Multi-instance learning based lung nodule system for assessment of CT quality after small-field-of-view...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Machine Learning for Image Analysis

Background:

• Small-field-of-view CT (sFOV-CT) enhances pixel density and reduces partial volume effects in airway structures.
• Multi-instance learning (MIL) is a weakly supervised machine learning approach for automated image quality assessment.
• Evaluating disparities between conventional CT (c-CT) and sFOV-CT is crucial for optimizing lung nodule detection.

Purpose of the Study:

• To compare image quality and nodule detection between c-CT and sFOV-CT.
• To assess the efficacy of a MIL-based lung nodule system versus radiologist evaluations.
• To identify quantitative differences in image features and signal-to-noise ratios.

Main Methods:

• Retrospective analysis of 112 patients' chest CT scans (c-CT and reconstructed sFOV-CT).
• Subjective assessment of nodule features by two radiologists.
• Objective analysis using a MIL-based lung nodule system (c-MIL and sFOV-MIL) and calculation of SNR-lung and CNR-nodule.

Main Results:

• Radiologist evaluation found minimal CT value as a significant difference between c-CT and sFOV-CT.
• The MIL system detected statistically significant differences in most features between c-MIL and sFOV-MIL, surpassing radiologist findings.
• sFOV-CT demonstrated a significantly higher contrast-to-noise ratio of nodules (CNR-nodule) compared to c-CT, with no significant difference in SNR-lung.

Conclusions:

• sFOV-CT exhibits superior image quality, particularly in CNR-nodule, compared to c-CT.
• The MIL-based system is more sensitive than radiologists in detecting subtle image quality differences between CT modalities.
• sFOV-CT holds potential for improved lung nodule characterization and detection in clinical practice.