Automatic attenuation map estimation from SPECT data only for brain perfusion scans using convolutional neural...

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Artificial Intelligence in Healthcare

Background:

• Accurate quantitative imaging in SPECT requires photon attenuation correction.
• Traditional methods rely on CT/MRI scans, which may not always be available or perfectly registered.
• Registration errors and lack of auxiliary scans limit quantitative SPECT accuracy.

Purpose of the Study:

• To develop and validate a SPECT-only strategy for estimating attenuation maps.
• To implement a convolutional neural network (CNN) for attenuation map generation.
• To assess the accuracy of CNN-based attenuation correction compared to traditional methods for brain perfusion imaging.

Main Methods:

• A CNN-based approach was developed for SPECT-only attenuation map estimation.
• The method utilized energy information from primary and scattered photons in list-mode data.
• Multiple SPECT reconstructions across a wide energy range served as input for the CNN.

Main Results:

• CNN-estimated attenuation maps achieved accuracy comparable to ground truth maps.
• Quantitative analysis showed a mean deviation of only 1.8% in standardized uptake values.
• Visual assessment confirmed the high quality of CNN-based attenuation correction.

Conclusions:

• A CNN-based SPECT-only method provides accurate and automatic attenuation correction.
• This technique is independent of other imaging modalities (CT/MRI) and human interpretation.
• The proposed method enhances the reliability of quantitative SPECT imaging for brain perfusion.