Deteksi Mycobacterium Tuberculosis Pada Citra Dahak Menggunakan RT-DETR

Umam, Khairul Umam (2025) Deteksi Mycobacterium Tuberculosis Pada Citra Dahak Menggunakan RT-DETR. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022222006_Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 April 2027. Download (20MB) | Request a copy

Abstract

Tuberkulosis (TB) adalah penyakit menular dan mematikan yang disebabkan oleh bakteri Mycobacterium Tuberculosis(MT). Menurut World Health Orga nization (WHO), TB merupakan 10 penyakit yang menyebabkan kematian terbesar di dunia. Dengan proporsi bakteri MT yang relatif kecil dengan estimasi 1,7 miliar orang yang terinfeksi akan berkembang selama hidupnya. Pemerikasaan dahak merupakan salah satu metode diagnosis yang banyak digunakan di negara-negara berkembang. Beberapa dataset publik yang tersedia terdiri dari citra dahak yang merupakan hasil dari pemeriksaan dahak dengan variasi background dan cara pengambilan berbeda. Hal ini menjadikan suatu tantangan untuk membandingkan performa suatu model dalam suatu penelitian lebih lanjut untuk deteksi keberadaan lokasi dari Mycrobacterium Tuberculosis atau Bakteri Tahan Asam (BTA) untuk diagnosis cepat bagi ahli medis. Metode deteksi bakteri dari berbagai pendekatan telah di coba seperti pendekatan pemrosesan gambar, pembelajaran mesin dan pembela jaran mendalam. Penelitian ini mengusulkan untuk mendeteksi keberadaan BTA pada citra dahak menggunakan tiga dataset publik menggunakan model berfokus pada one-stage dan efisien seperti RTDETR (Real-Time Detection Transformer). Penelitian ini berhasil mendeteksi BTA dengan baik. Beberapa percobaan dilakukan dan menghasilkan evaluasi didalamnya. Evaluasi pemodelan di dataset 1 model RTDETR-l memiliki performa yang lebih baik dari model RTDETR-x. Sedangkan pada dataset 2 dan dan dataset 3 yang di bagi menjadi K-fold 5 dan K-fold 10 pada masing- masing dataset , memiliki peningkatan kedua model berada pada K-fold 10. Dari keseluruhan dataset yang di coba pada kedua model. Dataset 2 memiliki performa lebih bagus dari dataset 1 dan dataset 3. Penelitian ini juga melakukan percobaan menggunakan Yolov8. Namun model RTDETR hampir keseluruhan percobaan unggul dalam matrix evaluasi pada masing-masing dataset. Secara visualisasi kedua model RTDETR keseluruhan BTA yang ada di ground truth terdeteksi semua. Akan tetapi tedapat objek yang menyerupai BTA yang masih memerlukan analisis dari tenaga ahli medis.
================================================================================================================================
Tuberculosis (TB) is a contagious and potentially fatal disease caused by the bacterium Mycobacterium tuberculosis (MT). According to the World Health Organization (WHO), TB is one of the top 10 causes of death worldwide. With a relatively small proportion of MT bacteria, it is estimated that 1,7 billion people are infected and may develop the disease during their lifetime. Sputum examination is one of the most widely used diagnostic methods, especially in developing countries. Several publicly available datasets consist of sputum images obtained from sputum examinations with varying backgrounds and acquisition techniques. These variations present a challenge when comparing the performance of different models in further studies aimed at detecting the presence and location of Mycobacterium tuberculosis or Acid-Fast Bacilli (AFB) for rapid diagnosis by medical experts. Numerous approaches have been explored for bacterial detection, including image processing, machine learning, and deep learning techniques. This study proposes detecting the presence of AFB in sputum images using three public datasets, employing one-stage and efficient models such as RTDETR (Real-Time Detection Transformer). The study successfully detected AFB with promising results. Several experiments were conducted, accompanied by evaluation metrics. In Dataset 1, the RTDETR-l model demonstrated superior performance compared to the RTDETR-x model. Meanwhile, in Dataset 2 and Dataset 3, which were divided into K-fold 5 and K-fold 10, both models showed performance improvements at K-fold 10. Among all datasets tested, Dataset 2 exhibited better performance compared to Dataset 1 and Dataset 3. Additionally, experiments were conducted using YOLOv8 for comparison. However, the RTDETR model outperformed YOLOv8 in almost all evaluation metrics across the datasets. In terms of visualization, the RTDETR models successfully detected all AFB instances present in the ground truth. Nevertheless, there were objects resembling AFB that still require further analysis by medical experts.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Tuberkulosis, Mycobacterium Tuberculosis, Deteksi, RTDETR =========================================================== Tuberculosis, Mycobacterium Tuberculosis, Detection, RTDETR
Subjects:	R Medicine > R Medicine (General) > R858 Deep Learning
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Khairul Umam
Date Deposited:	22 Jan 2025 06:15
Last Modified:	22 Jan 2025 06:15
URI:	http://repository.its.ac.id/id/eprint/116578

Actions (login required)

View Item