A Deep Auto-Encoder Approach for Prostate Cancer Detection Using Meta-Hybrid Learning and Ensemble Methods

Prostate cancer incidence has steadily climbed globally over time, particularly as the share of the older population has increased. Prostate cancer is discovered early, while it is still limited to the prostate glands, having a high potential of being successfully treated and has a higher survival rate. The main contribution of the proposed work is to develop an automated prostate cancer diagnosis and segmentation system with the use of groundbreaking medical image processing techniques. To achieve this, it is necessary to design a system that can reliably extract fine-grained features, distinguish between benign and cancerous structures, and contribute to more accurate diagnoses. In order to enhance the training and testing performance of the cancer detection system, the most important features from the preprocessed prostate MRI are selected using a novel Wild Horse Optimized (WHO) feature selection method. Subsequently, a collection of deep learning models, comprising Convolutional Neural Network (CNN), Residual Network (ResNet), and Generative Adversarial Network (GAN) approaches, is employed to accurately classify prostate photos pertaining to cancer and those that are not. The voting process determines which prediction model is the best, and the final classification result is determined by that model's anticipated outcome. Additionally, a new technique called Dual Swin Transformer UNet Segmentation (DSTra-UNet) is used to separate the prostate image's cancer-affected area. Numerous performance assessment metrics, including accuracy, sensitivity, precision, recall, and mistake rate, are compared and validated in this study.