MGM-TB-Net: A Mamdani Fuzzy Inference–Gated Multimodal Hybrid Transformer with Residual Fallbacks for Tuberculosis Detection

Tuberculosis (TB) is a lethal bacterial disease, caused by Mycobacterium tuberculosis , and is currently one of the leading causes of death. Although TB is preventable and curable, it can be fatal without early detection and treatment. Computer Aided TB detection has been an emerging field for early and accurate detection of TB. While previous works have employed deep learning (DL) for TB from a single modality, often chest X-rays (CXRs), operate as a black-box system. Furthermore, existing multimodal approaches fail to explore modality reliability in multimodal settings. In real-life clinical scenarios, CXR is used for the primary screening of TB followed by a confirmatory sputum test. The input quality and reliability of the modality may vary. To address these limitations, we propose MGM-TB-Net, a new Mamdani fuzzy inference–gated multimodal hybrid Transformer architecture for reliability-aware TB detection, motivated by the hierarchical nature of clinical TB screening. The proposed framework utilizes dual hybrid CNN–Transformer encoders for modality-specific feature extraction, followed by uncertainty estimation using a differentiable Mamdani Fuzzy Inference System (FIS) to generate confidence scores for each modality. During fusion, the confidence score of the auxiliary modality (sputum) is used to modulate the cross-attention mechanism. Furthermore, a residual fallback is used to preserve the primary CXR representation under degradation of the auxiliary modality. Under controlled experimental settings on a curated multimodal dataset, the proposed MGM-TB-Net architecture shows improved performance compared to baseline models, achieving an accuracy of 99% and AUC-ROC of 0.9976. Furthermore, we have evaluated MGM-TB-Net on two standard unimodal datasets, namely Dataset of Tuberculosis Chest X-rays Images and TB Chest X-ray Database. The experimental results on the unimodal datasets highlight that MGM-TB-Net shows strong performance against widely used DL architectures.

Authors

Institutions

Publication Details

Journal
International Journal of Computational Intelligence Systems
Published
2026-11-21
DOI
https://doi.org/10.1007/s44196-026-01689-y
Primary Topic
COVID-19 diagnosis using AI
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

MGM-TB-Net: A Mamdani Fuzzy Inference–Gated Multimodal Hybrid Transformer with Residual Fallbacks for Tuberculosis Detection

Debgandhar Ghosh, Vyacheslav V. Gulvanskii, Dmitrii Kaplun, Pawan Kumar Singh et al.
International Journal of Computational Intelligence Systems
COVID-19 diagnosis using AI
article

MGM-TB-Net: A Mamdani Fuzzy Inference–Gated Multimodal Hybrid Transformer with Residual Fallbacks for Tuberculosis Detection

Debgandhar Ghosh, Vyacheslav V. Gulvanskii, Dmitrii Kaplun, Pawan Kumar Singh, Asya I. Lyanova
article en

Abstract

Tuberculosis (TB) is a lethal bacterial disease, caused by Mycobacterium tuberculosis , and is currently one of the leading causes of death. Although TB is preventable and curable, it can be fatal without early detection and treatment. Computer Aided TB detection has been an emerging field for early and accurate detection of TB. While previous works have employed deep learning (DL) for TB from a single modality, often chest X-rays (CXRs), operate as a black-box system. Furthermore, existing multimodal approaches fail to explore modality reliability in multimodal settings. In real-life clinical scenarios, CXR is used for the primary screening of TB followed by a confirmatory sputum test. The input quality and reliability of the modality may vary. To address these limitations, we propose MGM-TB-Net, a new Mamdani fuzzy inference–gated multimodal hybrid Transformer architecture for reliability-aware TB detection, motivated by the hierarchical nature of clinical TB screening. The proposed framework utilizes dual hybrid CNN–Transformer encoders for modality-specific feature extraction, followed by uncertainty estimation using a differentiable Mamdani Fuzzy Inference System (FIS) to generate confidence scores for each modality. During fusion, the confidence score of the auxiliary modality (sputum) is used to modulate the cross-attention mechanism. Furthermore, a residual fallback is used to preserve the primary CXR representation under degradation of the auxiliary modality. Under controlled experimental settings on a curated multimodal dataset, the proposed MGM-TB-Net architecture shows improved performance compared to baseline models, achieving an accuracy of 99% and AUC-ROC of 0.9976. Furthermore, we have evaluated MGM-TB-Net on two standard unimodal datasets, namely Dataset of Tuberculosis Chest X-rays Images and TB Chest X-ray Database. The experimental results on the unimodal datasets highlight that MGM-TB-Net shows strong performance against widely used DL architectures.

International Journal of Computational Intelligence Systems
Jadavpur University (IN), China University of Mining and Technology (CN), Saint Petersburg State Electrotechnical University (RU), Innopolis University (RU)
COVID-19 diagnosis using AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon