Deep-learning-based Attenuation Correction for ⁶⁸Ga-DOTATATE Whole-body PET Imaging: A Dual-center Clinical Study

Area of Science:

• Nuclear Medicine
• Medical Imaging
• Artificial Intelligence

Background:

• Attenuation correction is crucial for quantitative positron emission tomography (PET) imaging.
• Computed tomography (CT) is used for attenuation correction but increases patient radiation dose.

Purpose of the Study:

• To develop a deep learning model for attenuation correction of whole-body 68Ga-DOTATATE PET images.
• To evaluate the model's performance and compare it with traditional CT-based methods.

Main Methods:

• A residual deep learning model was implemented using the NiftyNet framework.
• The model was trained and evaluated on whole-body 68Ga-DOTATATE PET images from 118 patients across two centers.
• Image quality was assessed using Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), Mean Square Error (MSE), and Root Mean Square Error (RMSE).

Main Results:

• The deep learning model achieved high quantitative accuracy, with PSNR values up to 52.86±6.6 and SSIM values up to 0.99±0.003.
• Using datasets from the same imaging center yielded the highest PSNR, while combining data from both centers resulted in the best SSIM and lowest MSE/RMSE.
• The model demonstrated robust performance across different evaluation metrics.

Conclusions:

• Deep learning provides an accurate method for attenuation correction in 68Ga-DOTATATE PET imaging.
• This approach can potentially eliminate the need for CT scans, thereby reducing patient radiation exposure.
• The findings support the clinical utility of AI-driven attenuation correction in PET imaging.