Deep Learning-Based Computed Tomography Perfusion Mapping (DL-CTPM) for Pulmonary CT-to-Perfusion Translation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Pulmonary perfusion imaging is crucial for diagnosing lung diseases.
• Current methods like SPECT/CT are effective but involve radiation and contrast agents.
• There is a need for non-invasive, accurate methods to assess lung perfusion.

Purpose of the Study:

• To develop a deep learning-based computed tomography (CT) perfusion mapping (DL-CTPM) method.
• To synthesize lung perfusion images directly from CT images.
• To evaluate the accuracy of DL-CTPM against SPECT/CT in estimating lung perfusion.

Main Methods:

• Retrospective analysis of 146 pulmonary SPECT/CT scans from 73 patients.
• Development of a 3D attention residual neural network for feature extraction and image reconstruction.
• Voxel-wise agreement assessed using Spearman's correlation (R) and structural similarity index measure (SSIM).
• Function-wise concordance evaluated using Dice Similarity Coefficient (DSC) for high/low functional lung volumes.

Main Results:

• DL-CTPM showed moderate-to-high voxel-wise agreement with SPECT (R=0.6733, SSIM=0.7635).
• Average DSC for high-functional lungs was 0.8183 and for low-functional lungs was 0.6501.
• 94% of test cases achieved high concordance (DSC >0.7) for high-functional lung volumes.

Conclusions:

• A novel DL-CTPM method effectively estimates lung perfusion from CT images.
• The 3D attention residual neural network provides moderate-to-high voxel-wise approximations of lung perfusion.
• Further validation in multi-institutional, large-cohort studies is recommended for clinical application.