Electronic Theses and Dissertation

Tuberkulosis (TB) merupakan penyakit infeksi dengan tingkat kasus dan kematian yang masih tinggi pada tahun 2023, sehingga deteksi dini memegang peranan penting dalam menekan laju penyebaran. Napas manusia diketahui mengandung volatile organic compounds (VOCs) yang dapat dijadikan biomarker dalam merefleksikan kondisi kesehatan tertentu, termasuk TB. Electronic nose (eNose) dibanyak penelitian menawarkan pendekatan cepat, murah, dan non-invasif dalam menganalisis VOCs untuk diagnosis penyakit. Penelitian ini berfokus pada menilai potensi eNose dan mengembangkan model machine learning berbasis data napas dalam mendeteksi TB. Prototipe eNose dan model deteksi dirancang dengan mengumpulkan sampel VOCs dari pasien TB dan non-TB, kemudian diolah menggunakan algoritma machine learning untuk mengidentifikasi pola spesifik VOCs. Data napas diperoleh menggunakan eNose yang dirancang dengan enam jenis sensor gas berbeda, mencatat perubahan sinyal saat mendeteksi VOCs yang dihasilkan dari napas. Data mentah dikumpulkan dan diproses menggunakan teknik preprocessing dan feature extraction berbasis statistik sederhana sebelum dilatih dengan beberapa algoritma machine learning. Validasi kinerja dilakukan melalui cross-validation, menghasilkan akurasi puncak sebesar 85% dengan menggunakan algoritma Artificial Neural Network. Hasil ini menunjukkan bahwa karakteristik VOCs dari napas pasien positif TB dan non-TB dapat dibedakan dengan cukup baik oleh sistem yang dikembangkan. Dengan kemampuan non-invasif dan waktu uji yang singkat, pendekatan eNose ini berpotensi menjadi metode pendukung deteksi dini TB serta mempermudah pemantauan pasien di lapangan.

Tuberculosis (TB) is an infectious disease with high case and mortality rates in 2023, making early detection crucial in curbing its spread. Human breath is known to contain volatile organic compounds (VOCs) that can serve as biomarkers reflecting certain health conditions, including TB. The electronic nose (eNose) in many studies offers a fast, inexpensive, and non-invasive approach to analyzing VOCs for disease diagnosis. This research focuses on assessing the potential of eNose and developing a machine learning model based on breath data to detect TB. The eNose prototype and detection model were designed by collecting VOC samples from TB and non-TB patients, then processed using machine learning algorithms to identify specific VOC patterns. Breath data were obtained using an eNose designed with six different gas sensors, recording signal changes when detecting VOCs produced from breath. Raw data were collected and processed using simple statistical-based preprocessing and feature extraction techniques before being trained with several machine learning algorithms. Performance validation was conducted through cross-validation, yielding a peak accuracy of 85% using the Artificial Neural Network algorithm. These results indicate that the VOC characteristics of breath from TB-positive and non-TB patients can be distinguished quite well by the developed system. With its non-invasive capability and short test time, this eNose approach has the potential to become a supporting method for early TB detection and facilitate patient monitoring in the field.