Multi-Objective Butterfly Optimization for Feature and classifier parameter's selection in Diagnosis of Heart Failure types Using CMR images

Heart Failure (HF) ranks among the fastest growing abnormal cardiovascular conditions. There is considerable heterogeneity in external symptom burdens that are predominantly neglected, especially during the transition from one stage to another. Classifying patients into different phenotypic groups is essential to plan effective treatment. These disorders result in deteriorated heart muscles in CMR images. The proposed work aims at determining the variations of optimized anatomic and texture features extracted from the segmented left ventricle myocardium. In the approach followed, the Co-occurrence of Adjacent sparse Local Ternary Pattern and Krawtchouk moment are employed to extricate the texture and anatomic measures in various frames of the left ventricle myocardium. However, the high dimensional feature vector necessitates an efficient feature selection algorithm. The main advantage of the nature-inspired butterfly optimization algorithm (BOA) is its capability to perform both global and local searches with less complex computation. This study recommends deploying multi-objective BOA to enhance the performance of the support vector machine classifier and identify an optimized percentage of extracted features. This addition has yielded the highest diagnosis rate (93.1%) of individuals with different HF stages. The optimized features have shown noticeable performance in discriminating patients with hyperdynamic and mild HF from healthy subjects. This study shows the remarkable impact of multi-objective feature selection on disease diagnosis. It also confirms that tissue characterization of CMR images will help effective HF diagnosis.