MULTI-SCALE CNN-BASED EARLY DETECTION OF LUNG NODULES FROM LOW-DOSE CT SCANS

Introduction: Detecting lung nodules early is crucial for enhancing clinical results in individuals vulnerable to lung cancer. Standard CT-based screening faces difficulties due to faint nodule shapes, diverse dimensions, and image noise in low-dose CT (LDCT) examinations. New developments in artificial intelligence, especially deep learning, provide encouraging diagnostic assistance. The purpose of this research is to assess the diagnostic accuracy of a multi‑scale convolutional neural network (CNN) for automated identification of pulmonary nodules in LDCT images.

Materials and Methods: A selected collection of 135 LDCT lung scans was utilized. All images underwent preprocessing through normalization, noise suppression, and resizing. Data augmentation techniques were employed to enhance robustness. A bespoke multi‑scale CNN model was trained to capture both high‑resolution details and broader lung patterns. The collection was split into training (80%) and testing (20%) subsets. Evaluation was performed using accuracy, precision, recall, F1‑score, and ROC‑AUC metrics.

Result: The multi‑scale CNN showed robust diagnostic performance. The system recorded an accuracy of 92.4%, a precision of 90.1%, a recall of 93.8%, and an F1‑score of 91.9%. Moreover, the ROC‑AUC of 0.95 reflected outstanding discrimination between nodule‑positive and nodule‑negative scans. It successfully detected small, subtle nodules that are frequently difficult to spot on LDCT images.

Conclusion: Multi-scale CNN-based analysis of LDCT images offers an efficient and dependable approach for early lung nodule detection. The model’s strong performance indicates its possible role as a supportive instrument in lung cancer screening workflows. Employing multi-scale feature extraction markedly improves nodule identification, underscoring its value in low‑contrast clinical imaging scenarios.