Multi-Head Probsparse Spherical Convolutional Network With Lyrebird Optimization Algorithm For Prostate Cancer Identification

Abstract

Prostate cancer remains a leading cause of mortality among men worldwide, and accurate early detection is critical for improving treatment outcomes. Traditional diagnostic methods and conventional imaging analyses often suffer from low Accuracy, high inter-observer variability, and limited capacity to capture complex tissue structures in MRI images. To address these limitations, this study proposes a novel Multi-Head ProbSparse Spherical Convolutional Network with Lyrebird Optimization Algorithm (M-HPSCNet-LOA) for prostate cancer identification from the SPIE-AAPM-NCI Prostate dataset using  MRI images. The framework incorporates Deep Attentional Guided Image Filtering (DAGIF) for pre-processing, Single-Head Vision Transformer (SHViT) for segmentation, and Dimba: Transformer-Mamba Diffusion Models (TMDM) for feature extraction. Classification is achieved through the Multi-Head ProbSparse Spherical Convolutional Network (MHPSCNet), which integrates Multi-Head ProbSparse Self-Attention (MHPSAN) and Spherical Convolutional Neural Network (SCNN) to capture both global tissue structures and localized intensity patterns. The Lyrebird Optimization Algorithm (LOA) further enhances model performance by fine-tuning parameters efficiently. Experimental results demonstrate that M-HPSCNet-LOA achieves 99.85% accuracy, 99.58% F1-score, and 99.62% precision, highlighting its robustness, stability, and applicability in MRI-based prostate cancer detection.