Klasifikasi Tuberkulosis Menggunakan Focal Transformer Pada Citra X-ray Dada

Rachman, Muhammad Arief Rachman (2025) Klasifikasi Tuberkulosis Menggunakan Focal Transformer Pada Citra X-ray Dada. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5002211164-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (21MB) | Request a copy

Abstract

Tuberkulosis (TB) merupakan salah satu penyakit menular paling mematikan di dunia, dan Indonesia termasuk negara dengan jumlah kasus tertinggi setiap tahunnya. Citra X-ray dada merupakan metode umum dalam diagnosis tuberkulosis.
Namun, kemiripan visual antara paru-paru yang terinfeksi dan yang normal sering kali menyebabkan kesalahan klasifikasi. Untuk mengatasi tantangan tersebut, penelitian ini mengusulkan sistem diagnosis otomatis berbasis kecerdasan buatan untuk melakukan klasifikasi TB dari citra X-ray dada menggunakan model Focal Transformer. Model ini digunakan karena kemampuannya dalam menangkap hubungan spasial lokal dan global melalui mekanisme focal self-attention, yang bekerja pada dua tingkat perhatian: tingkat halus untuk area terdekat dan tingkat kasar untuk area yang lebih jauh. Pendekatan ini membantu model membedakan fitur yang merupakan karakteristik utama TB dari kondisi paru normal yang secara visual dapat tampak serupa, dengan mengenali detail detail lokal pada citra serta memahami konteks spasial secara menyeluruh. Penelitian ini menggunakan tiga varian Focal Transformer, yakni Focal-Tiny, Focal-Small, dan Focal Base, yang dilatih pada dataset Kaggle Combined Unknown Pneumonia and Tuberculosis dengan dua kategori: tuberkulosis dan normal. Pengujian dilakukan pada citra tanpa dan dengan gangguan untuk mengevaluasi ketahanan model. Focal-Base menunjukkan performa terbaik di antara ketiganya dengan akurasi 93,47% pada citra tanpa gangguan;
90,09% pada citra dengan Gaussian Noise; 91,08% pada Salt and Pepper; 92,57% pada Motion Blur; 80,67% pada Brightness and Contrast Distortion; dan 91,20% pada Partial Occlusion.

============================================================================================================================================

Tuberculosis (TB) is one of the most deadly infectious diseases in the world, and Indonesia is among the countries with the highest number of cases each year. Chest X-ray imaging is a common method used for diagnosing tuberculosis. However, the visual similarity between infected and normal lungs often leads to classification errors. To address this challenge, this study proposes an automated diagnosis system based on artificial intelligence to classify TB from chest X-ray images using the Focal Transformer model. This model was chosen for its ability to capture both local and global spatial relationships through a focal self-attention mechanism, which operates on two levels of attention: fine-grained for nearby regions and coarse-grained for distant regions. This approach enables the model to distinguish features that are key characteristics of TB from visually similar normal lung conditions by identifying local image details and understanding the overall spatial context. This study utilizes three variants of the Focal Transformer—Focal-Tiny, Focal-Small, and Focal-Base—which were trained on the Kaggle Combined Unknown Pneumonia and Tuberculosis dataset with two categories: tuberculosis and normal. The models were tested on both clean and distorted images to evaluate their robustness. Among the three, Focal-Base achieved the highest performance, with an accuracy of 93.47% on clean images; 90.09% on images with Gaussian Noise;
91.08% on Salt and Pepper noise; 92.57% on Motion Blur; 80.67% on Brightness and Contrast Distortion; and 91.20% on Partial Occlusion.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Tuberkulosis, Focal Transformer, X-ray dada.
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence
Divisions:	Faculty of Mathematics and Science > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Arief Rachman
Date Deposited:	01 Aug 2025 03:16
Last Modified:	01 Aug 2025 03:16
URI:	http://repository.its.ac.id/id/eprint/125083

Actions (login required)

View Item