Pre-trained convolutional neural networks as feature extractors for tuberculosis detection.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Pathology

Background:

• Tuberculosis (TB) caused an estimated 1.8 million deaths in 2015, primarily in developing nations.
• Early TB detection is crucial for preventing mortality, but advanced diagnostic methods are often too expensive for widespread use.
• Current chest radiography analysis for TB relies on trained radiologists, limiting scalability and increasing costs.

Purpose of the Study:

• To investigate the application of pre-trained convolutional neural networks (CNNs) for automated tuberculosis detection from frontal thoracic radiographs.
• To propose and evaluate three distinct methods for utilizing CNNs as feature extractors in TB diagnosis.
• To advance research in AI-driven medical image analysis for infectious diseases.

Main Methods:

• Implementation of three distinct proposals using pre-trained CNNs as feature extractors.
• Application of these methods to analyze frontal thoracic radiographs for tuberculosis detection.
• Comparative analysis of the proposed methods against existing literature and current standards.

Main Results:

• The implemented deep learning models achieved competitive results compared to published works in tuberculosis diagnosis.
• Demonstrated the effectiveness of pre-trained CNNs as powerful feature extractors for medical image analysis.
• Showcased the potential for automated systems to improve the accuracy and efficiency of TB detection.

Conclusions:

• Pre-trained convolutional neural networks show significant promise for automated tuberculosis detection in chest radiographs.
• These AI-driven approaches can potentially reduce diagnostic costs and improve early detection rates.
• Further research in this area could lead to more accessible and effective TB screening tools globally.