Deteksi Dan Perhitungan Mycobacterium Tuberculosis Dengan Metode Deep Learning

Fadhila, Farah Dhia (2025) Deteksi Dan Perhitungan Mycobacterium Tuberculosis Dengan Metode Deep Learning. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025211030-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Tuberkulosis merupakan penyakit menular yang disebabkan oleh bakteri Mycobacterium tuberculosis yang biasanya menyerang paru-paru. Berdasarkan Global Tuberculosis Report 2023 oleh World Health Organization, pada 2022, penyakit Tuberkulosis menjadi penyakit menular mematikan kedua setelah COVID-19 dan penyebab kematian hampir dua kali lebih banyak dari HIV/AIDS. Di Indonesia, jumlah pasien yang terdiagnosis Tuberkulosis semakin meningkat setiap tahun dan menyentuh angka 1,06 juta kasus pada 2023. Umumnya diagnosis Tuberkulosis dilakukan secara manual dengan analisis pada pemeriksaan Sputum Smear Microscopy. Namun seiring dengan banyaknya kasus penyakit ini, metode manual tersebut menjadi kurang efisien dikarenakan membutuhkan waktu yang lama dan bersifat subjektif tergantung kemampuan dari para pakar yang menganalisis sampel apus dahak yang diuji. Oleh karena itu, perlu adanya metode automasi yang dapat membantu para pakar dalam menganalisis sampel apus dahak tersebut secara lebih efisien. Penelitian ini dibangun demi mengatasi permasalahan efisiensi waktu saat menganalisis sampel apus dahak yang sedang diuji yakni dengan model automasi untuk mendeteksi objek bakteri dan menghitung total bakteri yang terdeteksi oleh model. Model deteksi objek ini dibangun menggunakan metode deep learning dengan pendekatan transfer learning yang memanfaatkan model pretrained YOLOv9 untuk dilakukan pelatihan ulang agar dapat mempelajari dataset medis yang tersedia. Tahapan penelitian ini mencakup pengumpulan dataset publik dan dataset privat, uji coba dua dataset yang berbeda, uji coba seluruh seri YOLOv9, dan uji coba tuning parameter model pretrained. Tahap terakhir pada penelitian ini adalah evaluasi performa deteksi dan perhitungan objek hasil uji coba. Hasil penelitian ini menunjukkan bahwa model dengan evaluasi deteksi dan perhitungan terbaik, serta waktu pelatihan tercepat ada pada model YOLOv9s parameter tuning dengan precision sebesar 0,794, recall sebesar 0,763, mAP sebesar 0,826, precision counting sebesar 0,904, dan waktu pelatihan selama 358 detik. Sedangkan, waktu inferensi paling cepat ada pada YOLOv9t parameter tuning dengan waktu 1,6 mili detik, serta penggunaan memori yang paling kecil ada pada YOLOv9t dengan total 4,516 MB. Dari hasil penelitian ini diharapkan dapat membantu dalam hal deteksi dan perhitungan bakteri Mycobacterium tuberculosis dalam sampel apus dahak serta dapat membantu pakar untuk menganalisis sampel dengan efektif dan efisien.
=======================================================================================================================================
Tuberculosis is an infectious disease caused by the Mycobacterium tuberculosis bacteria, which primarily affects the lungs. According to the Global Tuberculosis Report 2023 by the World Health Organization, tuberculosis was the second deadliest infectious disease in 2022, following COVID-19, and caused nearly twice as many deaths as HIV/AIDS. In Indonesia, the number of diagnosed tuberculosis patients continues to rise each year, reaching 1,06 million cases in 2023. Typically, tuberculosis diagnosis is performed manually through analysis using Sputum Smear Microscopy. However, given the increasing number of cases, this manual method has become less efficient due to its time-consuming nature and its subjectivity, which depends on the expertise of the specialists analyzing the sputum samples. Therefore, an automated method is needed to assist medical professionals in analyzing these samples more efficiently. This research was conducted to address the issue of time efficiency in analyzing sputum smear samples by developing an automated model for detecting bacteria objects and counting the total number of detected bacteria. The object detection model was built using deep learning
techniques with a transfer learning approach, leveraging the pretrained YOLOv9 model, which was retrained to learn from the available medical dataset. The stages of the research included collecting both public and private datasets, evaluating two different datasets, testing the full series of YOLOv9 models, and tuning the parameters of the pretrained model. The final stage of the research involved evaluating the performance of the model in terms of object detection and counting. The results of this research indicate that the model yielding the best detection and counting performance, along with the fastest training time, was the YOLOv9s with parameter tuning. It achieved a detection precision of 0,794, recall of 0,763, mean average precision (mAP) of 0,826, precision counting of 0,904, and a training time of 358 seconds. Meanwhile, the fastest inference time was recorded by the YOLOv9t with parameter tuning at 1,6 milliseconds, and the lowest memory usage was achieved by the YOLOv9t at 4,516 MB. It is expected that the outcomes of this research can support the detection and counting of Mycobacterium tuberculosis bacteria in sputum smear samples and assist experts in conducting
analyses more effectively and efficiently.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Tuberkulosis, Deteksi Objek, Deep learning, YOLOv9, Tuberculosis, Object Detection, Deep learning, YOLOv9
Subjects:	T Technology > T Technology (General) > T11 Technical writing. Scientific Writing T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T57.62 Simulation
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Farah Dhia Fadhila
Date Deposited:	14 Jul 2025 02:27
Last Modified:	14 Jul 2025 02:27
URI:	http://repository.its.ac.id/id/eprint/119581

Actions (login required)

View Item