Lung cancer continues to be the predominant cause of cancer-related mortality globally, representing 11.6% of all cancer cases as reported by WHO, with approximately 2.2 million new diagnoses and 1.8 million fatalities recorded in 2020. For lung cancer diagnosis, Computed Tomography (CT) imaging serves as a critical tool in identifying both solid and subsolid ("ground glass") nodules. Although CT image segmentation has demonstrated significant clinical value, healthcare practitioners require a comprehensive understanding of the underlying algorithmic mechanisms to ensure diagnostic precision. This research investigates the application of interpretable machine learning methods for feature extraction from lung CT imaging data. We conduct a comparative analysis between transparent and opaque classification algorithms utilizing a comprehensive dataset comprising 1,229 normal and 1,010 abnormal pulmonary CT scans. The processed data undergoes evaluation using both interpretable models (including Logistic Regression, Decision Trees, and K-Nearest Neighbor) and black-box models (such as Multi-Layer Perceptrons, Convolutional Neural Networks, and Support Vector Machines). Our findings indicate that interpretable algorithms consistently outperform their black-box counterparts across multiple metrics. The evaluation framework incorporates accuracy, F1 score, precision, recall, computational efficiency, and resource utilization measurements. Results demonstrate exceptional classification accuracy for pulmonary malignancy detection while preserving explanatory capability, thereby providing clinicians with both practical and transparent diagnostic assistance. This investigation contributes to the development of accountable artificial intelligence systems for deployment in mission-critical healthcare environments.
Tanthanawarakun et al. (Sun,) studied this question.