Computer-aided detection of solid lung nodules in lossy compressed multidetector computed tomography chest exams

Area of Science:

• Radiology and Medical Imaging
• Image Processing and Compression
• Artificial Intelligence in Healthcare

Background:

• Multidetector computed tomography (MDCT) chest scans are crucial for lung nodule detection.
• Lossy image compression techniques are explored to reduce data storage and transmission burdens.
• The impact of such compression on computer-aided detection (CAD) systems requires careful evaluation.

Purpose of the Study:

• To evaluate the effect of three-dimensional (3D) lossy image compression on CAD performance for solid lung nodules (>4 mm).
• To determine acceptable compression levels for maintaining diagnostic accuracy in lung nodule detection.

Main Methods:

• 120 MDCT chest scans (including low-dose) were analyzed.
• Images were compressed using the 3D Set Partitioning in Hierarchical Trees algorithm at 24:1, 48:1, and 96:1 ratios.
• CAD sensitivity was assessed at 2.5, 5, and 10 false marks per case using McNemar's test and logistic regression.

Main Results:

• No significant degradation in CAD sensitivity was observed across studied compression levels (24:1, 48:1, 96:1) compared to uncompressed scans.
• The 24:1 compression level showed significantly better performance than 96:1 and was occasionally superior to no compression.
• Nodule location (juxtapleural) was a significant predictor of lower CAD sensitivity; nodule size, dose, and contrast were not significant.

Conclusions:

• Computer-aided detection (CAD) performance for solid lung nodules is not significantly compromised by 3D lossy image compression up to a 48:1 ratio.
• MDCT chest scan compression is feasible without substantial loss of CAD detection capability for nodules >4 mm.