Abstract
Tuberculosis is an infectious disease with increasing fatalities around the world. The diagnosis of the disease
is a major challenge to its control and management due to the lack of adequate diagnostic tools, contributing
significantly to the prevalence of drug-resistant tuberculosis. Convolutional Neural Network (CNN) models
have recently been developed to detect drug-resistant tuberculosis by analyzing chest radiograph images from
the TB portal, but the classification results are low. This is because CNNs struggle to capture complex global
and overlapping features in medical imaging, such as chest radiographs of drug-resistant tuberculosis. In
contrast, transformers excel in these areas by utilizing self-attention mechanisms that detect inherent subtle and
long-range dependencies across images. In this study, we used a pretrained data-efficient image transformer
(DEiT) model to enhance the diagnosis of drug-resistant tuberculosis and differentiate it from drug-sensitive
tuberculosis. The new model achieved an AUC of 80% in the detection of drug-resistant tuberculosis, an
improvement of 13% in the AUC compared to current CNN models using data from the same source. The
bootstrap significance test shows that the difference in AUCs is statistically significant. The results of the
study can help healthcare providers improve drug-resistant tuberculosis diagnostic accuracy and treatment
outcomes.
is a major challenge to its control and management due to the lack of adequate diagnostic tools, contributing
significantly to the prevalence of drug-resistant tuberculosis. Convolutional Neural Network (CNN) models
have recently been developed to detect drug-resistant tuberculosis by analyzing chest radiograph images from
the TB portal, but the classification results are low. This is because CNNs struggle to capture complex global
and overlapping features in medical imaging, such as chest radiographs of drug-resistant tuberculosis. In
contrast, transformers excel in these areas by utilizing self-attention mechanisms that detect inherent subtle and
long-range dependencies across images. In this study, we used a pretrained data-efficient image transformer
(DEiT) model to enhance the diagnosis of drug-resistant tuberculosis and differentiate it from drug-sensitive
tuberculosis. The new model achieved an AUC of 80% in the detection of drug-resistant tuberculosis, an
improvement of 13% in the AUC compared to current CNN models using data from the same source. The
bootstrap significance test shows that the difference in AUCs is statistically significant. The results of the
study can help healthcare providers improve drug-resistant tuberculosis diagnostic accuracy and treatment
outcomes.
Original language | English |
---|---|
Number of pages | 11 |
Publication status | Accepted/In press - 4 Dec 2024 |
Event | 18th International Conference on Health Informatics - Porto, Portugal Duration: 20 Feb 2025 → 22 Feb 2025 |
Conference
Conference | 18th International Conference on Health Informatics |
---|---|
Country/Territory | Portugal |
City | Porto |
Period | 20/02/25 → 22/02/25 |
Keywords
- Tuberculosis, Drug Resistance, Deep Learning, Vision Transformer, Data-Efficient Image Transformer, Transfer Learning, Chest X-rays