing dengan 2 metode yaitu Based-Time dan Based-Cycle. Metode Based-Time memotong sinyal dengan interval setiap satu detik, sehingga dapat menghasilkan data yang kurang atau lebih dari satu cycle dan terikut data yang kurang baik. Sebaliknya metode Based-Cycle memotong sinyal setiap satu cycle, dan dapat dilakukan seleksi sehingga diperoleh sinyal-sinyal yang informatif saja. Sinyal PPG kemudian diklasifikasi oleh model Convolutional Neural Network (CNN) dengan arsitektur ResNet34. Hasil yang didapat pada penelitian ini yaitu akurasi testing, presisi, sensitivitas, dan F1 score masing-masing sebesar 75,43%; 75,69%; 87,90%; 81,34% dengan menggunakan windowing Based-Time dan 92.06%; 91,18%; 93,94%; 92,54% dengan menggunakan windowing Based-Cycle. Hasil menunjukkan alat ini dapat memberikan pendekatan terhadap diagnosis diabetes.

Diabetes is a chronic metabolic disease characterized by elevated blood glucose levels, causing serious damage to various organs. This study offers a non-invasive blood glucose monitoring solution by detecting photoplethysmography (PPG) signals through sensors on the skin surface, which is more practical and convenient than traditional invasive methods. The PPG signal is windowed by 2 methods, namely Based-Time and Based-Cycle. The Based-Time method cuts the signal at intervals of every one second, so it can produce data that is less or more than one cycle and includes poor data. Conversely, the Based-Cycle method cuts the signal every one cycle, and selection can be made so that only informative signals are obtained. PPG signals are then classified by a Convolutional Neural Network (CNN) model with ResNet34 architecture. The results obtained in this study are testing accuracy, precision, sensitivity, and F1 score of 75.43%; 75.69%; 87.90%; 81.34% respectively using Based-Time windowing and 92.06%; 91,18%; 93,94%; 92,54% using Based-Cycle windowing. The results show that this tool can provide an approach to the diagnosis of diabetes.