Skip to content
SDGs Center UDINUS

Evidence record · UDINUS · SDGs

Deep Learning Using DeiT Transformer and Simple ResNet-16 Architecture for Tuberculosis Classification

Rektorat UDINUS 2025 Published 2026-10-30

Evidence record · UDINUS · SDGs

Unit
Rektorat UDINUS
Period
2025
Goals
Record
d5a1f633-6765-4d56-9fc7-a24e809a18e0
Published
2026-10-30

Universitas Dian Nuswantoro conducted an activity to develop a tuberculosis diagnosis method based on X-ray images. The activity involved Aini, Devita Nurul, Shidik, Guruh Fajar, and Purwanto as participants. Tuberculosis is described as a contagious disease that usually attacks the lungs and is caused by the Mycobacterium tuberculosis virus. The activity addressed limitations in traditional diagnostic methods, including sputum microscopy and expensive molecular tests.

The developed method combined a Data-Efficient Image Transformer, known as DeiT, with a Convolutional Neural Network using a simple ResNet-16 architecture. The model was designed to reduce model complexity and to support tuberculosis diagnosis in resource-limited areas. The approach did not include preprocessing stages such as cropping and data augmentation. The activity used the TBX11K dataset, which contains three classes: healthy, non-TB disease, and TB. The dataset was divided into training, validation, and testing sets in a ratio of 80:10:10.

The source reports that the dataset contained 8400 X-ray images. The model achieved 99.81% accuracy, 99.81% sensitivity, and 99.84% specificity. These results were reported as surpassing previous studies through a simpler and more efficient approach. The model was identified as having potential as a computer-aided diagnosis solution for tuberculosis detection, particularly in developing countries. The study recommends further development using more diverse datasets and exploration of transfer learning methods to improve model generalization in real-world conditions. This activity documents UDINUS participation in research addressing tuberculosis diagnosis through an image-based method.

Reference https://www.scopus.com/pages/publications/105031604023?origin=resultslist