Lung nodule classification using radiomics model trained on degraded SDCT images

Area of Science:

• Medical Imaging
• Radiomics
• Artificial Intelligence in Medicine

Background:

• Low-dose computed tomography (LDCT) screening reduces lung cancer mortality but faces challenges with high false positive rates and limited annotated datasets.
• Synthetic LDCT images were generated from standard-dose CT (SDCT) scans to address data scarcity.

Purpose of the Study:

• To develop and validate an interpretable radiomics-based model for distinguishing benign from malignant pulmonary nodules.
• To leverage synthetic LDCT images for improved lung nodule classification.

Main Methods:

• SDCT scans were degraded in the sinogram domain to create synthetic LDCT images for training.
• Radiomic features were extracted from nodules, and models were trained and validated using both synthetic and real LDCT data.
• A systematic pipeline optimized feature sets and evaluated machine learning models.

Main Results:

• A logistic regression model using only shape and size features achieved a mean balanced accuracy of 0.81, sensitivity of 0.76, specificity of 0.85, and AUC-ROC of 0.87 on the test set.
• The model demonstrated effectiveness in classifying lung nodules using synthetic LDCT data.

Conclusions:

• The study confirms the feasibility and effectiveness of using synthetic LDCT images for developing accurate radiomics models in lung nodule classification.
• This approach offers a potential solution to challenges in LDCT screening, aiming to improve lung cancer detection and reduce false positives.