Patient specific tumor growth prediction using multimodal images

Area of Science:

• Biomedical Engineering
• Computational Biology
• Medical Imaging

Background:

• Personalized tumor growth models are crucial for accurate cancer staging and effective therapy planning.
• Integrating multimodal imaging data enhances model precision for individual patient cases.

Purpose of the Study:

• To present a patient-specific tumor growth model utilizing longitudinal dual-phase CT and FDG-PET imaging.
• To incorporate Intracellular Volume Fraction (ICVF) from CT and Standardized Uptake Value (SUV) from PET into a Reaction-Advection-Diffusion model.

Main Methods:

• Developed a Reaction-Advection-Diffusion model integrating proliferation, infiltration, metabolic rate, and biomechanics.
• Formulated parameter estimation as a Partial Differential Equations (PDE)-constrained optimization inverse problem.
• Solved the optimality system using the Finite Difference Method.

Main Results:

• The model accurately predicted tumor boundaries and characteristics in six pancreatic neuroendocrine tumor patients.
• Achieved an average surface distance (ASD) of 2.4±0.5mm between predicted and reference tumors.
• Reported low discrepancies in Intracellular Volume Fraction (RMSD: 4.3±0.4%, AICVFD: 2.6±0.6%) and volume (RVD: 7.7±1.3%).

Conclusions:

• The developed patient-specific model effectively integrates multimodal imaging data for tumor growth prediction.
• This approach shows significant potential for improving tumor staging and personalized therapy planning.
• The model's accuracy was validated through quantitative comparisons with observed tumor data.