Optimising Multimodal Image Registration Techniques: A Comprehensive Study of Non-Rigid and Affine Methods for PET/CT...

Area of Science:

• Medical Imaging
• Image Processing
• Radiology

Background:

• Multimodal image registration is crucial for integrating data from different medical imaging modalities like PET and CT.
• Accurate alignment of PET/CT images is essential for diagnosis and treatment planning.

Purpose of the Study:

• To compare the performance of Demons, MIRT, and MATLAB Intensity-Based Registration for PET/CT image alignment.
• To evaluate the impact of preprocessing techniques on registration accuracy.

Main Methods:

• Analysis of 100 matched PET/CT slices using Demons, MIRT, and MATLAB Intensity-Based Registration.
• Application of preprocessing techniques including histogram equalization and contrast enhancement.
• Evaluation using metrics like RMSE, MSE, MAE, PCC, and STD under varying parameters.

Main Results:

• MATLAB Intensity-Based Registration with contrast enhancement achieved the highest accuracy (RMSE = 0.1317).
• Demons registration demonstrated superior computational efficiency (RMSE = 0.1529 at sigma fluid = 6).
• MIRT showed adaptability to complex deformations (RMSE = 0.1725). Preprocessing improved accuracy by up to 16%.

Conclusions:

• Each registration method offers distinct advantages: Demons for speed, MIRT for precision, and MATLAB for flexibility.
• The findings provide a framework for optimizing PET/CT image registration in clinical and research settings.