Statistical tools for the temporal analysis and classification of lung lesions

Area of Science:

• Medical Imaging Analysis
• Quantitative Pathology
• Radiomics

Background:

• Lung cancer is a leading global malignancy.
• Temporal evaluation of lung lesions aids in assessing treatment response and differentiating benign from malignant nodules.
• Accurate characterization of lung lesions is critical for patient management.

Purpose of the Study:

• To develop and validate a methodology for analyzing, quantifying, and visualizing local and global changes in lung lesions over time.
• To extract textural features for improved classification of lung lesions as benign or malignant.
• To assess the utility of temporal changes in density and volume for lesion characterization.

Main Methods:

• Employed statistical uncertainty to associate voxel-level probabilities of change within lesions.
• Utilized Jensen divergence and hypothesis testing for local (voxel-to-voxel) and global (volume) change detection.
• Performed texture analysis on regions exhibiting density changes for lesion classification.

Main Results:

• Local hypothesis testing revealed density changes ranging from 3.84% to 40.01% in malignant lesions and 5.76% to 35.43% in benign nodules.
• Texture analysis of changing regions achieved 98.41% accuracy in discriminating between benign and malignant lung lesions.
• Quantified temporal changes in lesion density and volume, correlating divergence values with lesion characteristics.

Conclusions:

• The methodology effectively quantifies temporal density and volume changes in lung lesions.
• Significant density changes, even with small volume alterations, are indicative of malignant lesions.
• The approach demonstrates that even seemingly stable lesions exhibit density changes, aiding in the characterization of benign nodules.