Embedded AMIS-Deep Learning with Dialog-Based Object Query System for Multi-Class Tuberculosis Drug Response...

Area of Science:

• Medical Diagnostics
• Computational Biology
• Infectious Diseases

Background:

• Drug-resistant tuberculosis (DR-TB) presents significant treatment challenges, requiring longer, more complex protocols.
• Early detection of DR-TB is crucial for reducing treatment duration and costs.
• Accurate classification of DR-TB subtypes (DS-TB, DR-TB, MDR-TB, XDR-TB) is essential for effective management.

Purpose of the Study:

• To develop a fast and effective classification scheme for four TB subtypes.
• To introduce the Drug Response Classification System (DRCS) as a tool for DR-TB subtype classification.
• To integrate the developed model into a dialog-based object query system for clinical support.

Main Methods:

• Ensemble deep learning (EDL) was employed for classification.
• The EDL model incorporated two image preprocessing techniques, four CNN architectures, and three decision fusion methods.
• The classification model was designed for integration into a dialog-based object query system (DBOQS).

Main Results:

• The EDL model demonstrated improved classification performance for DR-TB, with gains of 1.17-43.43% over existing methods.
• Compared to classic deep learning, EDL achieved 31.25% higher accuracy.
• The DRCS achieved 95.8% accuracy and 95.1% user trust, with 99.70% of users willing to continue its use.

Conclusions:

• The developed EDL-based DRCS offers a fast, accurate, and effective method for classifying DR-TB subtypes.
• The system shows high potential as a supportive diagnostic tool for medical professionals.
• Integration into DBOQS facilitates practical application in clinical settings for improved DR-TB diagnosis.