Fast direct reconstruction strategy of dynamic fluorescence molecular tomography using graphics processing units

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computational Science

Background:

• Dynamic fluorescence molecular tomography (DFMT) is crucial for in vivo metabolic assessment.
• Direct reconstruction methods enhance DFMT temporal resolution but are time-consuming.
• Computational efficiency is a bottleneck for advanced DFMT applications.

Purpose of the Study:

• To accelerate the direct reconstruction method for DFMT.
• To reduce the computational time of DFMT without sacrificing image quality.
• To enable faster in vivo metabolic evaluations using DFMT.

Main Methods:

• Implemented conjugate gradient optimization using Compute Unified Device Architecture (CUDA) for GPU acceleration.
• Developed and validated a GPU-accelerated direct reconstruction algorithm for DFMT.
• Performed numerical simulations and in vivo experiments to assess performance.

Main Results:

• The GPU-accelerated DFMT reconstruction achieved a time reduction of approximately 90% compared to traditional methods.
• No degradation in image quality was observed with the accelerated strategy.
• Feasibility was confirmed through both simulations and in vivo studies.

Conclusions:

• GPU acceleration is a feasible and effective strategy to overcome the time limitations of direct DFMT reconstruction.
• This advancement significantly improves the efficiency of DFMT, facilitating rapid in vivo metabolic imaging.
• The developed method holds promise for broader clinical and research applications of DFMT.