ROC and localization ROC analyses of lesion detection in whole-body FDG PET: effects of acquisition mode, attenuation...

Area of Science:

• Nuclear Medicine
• Medical Imaging Analysis

Background:

• Positron Emission Tomography (PET) is crucial for lung cancer staging.
• Accurate detection of small lung lesions is vital for effective staging.

Purpose of the Study:

• To compare the effectiveness of different image reconstruction methods for detecting lung cancer lesions on PET scans.
• To evaluate the impact of acquisition dimensions and attenuation correction techniques on lesion detectability.

Main Methods:

• Receiver Operating Characteristic (ROC) and Localization ROC (LROC) analyses were employed.
• Simulated lung cancer lesions were added to PET data from healthy volunteers.
• Four reconstruction methods were compared: 2D FBP without AC, 3D FBP without AC, 2D FBP with segmented AC, and 2D iterative MAP with AC.

Main Results:

• Three-dimensional acquisition and segmented attenuation correction significantly degraded lesion detection performance compared to 2D FBP without AC.
• Iterative MAP reconstruction showed a non-statistically significant improvement in detection performance over 2D FBP.
• Average Az for 2D FBP without AC was 0.79, while 3D FBP was 0.51 and segmented AC was 0.59.

Conclusions:

• 3D acquisition and segmented attenuation correction do not enhance lung lesion detection on whole-body PET images.
• Iterative MAP reconstruction offers only a marginal, non-significant improvement in lesion detection compared to standard 2D FBP.