Deep Learning-Based Medical Image Classification Segmented with Particle Swarm Optimization Technique

Medical image segmentation and classification is a prior task in medical image analysis. Therefore, a generalized method is required to process these images efficiently. Objective of our work is to use a nature inspired metaheuristic algorithm, particle swarm optimization (PSO). The algorithms is applied on three types of databases available in repositories. These databases include breast cancer ultrasound, chest-X- Rays, and carotid ultrasound images. After performing PSO clustering, medical images are classified into binary class using state of the art deep learning models VGG16, GoogleNet and ResNet50 models. Out of these DL models GoogleNet performs better classification performance for all types of database. The PSO-based approach effectively produces distinct ROI clusters for cluster 3, while the cluster 4-based method highlights variations in ROI tissue density. In summary, the PSO-based method stands out as a robust tool for delineating regions of interest in medical images.