A four-layer Digital Twin–Metaverse framework for lung cancer detection: Empirical evaluation of YOLOv۸ optimization strategies

Lung cancer remains one of the leading causes of cancer-related mortality worldwide; however, its accurate and timely diagnosis continues to face limitations. Addressing this challenge demands not only advances in detection accuracy but a fundamental rethinking of the diagnostic ecosystem toward frameworks that are immersive, collaborative, and capable of continuous patient-specific modeling. This paper presents a four-layer architectural framework that conceptually integrates Digital Twin and Metaverse technology to redefine lung cancer diagnostics. Within this framework, a dynamically updated virtual replica of the patient's pulmonary system serves as the three-dimensional operational substrate, which is accessible to clinical teams. To empirically ground the detection component of this framework, we conduct a rigorous comparative evaluation of three optimization algorithms Stochastic Gradient Descent (SGD), Adam, and AdamW applied to YOLOv۸ on a four class lung Computed Tomography (CT) dataset encompassing adenocarcinoma, large cell carcinoma, squamous cell carcinoma, and normal tissue. AdamW achieved the highest classification performance, attaining an accuracy of ۹۳.۹۷%, precision of ۹۴.۰۶%, recall of ۹۲.۸۷%, F۱-score of ۹۳.۴۵%, and mAP۵۰ of ۰.۹۳۷۱. SGD demonstrated competitive localization fidelity with the highest mAP۵۰ of ۰.۹۷۴۶, while Adam exhibited overfitting signatures manifested as validation loss divergence and asymmetric class-level error distribution. These results establish AdamW-optimized YOLOv۸ as a deployment-ready detection backbone for integration within the proposed framework.

کلیدواژه ها:

Lung Cancer Detection ، Digital Twin ، Metaverse ، YOLOv۸ ، Optimizer Comparison ، Deep Learning ، Medical Imaging