Denoising PET images for proton therapy using a residual U-net

Area of Science:

• Medical Physics
• Radiotherapy
• Image Processing

Background:

• Proton therapy offers precise dose delivery but suffers from range uncertainty, impacting normal tissue sparing.
• Proton-induced positron emission tomography (pPET) visualizes proton irradiation but requires denoising for accurate analysis.
• Conventional denoising methods like Gaussian filters can reduce spatial resolution and distort pPET image features.

Purpose of the Study:

• To develop and evaluate a novel denoising method for pPET images using a Residual U-Net architecture.
• To assess the performance of the proposed Residual U-Net against conventional filters for pPET image enhancement.
• To determine the impact of the denoising method on proton range estimation accuracy and image quality.

Main Methods:

• Developed a denoising algorithm based on a Residual U-Net convolutional neural network.
• Acquired pPET data through Monte Carlo simulations and experimental irradiation of a human phantom.
• Compared the Residual U-Net method with Gaussian, median, BM3D, and total variation filters using range estimation accuracy and image similarity metrics.

Main Results:

• The Residual U-Net demonstrated effective proton range estimation comparable to conventional methods.
• Peak-signal-to-noise ratio results for the Residual U-Net were similar to those achieved by Gaussian, median, BM3D, and TV filters.
• The proposed method showed potential for enhancing pPET image quality without compromising essential features.

Conclusions:

• The Residual U-Net based denoising method shows promise for improving pPET image analysis in proton therapy.
• Accurate pPET imaging is crucial for verifying proton treatment delivery and optimizing dose distribution.
• This approach could lead to enhanced accuracy in treatment verification and reduced PET measurement durations.