A Confidence-Optimised and Edge-Guided Deep Learning Framework for Precise Lung Cancer Identification and Risk Assessment

Lung cancer remains one of the leading causes of cancer-related mortality worldwide due to challenges in early detection and accurate risk stratification. Recent advancements in deep learning have revolutionized medical imaging, enabling precise tumor detection, classification, and prognosis. However, conventional deep learning systems often lack interpretability, robustness, and confidence-aware prediction, making clinical adoption limited. This paper proposes a Confidence-Optimized and Edge-Guided Deep Learning Framework (COEG-DLF) for lung cancer identification and risk assessment. The framework integrates edge-preserving segmentation, convolutional feature extraction, and probabilistic confidence calibration to ensure robust tumor boundary delineation and reliable risk stratification. We incorporate attention-guided convolutional neural networks (CNNs) for high-level feature extraction and a Bayesian confidence optimization layer for uncertainty estimation. A large-scale survey of existing methods is presented to benchmark the strengths and limitations of prior models. Experimental evaluation using publicly available lung cancer datasets (LIDC-IDRI, TCIA) demonstrates that the proposed framework outperforms traditional CNN and transformer-based approaches in terms of accuracy, precision, and reliability of predictions. This work contributes to the field by introducing an interpretable, clinically reliable, and computationally efficient framework that supports oncologists in early detection and personalized treatment planning.