Developing and verifying automatic detection of active pulmonary tuberculosis from multi-slice spiral CT images based...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Diagnosis of tuberculosis (TB) using computed tomography (CT) is challenging in resource-limited areas due to a shortage of experienced radiologists.
• Active tuberculosis (ATB) detection in multi-slice spiral CT images requires advanced diagnostic tools.

Purpose of the Study:

• To develop an automated artificial intelligence (AI) system for the detection and diagnosis of active tuberculosis (ATB) in CT images.
• To enhance diagnostic accuracy and simplify the process of identifying ATB in clinical settings.

Main Methods:

• A dataset of 846 patients' CT images was utilized, with a U-Net deep learning algorithm applied for automatic detection and segmentation of ATB lesions.
• Image processing techniques were employed to refine 2D lesion detection and reconstruct 3D representations of ATB from consecutive annotations.

Main Results:

• The AI tool achieved a high performance on independent test data, with an Area Under the Curve (AUC) of 0.980.
• Excellent diagnostic metrics were reported, including accuracy (0.968), sensitivity (0.964), specificity (0.971), positive predictive value (0.971), and negative predictive value (0.964).

Conclusions:

• A successful AI tool for the automatic detection of ATB in chest CT scans has been developed.
• The AI tool accurately identifies ATB patients and distinguishes them from non-ATB cases, streamlining diagnosis for clinical application.