Computational Intelligence Re-meets Medical Image Processing
Analysis of Machine Learning Algorithms for Diagnosis of Diffuse Lung Diseases
Abstract
Background In the last decades, new optimization methods based on the nature's intelligence
were developed. These metaheuristics can find a nearly optimal solution faster than
other traditional algorithms even for high-dimensional optimization problems. All
these algorithms have a similar structure, the difference being made by the strategies
used during the evolutionary process.
Objectives A set of three nature-inspired algorithms, including Cuckoo Search algorithm (CSA),
Particle Swarm Optimization (PSO), and Multi-Swarm Optimization (MSO), are compared
in terms of strategies used in the evolutionary process and also of the results obtained
in case of particular optimization problems.
Methods The three algorithms were applied for biomedical image registration (IR) and compared
in terms of performances. The expected geometric transform has seven parameters and
is composed of rotation against a point in the image, scaling on both axis with different
factors, and translation.
Results The evaluation consisted of 25 runs of each IR procedure and revealed that (1) PSO
offers the most precise solutions; (2) CSA and MSO are more stable in the sense that
their solutions are less scattered; and (3) MSO and PSO have a higher convergence
speed.
Conclusions The evaluation of PSO, MSO, and CSA was made for multimodal IR problems. It is possible
that for other optimization problems and also for other settings of the optimization
algorithms, the results can be different. Therefore, the nature-inspired algorithms
demonstrated their efficacy for this class of optimization problems.
Keywords
nature-inspired algorithms - metaheuristic - optimization - image registration
