Computer-aided diagnosis using embedded ensemble deep learning for multiclass drug-resistant tuberculosis...

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Computational Biology

Background:

• Tuberculosis (TB) poses a significant global health challenge, with drug resistance complicating treatment and increasing mortality.
• Accurate classification of TB strains, including drug-sensitive TB (DS-TB) and various forms of drug-resistant TB (DR-TB) like multidrug-resistant TB (MDR-TB), pre-extensively drug-resistant TB (pre-XDR-TB), and extensively drug-resistant TB (XDR-TB), is crucial for effective patient management.
• Existing diagnostic methods may lack the speed and accuracy required for rapid clinical decision-making.

Purpose of the Study:

• To develop and evaluate the TB-DRD-CXR web application for classifying TB patients into drug resistance subgroups.
• To implement an ensemble deep learning model capable of distinguishing between five TB subtypes: DS-TB, DR-TB, MDR-TB, pre-XDR-TB, and XDR-TB.

Main Methods:

• An ensemble deep learning model was developed, incorporating novel fusion techniques, image segmentation, data augmentation, and learning rate strategies.
• The model's performance was benchmarked against state-of-the-art techniques and standard Convolutional Neural Network (CNN) architectures.
• The TB-DRD-CXR application was developed and usability tested with medical staff.

Main Results:

• The proposed ensemble deep learning model demonstrated superior performance over existing methods, with accuracy increases ranging from 4.0% to 33.9%.
• Significant accuracy improvements were observed compared to standard CNN models (e.g., DenseNet201, NASNetMobile, EfficientNetV2B3), with gains up to 93.4%.
• The TB-DRD-CXR application achieved a high accuracy rate of 96.7%, excellent time-based efficiency (4.16 goals/minute), and overall relative efficiency (100%).

Conclusions:

• The TB-DRD-CXR web application effectively categorizes TB patients based on drug resistance levels using an advanced ensemble deep learning model.
• The system exhibits high accuracy, efficiency, and user satisfaction (SUS score of 96.7%), indicating its potential for clinical adoption.
• The developed application represents a significant advancement in TB diagnostics, offering improved performance over current methodologies.