Computer-Aided Detection of Pulmonary Nodules in Computed Tomography Using ClearReadCT

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Pulmonary nodular lesions (PNL) detection in computed tomography (CT) is crucial for early diagnosis.
• Computer-aided detection (CAD) systems aim to improve the accuracy and efficiency of PNL identification.
• Evaluating CAD performance under various imaging parameters is essential for clinical optimization.

Purpose of the Study:

• To assess the accuracy of the ClearReadCT (Riverain Technologies) CAD application for PNL detection.
• To investigate the impact of contrast enhancement, kernel reconstruction, and slice thickness on CAD performance.
• To identify optimal imaging parameters for enhanced CAD accuracy in CT scans.

Main Methods:

• Retrospective analysis of 106 biopsied PNLs from 100 patients undergoing CT scans.
• Comparison of CAD performance between contrast-enhanced (CECT) and non-contrast-enhanced (NECT) scans.
• Evaluation of different slice thicknesses (0.75 mm, 1.5 mm, 3.0 mm) and kernel reconstructions (soft-tissue, lung).

Main Results:

• Contrast enhancement significantly increased CAD sensitivity from 60% (NECT) to 80% (CECT) (P=0.025).
• CAD sensitivity was higher with 3.0 mm slices (84%) compared to 0.75 mm slices (68%), though not statistically significant (P>0.2).
• Lung kernel reconstructions increased the false positive rate without affecting sensitivity (P>0.05).
• ClearReadCT achieved an optimized sensitivity of 84% and PPV of 67% in enhanced scans with thick, soft kernel reconstructions.

Conclusions:

• ClearReadCT demonstrates improved sensitivity with contrast enhancement and thicker slices for PNL detection.
• Non-contrast imaging, thin slices, and lung kernel reconstructions are associated with inferior CAD performance.
• Optimized imaging protocols are crucial for maximizing the effectiveness of CAD in pulmonary nodule detection.