Long-range medical image registration through generalized mutual information (GMI): toward a fully automatic volumetric alignment

TL;DR

This paper introduces a generalized mutual information (GMI) method for medical image registration, significantly extending the registration range and reliability for affine transformations in volumetric images.

Contribution

The paper proposes and evaluates a novel GMI function that improves the registration range and accuracy, overcoming local maxima issues in traditional mutual information methods.

Findings

GMI provides smooth isosurfaces guiding to global maxima.

Registration range extended to ±150mm translations and ±180° rotations.

Achieved over 99% success rate in simulated and real MRI image registration.

Abstract

Image registration is a key operation in medical image processing, allowing a plethora of applications. Mutual information (MI) is consolidated as a robust similarity metric often used for medical image registration. Although MI provides a robust medical image registration, it usually fails when the needed image transform is too big due to MI local maxima traps. In this paper, we propose and evaluate a generalized parametric MI as an affine registration cost function. We assessed the generalized MI (GMI) functions for separable affine transforms and exhaustively evaluated the GMI mathematical image seeking the maximum registration range through a gradient descent simulation. We also employed Monte Carlo simulation essays for testing translation registering of randomized T1 versus T2 images. GMI functions showed to have smooth isosurfaces driving the algorithm to the global maxima. Results show significantly prolonged registration ranges, avoiding the traps of local maxima. We evaluated a range of [-150mm,150mm] for translations, [-180{\deg},180{\deg}] for rotations, [0.5,2] for scales, and [-1,1] for skew with a success rate of 99.99%, 97.58%, 99.99%, and 99.99% respectively for the transforms in the simulated gradient descent. We also obtained 99.75% success in Monte Carlo simulation from 2,000 randomized translations trials with 1,113 subjects T1 and T2 MRI images. The findings point towards the reliability of GMI for long-range registration with enhanced speed performance