Hybrid Seeker Optimization Algorithm-based Accurate Image Clustering for Automatic Psoriasis Lesion Detection
Manoranjan Dash, Narendra D. Londhe, Subhojit Ghosh, Rajendra S. Sonawane
Abstract
Automatic lesion segmentation plays a significant role in disease diagnosis and severity detection. Specifically, in psoriasis, the lesion exhibits color diversity inside and around the boundary, which makes its extraction very challenging. The psoriasis lesion segmentation utilizing the classical K-means clustering technique was performed by our group. But it has been observed that the centroids of the clusters obtained from K-means are local optimum in nature. To overcome the limitation pertaining to trapping into the local optima and hence provide globally optimal centroids, we have designed a hybrid seeker optimization-based image clustering to improve the detection of psoriasis lesions. The proposed algorithm is executed in two steps. At the start of the proposed algorithm, seeker optimization algorithm (SOA) has been implemented to obtain the initial set of cluster centroids from the random search space. Further, these cluster centroids are refined using K-means algorithm. Integrating a global optimization technique with the local search technique such as K-means increases the convergence probability of a global optimum result with reduced computational cost, thereby resulting in improved clustering and detection. The efficacy of the hybrid optimization is evaluated by implementing it on 522 psoriasis images considering the different class of severity specifically mild (67), moderate (155), severe (172), and very severe (128). The performance has been evaluated using four metrics, namely accuracy (91.4%), sensitivity (93.1%), specificity (92.0%), and Jaccard Index (0.873). The mean lesion detection accuracy confirms the supremacy of hybrid seeker optimization algorithm.