Reproducible and Interpretable Spiculation Quantification for Lung Cancer Screening.

Area of Science:

• Radiomics
• Medical Imaging Analysis
• Computational Pathology

Background:

• Spiculations on pulmonary nodules are key indicators of lung cancer malignancy.
• Accurate quantification of spiculations is crucial for early cancer detection.
• Existing methods may lack interpretability or require parameter tuning.

Purpose of the Study:

• To develop an interpretable, parameter-free method for quantifying nodule spiculations using area distortion.
• To introduce novel spiculation scores and measures for improved characterization.
• To integrate this feature into a radiomics framework for lung cancer malignancy prediction.

Main Methods:

• Utilized conformal spherical parameterization to calculate area distortion metric.
• Developed semi-automatic segmentation for nodules, vessels, and wall attachments.
• Trained and validated radiomics models on the LIDC-IDRI and LUNGx datasets.

Main Results:

• Achieved Area Under the Curve (AUC) of 0.80 and 0.76 in external validation.
• Outperformed previous models on the LUNGx dataset (AUC 0.68 vs. 0.80).
• Demonstrated high correlation (ρ=0.44) between the new spiculation feature and radiologist scores.

Conclusions:

• The developed technique provides a reproducible and interpretable method for quantifying spiculations.
• The novel spiculation features enhance radiomics models for malignancy prediction.
• This approach shows promise for clinical application in lung cancer screening.