The continuous study of natural systems has led to the development of nature inspired computational systems and techniques. The candidate solutions for the optimization problems at hand take on the role of individuals in a population function while the quality of the solutions is determined by the fitness functions. A commonly used nature-inspired algorithm is the Cuckoo Search (CS) algorithm which is a heuristic search algorithm derived from the reproduction strategy of the cuckoo birds. The Cuckoo Search algorithm has widespread applications in various domains such as weight optimization in neural networks, job scheduling, image processing, radial basis functions and clustering among others. However, this paper briefs on the CS algorithm, its applications in image processing, and gives specific emphasis on medical imaging (Kamat and Karegowda).
In medical imaging systems, physicians use Content-Based Medical Image Retrieval (CBMIR) to convey effective decisions to patients as accurately as possible. Medical research students also use CBMIR for research to study the characteristics of medical images and process them based on visual features. The performance of medical imaging systems is however restricted by the high feature dimension characteristic of visual features. In order to reduce the high feature dimensions, several integrated approaches have been proposed, and these include extraction of visual features, selection of features, classification, and comparison of similarity measurements. In feature selection, the CS algorithm is applied in modified techniques known as Fuzzy based Cuckoo Search (FCKS). These techniques aim to reduce the high feature vector dimensionality, address the problems of feature optima surrounding feature vectors; and to determine the global optimum feature position that is special for all featured cuckoo hosts (Jaganathan and Vennila).
In CBMIR feature extraction where the CS algorithm is applied, visual features in medical images are extracted as numerical values and then stored in the form of feature vectors in a medical image database. Texture features are highly preferred for image retrieval due to their periodicity and the ease of scaling results when gathering data on semantic features in images. However, the technique also takes up a large number of finite grey levels within the selected images (Kamat and Karegowda).
In traditional image retrieval systems, statistical transforms and spatial frequency based texture descriptors were used. These include Gabor Filters (GF), Tamura Features (TF) and Gray level Co-occurrence Matrix (GLCM) methods (Jaganathan and Vennila). However, to overcome the limitations of the traditional techniques described earlier, the Cuckoo Search algorithm is suggested for use in modern medical image retrieval systems due to its powerful optimization techniques.
The CS algorithm emulates the brooding behavior of some cuckoo birds species that is also observed in the Levy flight behavior of some other birds and fruit flies. Essentially cuckoo birds lay their eggs in a communal nest. However, if the host bird finds that the eggs in its nest are alien, the bird either discards the alien eggs or simply abandons the nest and moves on to build a new one elsewhere. The CS algorithm is based on three idealized rules; one is that each cuckoo bird lays an egg at a time in a randomly chosen nest. The second rule is that the nests with highest quality nests carryover the next generation. Finally, it is assumed that the number of available hosts per nest is fixed, and that the probability of the host bird discovering an egg laid by a cuckoo is P ∈[0,1] (Kamat and Karegowda).
The CS algorithm begins with an obstinate number of nests that continues to increase as compared to the previous number of nests during the final stage of each cuckoo generation. The FCKS technique is used to choose the best solution that would result in the optimum number of nests alive to begin with for each upcoming generation. The fuzzy membership function is enclosed in the FCKS algorithm to with its goal being to maximize the fitness function. When FCKS is used, it can search for the best host nests in both a local and global scope, and also choose the appropriate nests to reject (Jaganathan and Vennila). By using this analogy in feature selection during image retrieval in medical images, the CS algorithm selects the best features that are needed in image retrieval and rejects those that are not needed.
Works Cited:
Jaganathan, Yogapriya, and Ila Vennila. 'AN INTEGRATED FRAMEWORK BASED ON TEXTURE FEATURES, CUCKOO SEARCH AND RELEVANCE VECTOR MACHINE FOR MEDICAL IMAGE RETRIEVAL SYSTEM'. American Journal of Applied Sciences 10.11 (2013): 1398. Print.
Kamat, Sanket, and Asha Gowda Karegowda. 'A Brief Survey on Cuckoo Search Applications'. Print.
