Robust filtering of thin-slice reconstructions improves lacunar stroke detection in CT perfusion imaging

Area of Science:

• Medical Imaging
• Neurology
• Radiology

Background:

• Reducing CT perfusion (CTP) slice thickness can improve lacunar stroke detection.
• Thinner slices increase image noise, requiring effective noise reduction filters.

Purpose of the Study:

• To assess the impact of slice thickness and noise filters on detecting lacunar strokes.
• Evaluate three advanced noise filters for CTP imaging.

Main Methods:

• Artificial lacunar infarcts (5-10mm) were added to CTP data.
• Compared temporal-average, time-intensity profile, and U-Net filters on thin (0.7mm) and thick (4.9mm) slices.
• Assessed infarct detectability using F1-score, CNR, and observer confidence.

Main Results:

• Temporal-average guided filtering yielded the highest F1-scores (0.70 thin, 0.49 thick slices).
• Thin-slice temporal-average filtering achieved the best contrast-to-noise ratio and observer confidence.
• Sensitivity for 7mm and 10mm lacunar infarcts reached 73% and 75% respectively with this method.

Conclusions:

• Robust noise filtering, specifically temporal-average guided filtering, is crucial for CTP.
• Thin-slice CTP with temporal-average filtering significantly improves detection of lacunar strokes ≥7mm.