Deep Learning-Based Slice Thickness Reduction for Computer-Aided Detection of Lung Nodules in Thick-Slice CT.

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Pulmonology

Background:

• Computer-aided detection (CAD) systems struggle with detecting small lung nodules on 5 mm CT scans.
• Missed nodules on standard CT scans can lead to delayed diagnosis.

Purpose of the Study:

• To evaluate a deep learning technique for reducing CT slice thickness from 5 mm to 1 mm.
• To assess the impact of this technique on CAD performance for lung nodule detection.

Main Methods:

• Retrospective analysis of 687 chest CT scans (355 with nodules, 332 without).
• CAD performance evaluated on nodules with radiologist consensus.
• Slice thickness reduction applied using a deep learning method.

Main Results:

• Scan-level AUC improved from 0.867 to 0.902 (p < 0.001).
• Nodule-level sensitivity increased from 0.826 to 0.916 (at 2 FPs/scan).
• Improvements were more significant for smaller nodules; qualitative analysis showed better differentiation of nodule types.

Conclusions:

• Deep learning-based slice thickness reduction significantly enhances CAD for lung nodule detection.
• Refined 1 mm CT scans improve diagnostic accuracy and support clinical adoption.
• This technique offers a promising approach for earlier and more accurate lung cancer diagnosis.