Lung Tumor Detection using LW-convMLP Segmentation with Improved Golden Jackal Optimization-based DRNet-MM Classification

Abstract

Lung cancer (LC) remains a critical global health issue, characterized by abnormal cell proliferation. Early detection is crucial, relying on imaging techniques such as CT, MRI, and ultrasound. This study leverages the Linear Imaging and Self-Scanning Sensor (LISS) and the Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI) datasets. CT images were preprocessed using local fractional entropy (LFE) and rolling guidance filtering (RGF). Tumor segmentation was performed using the Light Weight Medical Image Segmentation Network, integrating Convolutional Neural Network (CNN) and Multi-Layered Perceptron (MLP) based on the UNet architecture (LW-convMLP UNet). The study assessed the efficacy of the Deep Residual Neural Network with Masked Modeling (DRNet-MM) for LC classification, with hyperparameters optimized using the Improved Golden Jackal Optimization Algorithm (IGJOA). The proposed model exhibited outstanding performance, achieving precision and accuracy scores of 99.10% and 99.4% for LIDC-IDRI, and 90.3% and 90.25% for LISS. In conclusion, this method surpasses previous approaches, demonstrating its effectiveness in detecting and categorizing lung tumors.