The Indian journal of tuberculosis · 2025
AI-driven drug resistance profiling in tuberculosis patients: A transfer learning approach.
СтатьяОбзорНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Wakhare Prashant, Sheikh Shagufta Md S, Mahale Pragati, Patil Pritesh, Bhilegaonkar Sanjay, Gulhane Monali
Аннотация
BACKGROUND: Tuberculosis (TB) is still the most common infectious disease that kills people around the world. Drug-resistant types like multi-drug resistant (MDR-TB) and widely drug-resistant TB (XDR-TB) are becoming more common, which makes treatment much harder. Molecular tests and trait drug resistance testing are common ways to diagnose problems, but they take a long time to do or are hard to get to, especially in places with few resources. New developments in artificial intelligence (AI) and machine learning (ML) offer potential new ways to quickly and accurately profile drug resistance by using a variety of data sources.
METHODS: In this study, Pretrained CNNs (ResNet50, DenseNet121, EfficientNet) extracted CXR features, genomic mutations were modeled using transformer-based encoders, and clinical variables were encoded using dense layers. Modality-specific embeddings were fused using an attention-based architecture.
RESULTS: The proposed model achieved 91.2 % accuracy, 89.4 % sensitivity, and 92.7 % precision, outperforming single-modality baselines. Drug-specific AUC values exceeded 90 % for isoniazid, rifampicin, and ethambutol. Ablation studies confirmed substantial performance gains from tri-modal integration.
CONCLUSION: The results of this study show that using transfer learning along with multi-modal data merging can speed up and improve the identification of TB drug resistance. The framework provides an expandable, low-cost, and flexible way to help people make personalised treatment choices, even in places with limited testing tools. In the future, researchers will focus on adding to datasets and making sure they can be used in clinical settings by doing prospective studies.
METHODS: In this study, Pretrained CNNs (ResNet50, DenseNet121, EfficientNet) extracted CXR features, genomic mutations were modeled using transformer-based encoders, and clinical variables were encoded using dense layers. Modality-specific embeddings were fused using an attention-based architecture.
RESULTS: The proposed model achieved 91.2 % accuracy, 89.4 % sensitivity, and 92.7 % precision, outperforming single-modality baselines. Drug-specific AUC values exceeded 90 % for isoniazid, rifampicin, and ethambutol. Ablation studies confirmed substantial performance gains from tri-modal integration.
CONCLUSION: The results of this study show that using transfer learning along with multi-modal data merging can speed up and improve the identification of TB drug resistance. The framework provides an expandable, low-cost, and flexible way to help people make personalised treatment choices, even in places with limited testing tools. In the future, researchers will focus on adding to datasets and making sure they can be used in clinical settings by doing prospective studies.