Application of Chord Length Histogram to Enhance the Accuracy of Pulmonary Nodule Recognition on Computed Tomograms

The advancement of computed tomography (CT) technologies in pulmonary nodule medical diagnostics necessitates the development of efficient classification and segmentation methods. This study assesses the utility of the Chord Length Distribution Histogram (CLDH) in conjunction with machine learning techniques, including SVR, random forest, and gradient boosting, for comprehensive analysis of the three-dimensional (3D) geometry of pulmonary nodules. The results indicate that CLDH serves as a promising shape descriptor, providing high accuracy in classifying pulmonary nodules. Compared to neural network approaches, the CLDH-based method maintains effectiveness even when confronted with limited training data, thus optimizing computational resource utilization. The proposed methodology for characterizing pulmonary nodule morphology showcases robust data separability and achieves a classification accuracy rate of 0.8 on the LUNA16 dataset. These results emphasize the importance of integrating classical machine learning methodologies into the analysis of three-dimensional geometric structures to refine lung cancer diagnosis accuracy.